Click here to subscribe to print for your office or home.

We’ve barely scratched the surface of the total opportunity in robotics. If today’s startups achieve their ambitions, they will unlock over $1tn in revenue and disrupt how virtually every industry operates. 

Breakthroughs in robotics could massively expand existing verticals and unlock new ones. In established markets like manufacturing and logistics, there’s ample opportunity to increase robot adoption. In logistics, robots still leave 95% of the world’s warehouse space untouched—a $50bn revenue prize. 

Then there’s the long-held science fiction promise of helpful robots in our homes and daily lives. Enticing demo videos show robots doing complex tasks in homes—handling cooking utensils or navigating a messy living room to deliver a drink. But even simple robots could unlock huge consumer markets. The two-decade history of robot vacuums shows that a premium point solution for consumers is still worth billions of dollars annually. 

There’s immense value to unlock from getting more robots into the world. But the hard truth is that getting robots into the world is easier than keeping them there.

The challenge of commercializing robots means that companies have captured a vanishingly small fraction of the total opportunity on the table. Only two robot form factors have been broadly commercialized: industrial robot arms for manufacturing and robot vacuums. Other robot applications have struggled to gain traction. In the past decade, $26bn of venture funding has gone into US companies to bring robots to new markets. This investment has brought many robots out of the lab, but with little real progress past pilot testing. 

Most robotics startups—and their investors—have focused on hardware. But through our deep-dive research and conversations with the best technologists and founders working on robotics today, we’ve concluded that the real bottleneck is software. 

Robots are held back from wider commercialization by a lack of software that gives them the competence to reason about open-ended situations. For robots to reach their full potential in existing markets, and conquer new ones, they depend on software breakthroughs that make them more adaptable. With more adaptability, warehouse robots could take on broader tasks and better collaborate with people. Delivery robots could make smarter decisions about anomalous situations in hallways and on sidewalks. Consumer robots could expand their capabilities in messy, unpredictable home environments. 

Much of the current robotics zeitgeist hinges on the idea that today’s AI can supply these breakthroughs and equip robots with the same type of flexibility we’ve come to expect from large language models (LLMs). In the past year, robotics funding shifted as investors made unprecedented bets on this frontier. Investment in just six frontier robotics startups accounted for over 35% of all robotics venture investment in 2024. Companies working on AI models for robots raised over $1bn funding in 2024—nearly 10x what they raised in 2023. 

But the AI of today doesn’t automatically deliver new capabilities for robots. Instead, robotics companies will need to address both shortages and gridlocks in real-world robotics data in order to finally break the commercialization barrier. Only by addressing these systemic challenges can companies build mainstream-ready robots to operate in unstructured, everyday places. 

We believe today’s cohort of robotics startups will continue to stagnate in pilot testing unless they can overcome these obstacles. This is critical because robotics lacks the structural data advantages that helped other AI fields advance—fields that still took decades of financial and technical investment to mature. While the potential in robotics is significant, we expect progress to be a resource-intensive process more akin to the long development of autonomous driving, rather than an abrupt shift.

We share our findings to help investors scope the robotics frontier and identify the most important challenges, risks, and paths forward. Breaking down our approach:

Section 1: Provides our framework for investors to orient themselves on the robotics frontier.

Section 2: Analyzes exactly what current AI can (and can’t) do for the robotics stack, and what it will take to build generalist robots.

Section 3: Explores the challenges investors should expect robotics companies to face in breaking through the commercialization barrier.

Section 4: Presents our perspective on investable opportunities.

Section 1. Open Worlds are the Robotics Frontier

The robotics frontier has always been about progressing from constrained to open worlds. 

The best framework for investors to understand robotics is to look at how the evolution of software pushes robots out of factories and into new verticals. The ultimate prize is the ability to successfully operate in open worlds—places where robots encounter previously unknown things, locations, and people. 

Software has been the greatest limiting factor in this progression. Though there have been immense improvements and cost reductions in hardware over the past 70 years, better software will enable robots to move beyond assembly lines and into everyday life. The earliest robots could only complete simple, precise tasks in highly constrained environments on factory assembly lines. As robots were equipped with better software for perception, navigation, and control, they could be deployed in marginally more fluid environments, like warehouses. Further improvements could push robots into truly open environments—homes and unpredictable public places, alongside people.

Constrained worlds: concentrated markets for mature technologies

Constrained, closed worlds are systematic and predictable places where proximity to people is unsafe or unnecessary. Factories where robots can work in fixed workcells, often literally caged for safety, are the prototypical constrained environment. Even today, these robots rely on well-established software paradigms that don’t involve any learning or adapting. They follow predefined tasks where everything around them is kept as consistent as possible. 

Because of their relative simplicity, factories were the first environments where robots were deployed. The first industrial arm actually required no programming language—no real software at all. It was made by Unimation, the first robotics company founded in the 1950s. Unimation created the ‘Unimate’ heavy-duty arm which began earning its keep in 1961 by unloading metal castings from a die-cast press on an assembly line of the Ternstedt GM plant in New Jersey.^[1] Unimation itself didn’t last long—its hydraulic arms were quickly out-competed by electrical ones. But since the 1960s, industrial robots have proliferated for routine manufacturing. 

Manufacturing robotics is a mature, highly consolidated, and slow-growing space with little early-stage investment activity. Globally, $30–40bn is spent each year on purchases of new industrial arms. But this is a flat market, with sales growing at only 5% for the past six years globally and at less than 1% in North America.^{[2] [3]} Most industrial robot arms are sold by four incumbents—ABB, Fanuc, KUKA, and Yaskawa—which together hold 75% share.^[4] None are pure-play robotics companies—all are generalist suppliers of automation equipment that have been operating for much longer than the history of robotics. Because industrial robot arms are such a mature technology, there’s little startup activity in this space. Over the past decade, less than 10% of robotics VC funding has gone to industrial robots.

Structured worlds: active investment in current innovation

Outside of factories, there are opportunities for robots in places that aren’t as perfectly systematic as an assembly line, but are still highly structured. These structured world environments are often routine workplaces like warehouses, industrial sites, hotels, or office buildings. Here, robots must adapt a little bit but don’t face major changes to their surroundings or to the task at hand. They need some flexibility, but not full, open-world reasoning. 

Ingrained structure helps ensure that robots can complete high-level tasks while adapting to minor changes. A robot arm unloading items from a palette must adaptively decide how to pick up various objects that are jumbled together. A mobile robot in a warehouse must navigate freely through aisles while avoiding people. Consistent surroundings help robots adapt reliably. In a warehouse, good lighting, standardized aisle spacing, and orderly workflows like one-way aisles help robots avoid collisions. In hospitals and hotels, ADA-accessible interiors with wide doorways and elevators allow robots to maneuver around people.

The market for structured world robots is smaller than the industrial manufacturing market, but faster growing. Robots for these environments see about 200k new annual installs, 30% of which go to warehouses.^[2] These robots represent a $3.5–5bn annual revenue opportunity in terms of flat purchases. However, most are sold through subscription-based Robot-as-a-Service business models where customers rent robots, paying different rates for working and idle time. The size of this RaaS fleet is growing at over 20%, which is 4x faster than the industrial manufacturing market.^[2] 

Structured environments have been the realm of VC-backed robotics startups rather than older, self-sustaining incumbents like in the industrial robotics market. Half of such companies today have fewer than 50 employees.^[2] Early stage investor attention has concentrated on this segment for the past decade, accounting for over 90% of robotics VC funding since 2014. 

Structured world robotics represents many distinct applications across verticals, with varying degrees of technological maturity and market penetration. But across the board, their market penetration is <5% overall. So the potential for further growth is vast, with even cautious estimates placing it at over $300bn in revenue opportunity.

Warehouse robots

Warehouse robotics is one of the most commercialized applications of robots in structured environments today. In warehouses, robots require a minimal amount of additional adaptability to succeed, compared to busier places like hospitals or hotels. As a consequence, warehouse robotics companies have been the best bet for early stage investors in the past. Though returns are modest by VC standards, warehouse robotics has seen many of the most successful exits of robotics startups founded in the previous decade. 

Because warehouses are one of the more established robotics verticals outside of manufacturing, robots also have a fairly established role to play. Most common are autonomous robotic carts, such as those made by startups like Robust.AI or Locus Robotics.^[5] Autonomous carts are used to make independent deliveries across the warehouse, or follow a person as they pick items from shelves. Warehouses also make use of robotic arms for pick-and-place tasks like removing items from a palette or unloading boxes from a truck. Robots that can both move around and handle items are called ‘mobile manipulators’. For example, Agility Robotics frames its humanoid robot Digit—which is currently pilot testing in warehouses—as a mobile manipulator.^[6] While Digit’s legged design makes it humanoid, Agility sees Digit’s underlying purpose as the next-gen way to combine the core functionalities of movement and handling of objects.

Robust.AI

Mobile robots can work alongside people in warehouses, like the Carter made by Robust.AI.

About 80k new warehouse robots are installed each year globally, usually rented via a RaaS business model.^[2] This amounts to $2bn of new ARR coming online annually. And with only 5% market penetration in warehouses globally, this revenue opportunity could expand by tens of billions as demand for warehouse robotics accelerates.

Figure 1 Acquisitions of startups making robot arms for industrial or logistics applications, and those making warehouse robots. Though returns are modest by VC standards, they represent the most successful venture exits of robotics startups founded since 2014. Source: Pitchbook.

Autonomous vehicles

Autonomous vehicles are another application of robotics in structured environments that have seen early commercial success, capturing a small portion of expansive markets. Though a great variety of things happen on roads, they also have a lot of natural structure. Roads have rigid dimensions, consistent signage, reliable maintenance, and relatively strict behavior patterns. Because driving is so well-structured already, autonomous vehicles made initial progress very quickly. Serious efforts toward self-driving began with the DARPA Grand Challenges—government-sponsored research competitions wherein cars were autonomously racing through empty urban streets as early as 2007.^[7]

These first research efforts have grown into several multibillion dollar companies. Challenge winner Sebastian Thrun went on to lead Google’s Self-Driving Car Project in 2009, which evolved into the $45bn robotaxi company, Waymo. Several other autonomous trucking and logistics startups have reached unicorn status, like TuSimple and Nuro. Tesla takes a different approach, gradually increasing the autonomy available to its large base of 5 million customers through Tesla Fully Self Driving, which received a major overhaul and performance improvement in mid-2024.^[8]

While autonomous vehicles made relatively strong commercial progress versus other robotics applications, they still have a miniscule share of enormous markets. The rideshare and automotive markets are both hundreds of billions of dollars, and autonomous vehicles have barely begun to disrupt them. Today, autonomous driving companies are prioritizing software for their vehicles to adapt to rare situations like emergency vehicles, snow, cyclists, and road debris. This adaptability will allow them to operate in more geographies, enabling them to continue to encroach on legacy markets worth more than $500bn today.

Hospital delivery robots

Hospital delivery robots are another structured environment application—but one that has not seen as much commercial success. Hospitals have many routine workflows, like medication and linen distribution. They feature accessible interiors with wide hallways and consistent lighting. However, they are much more unpredictable than warehouse aisles. Hospital robots must be adaptable enough to politely get out of the way during an emergency, deal with blocked hallways, enter crowded elevators, and avoid mischievous people. 

Hospital delivery robots have been in development for over 30 years. The first was built by Joseph Engelberger, who also founded Unimation, the first industrial robot company. After Unimation, Engelberger created HelpMate, a refrigerator-esque robot for delivering linens and meals. HelpMate entered revenue-generating pilots in 1991, with its robots rented out for about $12 in today’s money to compete with labor costs.^[9] The company went public in 1997 with a fleet of ~100 robots and $4.5mn in revenue, but was acquired three years later by Cardinal Health and its robots were discontinued.^[10]

In the 30 years since HelpMate, several companies have launched more and more advanced hospital delivery robots. These include startups like Aethon (2002), Savioke (2014), Robotise (2016), and Diligent Robotics (2017), as well as incumbent projects like Panasonic’s Hospi robots. These companies built robots with the same fundamental form factor as HelpMate (though Diligent Robotics added an arm) and sold them with the same subscription-based business model, but have not achieved broad commercialization. 

Hank Morgan, Science Source; Lynn Nguyen

Hospital delivery robots carry linens, medication, or meals. They have been in development since the HelpMate robot (left) found its first paying customer in 1991. Today, robots like Diligent Robotics’ Moxi robot (right) use a similar form factor. 

If these companies can break through the commercialization barrier, they unlock a significant opportunity. Robots providing courier services inside hospitals have a clear value proposition—they can return tens of thousands of hours to clinical teams per year to spend with patient.^[11] Even a small fleet of delivery robots in every US hospital could represent $3–6bn of ARR. But hospital delivery robots’ long history of modest scale shows that the more adaptable a robot must be, the harder it is to commercialize.

Open worlds: the frontier

Open worlds are unpredictable places where anything can happen—in which robots encounter many possible tasks and interactions. The quintessential application of robots in open worlds is as consumer products in people’s homes—but it includes any highly variable place like sidewalks, airports, or shopping malls. In open worlds, robots need adaptable software that’s beyond the frontier of what works in orderly warehouse aisles or hotel hallways.

The challenge of building good robots for highly variable home environments means that almost no robots have succeeded in the consumer market. Many consumer robotics startups of the past decade have gone out of business without securing exits or further funding.^[12] Among incumbents, Bosch and Amazon both incubated ~$1,000 home robots—Mayfield Robotics’ Kuri in 2017 and Amazon’s Astro in 2021—both since pulled from the market.^{[13] [14]} These robots were meant to be generalists. Most were designed with a broad product surface area: games and entertainment, voice-enabled assistant services, audio or video calls, and video monitoring of pets or loved ones. In general, they didn’t gain traction because they weren’t uniquely good for these tasks compared to non-robotic solutions like home security systems and smart speakers.

Courtesy of Amazon, Embodied, NurPhoto, Seb Daly

Figure 2 Building good robots for the home is incredibly difficult. Consumer robotics startups and robots launched by large tech companies have both floundered. Of this set, the only one still available is Amazon Astro, which can be purchased on an invite-only basis.

The one exception to the fraught history of consumer robots is robot vacuums, which were designed to be specialists, not generalists. Robot vacuums have succeeded as consumer products because they’re limited to a relatively constrained task and can largely ignore any potential disarray around them. They’re an established technology that has sold millions of units per year since iRobot launched the first Roomba in 2002. Today, 93% of all consumer robots sold are robot vacuums (most of the non-vacuums are toys and educational tools).^[2]

Outside of robot vacuums, the consumer robotics market remains largely untapped as technological limitations have kept robots uninteresting to consumers, and unattractive to investors. But if its potential were fully realized, consumer-facing robots in homes and other everyday public places could become a huge market. People spend about $5bn every year globally on robot vacuums. There could be many other billion-dollar opportunities for specialist consumer robots. In fact, the Founder and former CEO of iRobot—Colin Angle—is launching a new home robot startup which raised its first round in December 2024.^[78]

Investor attention is shifting to the frontier

While open world robotics has always been the frontier from a technical perspective, it’s only within the last year that venture investors have been willing to seriously back startups in the space. It’s now become the hottest investment area for early-stage robotics companies. Frontier robotics startups approach open worlds from a variety of angles. Some aim to build the next generation of home robots or to fulfill the vision of generalist humanoids. Others are focused on tackling the AI layer itself and building the software necessary to power open world reasoning.

Figure 3 Robotics venture funding has primarily risen for the past decade. Funding decreased for two years after the 2021 secular peak in VC funding overall, but bounced back in 2024. Source: Pitchbook.

Megarounds raised by frontier robotics startups are changing the face of robotics capital markets. In general, robotics venture investment is increasing and becoming more concentrated. Funding decreased for a few years after the 2021 secular peak in VC funding overall, but bounced back in 2024 to a new high of $5.2bn. This recovery coincides with higher concentration. In 2021, the top decile of early stage robotics deals accounted for less than 50% of the total early stage robotics funding. But in 2024, the top decile of rounds accounted for over 80% of the total funding. Frontier startups explain this recovery and increase in funding concentration. Just six of them accounted for over one third of all venture funding for robotics in 2024. Two companies, Physical Intelligence and Skild AI, develop AI models for robots. The rest focus on open-world or generalist robots themselves.

Valuations of these frontier robotics startups are also growing quickly. For example, Physical Intelligence increased more than 4.2x in value in the seven months from March 2024, when it came out of stealth, to October 2024, when it released its first prototype model. Humanoid robotics company Figure AI increased its valuation nearly 5x in less than a year.

Figure 4 Robotics investors are shifting their attention to the frontier. In 2024, just six early-stage startups accounted for one third of all robotics venture funding. Source: Pitchbook.

Figure 5 Valuation step-ups for Physical Intelligence and Figure AI. Among all US startups in Q2 2024, median seed to Series A step-ups were 2.8x in two years. Median Series A to Series B step-ups were 2.3x in two-and-a-half years. Source: ^[77] Pitchbook data.

How iRobot sold a million Roombas in two years

iRobot launched the first Roomba in 2002 and sold a million robots within two years, a feat since unrepeated by any robotics company.^[16] Whereas many other robot vacuum companies built sophisticated, expensive robots, iRobot’s rapid early success came from building a maximally simple, affordable one.

Selling affordable, good products isn’t groundbreaking, but finding the right approach to do so in robotics is challenging. iRobot cycled through 14 failed business models from its founding in 1990 to the release of Roomba in 2002.^[17] The company’s first business plan was ‘build a robot, send it to the moon, and sell the movie rights’. This and other failures helped iRobot accumulate the expertise needed for Roomba’s success: iRobot learned how to manufacture at scale from its partnership with Hasbro (failed business model #3), learned how to clean floors from making industrial cleaning robots (failed business model #8), and learned how to do simple navigation from making bug-like mine detector robots for the military (failed business model #11). In fact, iRobot’s military robotics business helped support the early years of Roomba with revenue from defense contracts. Once Roomba could hold its own, iRobot spun off this division for $45mn in 2016.^[18] Today, iRobot’s military reconnaissance robots are sold by the sensor and equipment company Teledyne FLIR.^[18]

The lessons from its failed business models led iRobot to create a product that was technologically years behind the frontier. Roomba’s early competitors included nascent mapping and navigation algorithms that added expense and user hassle. In contrast, early Roombas used simple sensors and random navigation to keep costs low. This allowed iRobot to make a product more than 10x cheaper than the competition.^[17] The first Roomba was $199, which was intended to be just below the threshold at which you need to call your spouse before making the purchase.^[19]

iRobot’s revenue has fallen in recent years due to the expiration of key patents and loss of market share in Europe and Asia. Amazon planned to acquire the company in 2023, but the acquisition failed after being blocked in the EU. However, the lessons of the Roomba’s rapid success are worthy of reflection. 

It’s quite possible that the next consumer robotics company to sell a million units will use a sophisticated combination of very new technologies. But another strategy might be the Roomba approach—choosing the right problem and using tech years behind the frontier to solve it with an affordable, reliable product. Either way, iRobot’s remarkable, rapid commercialization is the seminal example of how product sense and early commitment to the right price point are essential to scale robots.

Figure 6 iRobot successfully scaled the Roomba incredibly quickly. The company committed to a simple technical approach that ensured the first Roomba was 10x cheaper than the competition. It also relied on another business line, making rugged military robots, to support Roomba’s early scaling. Source: iRobot public filings.

Section 2. Robots are Hardware Held Back by Software

From a roboticist’s perspective, hardware and software are equally important and interdependent. Robotics depends on integrating many mechanical and digital systems for robots to sense, think, and act in the world. From an investor perspective, we think hardware is nontrivial, but software is the more important barrier to commercialization today.

Hardware is already trending in the right direction—consistently getting better and cheaper. Fifteen years ago, the most cutting-edge home robot was the $400k humanoid PR2 (Personal Robot 2) made by the seminal robotics research company Willow Garage. When Willow engineer Melonee Wise left the company to develop more cost-effective platforms for Willow’s open-source software, she got the cost of similar robots down 10x to $35k by 2015.^[20] Even the most advanced humanoid robots today are expected to cost half of the original PR2. Agility Robotics’ Digit is $150k to purchase and Elon Musk intends Tesla’s Optimus humanoid to cost only around $20k.^[21]

With these hardware improvements, commercializing robots is all about connecting them to next-generation software so they can reach their full potential. In the last few years, LLMs have commercialized a new level of flexible workflows for abstract things like language and code. But these text-native models alone don’t automatically provide new capabilities for robots in the physical world. Companies working on AI for robotics face many technical challenges to give robots the versatility and competence we expect from AI today. They need to gather real-world robotics data with unprecedented scope and sophistication, while also developing a real-world robotics business to benefit from that data.

Robotics AI needs physical intelligence

Traditional approaches to deep learning in robotics devour time and resources, with no economies of scale. This is exactly what limited HelpMate Robotics’ commercial success with hospital delivery robots in the 1990s. Its robots relied on an ‘expert system’—a hard-coded logical flowchart of over 1,600 rules that needed manual updates every time a robot made a mistake.^[9] This process meant no machine learning for the robots, and no economies of scale for the company. Now, robots can learn, but usually rely on distinct programs for specific tasks, based on either manually labeled data or many hours of manual human demonstration. For instance, it can take a handful of hours’ work to complete 300 to 400 separate demonstrations (and perhaps an overnight data processing session) for a robot to learn to rotate a Rubik’s cube or flip a pancake.^{[22] [23]}

This approach doesn’t compound—teaching a robot one skill usually doesn’t make teaching the next any easier. Nor does it take advantage of what the robot has already learned. The labor intensity of data generation means that robotics datasets aren’t broad enough for robots to generalize. Narrow data causes robots to overfit, meaning small situational differences, like an object’s weight or table height, can lead to failures.^[24]

A frontier alternative to single-skill learning is the foundation model approach. Models like LLMs trained on an enormous ‘foundation’ of data have the scope to jump between professional emails, poetry, and code. The idea is that a similar tool could allow robots to adapt and generalize in the physical world. 

The prize to chase here is a generalist model for physical intelligence. In fact, one of the most valuable companies working on this problem today is simply called Physical Intelligence, and it’s valued at $2.4bn. Physical intelligence is our understanding of things and space in the world—the instinct we use to gingerly squeeze an overripe tomato, swirl clear liquid in an opaque cup, use a paintbrush or a paint roller interchangeably, and safely handle a hot drink. These tasks are hard for different reasons—adding a nut to a bolt requires precision, but handing someone a sharp tool requires consideration.^[24]

In many ways, developing production-ready AI for physical intelligence is a much higher bar than today’s systems like chatbots, image generators, or code generation tools. Robots need real-time reasoning to make dynamic, fast decisions and operate safely. Additionally, errors and hallucinations are far more safety-critical for robotics compared to other AI applications. While hallucinations from language and image generators are frustrating, analogous issues in a free-roaming robot that works alongside people could be quite dangerous.

We don’t have a commercialization-ready AI model for physical intelligence today, but we have a good toolkit. Transformer architectures—like the ones that power LLMs—have already proven themselves capable in robotics applications. But architecture alone doesn’t unlock physical intelligence; instead, it provides good scaffolding for robotics companies to tackle data challenges.

Apptronik

The uncrossed bridge between bits and atoms

While transformers show promise, there’s no free lunch in the leap from abstract to physical intelligence. Robotics has not seen as much progress as other domains in a post-generative AI world because today’s best models still struggle with 3D reasoning.^[25] LLMs are good at following instructions within the guardrails of a turn-taking text conversation, but can’t quite keep up with dynamic, fuzzy tasks in the physical world. They disappoint when it comes to realizing that ‘over there’ could mean a single shelf or an entire room, and that ‘the red cup’ might have moved since the last time it came up in conversation.^[26]

LLMs don’t fully solve robotics problems because robotics data is much more complicated than text.^[27] Even simple physical tasks could combine camera feeds with a robot’s sense of where its limbs are (from encoder sensors in its joints) or sense of touch (from force sensors in its hands). We’re still exploring the best way to wrangle and tokenize these multimodal inputs. For example, the extent to which robots could use computer vision alone, without a sense of touch, is still being explored.

Because of these limitations, most AI applications in robotics today are point solutions that focus on what models are already good at: words and images. They require a mediation layer connected to conventional robotic control systems to carry out actions. For example, DeepMind’s PaLM-E model uses an LLM to create high-level instructions that are then fed to a traditional, lower-level robot controller.^[28] Microsoft’s LATTE model uses an LLM to adjust a robot arm controller based on natural instructions like ‘go faster’.^[12] Robotics AI startup Cobot similarly uses a translation framework to convert language input like ‘we’re running low on supplies’ into precise commands that robots can execute.^[29] These solutions enhance parts of the robotics stack that use language and images but don’t meaningfully cross the bridge between bits and atoms.^[27]

Outside these point solutions, researchers have developed prototype full-stack robotics models that are good proofs of concept, but still quite limited. Many use turnkey vision-language models ‘grafted’ to a relatively tiny amount of robotics data. The result is technically a full-stack robotics model. However, the combination of lots of language data and little real-world data makes for a model with broad language abilities and narrow physical abilities.^{[30] [31]}

Grafted models enable robots to understand more variations of the behaviors in their limited wheelhouse, but don’t help them learn new behaviors. A robot could parse that ‘pick up the soda can’ means the same thing as ‘pick up the shiny thing’. But the robot wouldn’t then know how to pick up a tissue, plug in a USB drive, or guess what to do if the soda can rolls off the table.^[32] For example, the first Gemini Robotics model, released in March 2025, was trained on shirt-folding tasks using children’s shirts in a small range of colors. Impressively, it successfully generalizes to fold shirts of unseen colors, sizes, and sleeve lengths. However, it often fails to fold shirts that are rotated 180 degrees or presented face-down on the table, since neither situation occurred in its training data.^[81]

A key lesson from these grafted models is that even their initial progress was incredibly resource intensive. Take Google DeepMind’s Robot Transformer models RT-1 and RT-2. They were trained by combining a vision-language model with a small amount of robotics data, representing just eight simple actions like ‘knock object over’.^[31] But compiling enough examples of these actions took over a year of data collection from a fleet of 13 robots. The first Gemini Robotics models similarly required eight months of data collection from a team of 35 human operators. This model focused on a set of five more nuanced actions—including tying a shoe, stacking dishes, and inserting gears into a socket.^[81]

Commercially viable AI for robotics must break this mold. It’s easy to recognize that solving the limitations of AI in the robotics stack would be an incredibly valuable enabling technology to push robotics forward in new applications. Early-stage companies and their investors are turning attention to this gap in the robotics stack.

The (robot) arms race for physical intelligence

An arms race took off in 2024 to bring the capabilities of generative AI to physical intelligence. Robotics AI companies raised over $1bn last year—4x more than the previous four years combined. Frontier startups are beginning to release models based on their initial progress. Robotics AI startup Physical Intelligence released its first model, called 𝜋₀, in October 2024.^[33] Humanoid company Figure AI released its first model, called Helix, in February 2025.^[34] Both have a similar two-part architecture. The first part uses a vision-language model in conjunction with proprioceptive data about robots’ limb locations. This is then fed to another system (which Physical Intelligence calls an ‘action expert’ and Figure calls a ‘visuomotor policy’) that translates these inputs into new movements.

Tech incumbents are also reinvigorating their frontier robotics work. OpenAI is rebooting its internal robotics research team after previously disbanding it in 2021. Google DeepMind is investing heavily in robotics projects and has begun releasing Gemini Robotics models. Amazon acqui-hired robotics AI startup Covariant, which released its first model in May 2024 for pick-and-place robotic arms in warehouses.^[35]

Other incumbents are focused on winning share with platform products for robotics development. NVIDIA is vying to do so with its Isaac platform,^[36] an everything-but-the-robot ecosystem. Isaac is expansive already; it includes tools for embedded computing, training robots in simulation, CUDA-accelerated versions of familiar open-source software, and Gr00T, a research initiative to develop general-purpose robotics foundation models.^{[37] [38]} Meta has also founded a new team within Reality Labs to work on robotics tooling. Meta isn’t reportedly interested in building its own robot and is instead focused on how it can leverage its existing work on VR to create a robotics development platform.^[39] Hyundai’s Boston Dynamics is also developing a simulation and training platform for machine learning in robotics.^[40]

Solving the AI layer for robotics can enable robots to take on previously impossible tasks and perform routine tasks in previously unworkable places. The market for all non-manufacturing robots today is less than a $10bn annual revenue opportunity. But better enabling software that increases adoption could expand this market by hundreds of billions. Similarly, fewer than 20% of US households have a robot vacuum. It could easily be worth tens of billions for a new robot to reach every US household with a budget for smart appliances.

Figure 7 Investment in frontier robotics startups working on generalist AI models increased by nearly 10x from 2023 to 2024. Source: Pitchbook.

What will it take for robotics AI companies to build foundation models?

The principal challenge facing frontier robotics AI companies is building representative datasets. This means collecting data that captures the full range of situations a robot will face—especially infrequent, but mission-critical ones. Outside of robotics, AI models can adapt and generalize because we’ve already gone through the process of collecting internet-scale datasets, and curating them to span enough rare and everyday situations. Likewise, we need an abundance of realistic robotics data to build AI good enough to expand robotics markets.

Figure 8 The scale of robotics data today lags the scale of data used to develop LLMs. Training LLMs relied on curating vast datasets like Common Crawl. Analogs to Common Crawl in robotics, like DeepMind’s Open X-Embodiment Project, are still many orders of magnitude smaller. Source: Positive Sum analysis.

Other AI applications have seen breakthroughs because they benefited from large amounts of naturally representative data. LLMs benefited from many rich, realistic sources of text—social media, Wikipedia, Github, and the digitization of books. When ChatGPT was released in November 2022, the size of language training datasets had steadily grown by 10,000x in a decade.^[43] Much of this growth was achieved by the nonprofit Common Crawl, which maintains an immense open source repository of text. Commercializable LLMs would not exist without Common Crawl.^[42] The best LLMs today rely on highly curated subsets of this internet-scale slush pile of language for the majority of their training data—like Hugging Face’s FineWeb, or EleutherAI’s The Pile. 

Much like language, governments and other institutions have been gathering internet-scale autonomous driving data since the early 2000s.^[43] The size of these datasets has also steadily increased, by 10,000x in the past decade.^[43] By 2016, open-source driving dataset Oxford Robotcar was already 23 terabytes—about half the size of what it was rumored to take to train GPT4. That same year also marked Waymo’s first driverless launch on public roads. Since then, we estimate that Waymo has steadily collected over 130 petabytes of real-world driving data. 

Outside of the special case of autonomous driving, robotics lacks the internet-scale data collection efforts that provided the fuel for AI in other applications.^{[24] [44]} The best analog to Common Crawl or Robotcar today is DeepMind’s 2023 Open X-Embodiment Project (OXE), which combines small robotics datasets from 21 academic institutions.^[45] But OXE is only about one thousandth the size of Common Crawl or Robotcar, and about one ten-thousandth the size of the data that we estimate Waymo has collected. It would likely take ~2,300 more years of robot ‘man-hours’ for OXE to reach the size of Common Crawl, and 20,000 more years for it to reach the size of Waymo’s real-world data. 

Building AI for robots will require collecting much more data before we can curate a representative subset of the variances and corner cases robots will encounter in real-world situations. For example, much of the OXE data isn’t that realistic—toy wooden blocks appear 8x more often than kitchen tools. Already, researchers have curated a better subset of OXE, dubbed OXE Magic Soup.^[33] These early efforts foreshadow that robotics AI companies are just beginning serious, systematic data collection efforts that will require several orders of magnitude more data before models can adapt well in the real world.

The gridlock of real-world robotics data

Robotics AI companies need representative data to build good software. But the only place to get it is from robots in realistic situations. However, the software and data limitations we noted previously mean that it’s challenging to build a robot that can even be deployed into the real world to collect meaningful data. 

There are multiple strategies to break the stalemate. Some companies are prioritizing data first, and deploying robots second. They are focusing resources on releasing models trained with a fleet of in-house robots. For example, startup Physical Intelligence is collecting its own data from an in-house fleet of robots practicing various tasks—like unloading washing machines. It raised $470mn since its seed round in March 2024, and has already collected 10,000 hours of its own robotics data (about one year of robot man-hours).^[33] Figure AI’s Helix model was similarly trained on 500 hours of data from manually operated robots.^[34] 

Another strategy is to prioritize deploying robots first, and models second. Some companies, like Cobot, aim to build robots that can operate at a level of quality that’s nevertheless sufficient to drive sales. The business doesn’t need frontier AI models, but would benefit from them as soon as they are created.^{[23] [46]} The idea is that as soon as a robot is good enough to be worth paying for as a real product, realistic data becomes a byproduct of revenue.^[47] It might take 2,300 more years of robot man-hours to make a Common Crawl-sized robotics dataset, but we estimate that if every Walmart in America had just one robot restocking the shelves for just one hour at night, they would reach this in less than two months.

Simulation and synthetic data will do the heavy lifting for robotics AI companies building datasets and training new models, but they won’t solve the last mile in open worlds. Robots can learn a lot in simulation, but still need real-world experience to get the details right. This is known as the ‘Sim2Real training gap’, in which behaviors trained in simulation don’t quite hold up in the real world. For example, Agility Robotics found that humanoids trained to walk in simulation would lose their balance in the real world because of variations in the grippiness of real ground they couldn’t simulate.^[75] Simulation is also better at representing simple, structured places. For example, Robotics AI company Cobot reported that its robots learned to navigate a warehouse environment almost entirely in NVIDIA’s Isaac Sim simulator.^[38] But simulating a well-lit, clean warehouse is much more reliable than trying to simulate the endless chaos of a busy kitchen. So real-world data will remain especially important for frontier robotics companies working on open-world applications.

Autonomous driving foreshadows what it will take to solve physical intelligence

For investors to predict the path to future success in AI for physical intelligence, it’s helpful to understand how data challenges were solved in the specific case of autonomous driving. Autonomous vehicles—especially robotaxis—provide a good peek into the future of generalist robots because they’re ahead in both the transition towards generalized AI, and commercial presence in consumers’ everyday lives. 

Just like with physical manipulation tasks for robot arms, previous approaches to AI in self-driving involved distinct systems for single tasks like lane tracking, object detection, navigation, and the human interface.^[43] Today, companies are prototyping full-stack alternatives that combine all these functionalities into a single system. Tesla recently shifted to this approach with the Full Self-Driving (Supervised) v12 release,^[48] replacing an older system with a single end-to-end neural network.^[49] Waymo also unveiled a prototype full-stack model called EMMA (End-to-End Multimodal Model for Autonomous Driving) in 2024. This model generates brief driving sequences based on natural language input.^[50]

Curating representative, real-world data is key to success for autonomous vehicle companies—self-driving wasn’t built from racetracks or closed city streets. Companies have developed this real-world data flywheel in different ways.^[51] Waymo’s robotaxi service in San Francisco gathers data from near-complete autonomy in limited places. Tesla collects data from partial autonomy across its many customers in many locations. But both of these strategies involve natural access to real-world situations.

Autonomous driving companies have relied heavily on simulation, which has proven to be an accelerant, but not a full solution for real-world adaptability. Waymo and Cruise both relied heavily on simulation tools to parallelize training—Waymo logging 3mn simulated miles per day by 2016.^[52] But robust synthetic data and simulation efforts haven’t eliminated the need to gather realistic data from real deployments.

Zuma Press, Waltarrrrr

Stanley (left) was the winner of the 2005 DARPA Grand Challenge created by Sebastian Thrun and the Stanford Racing Team. Thrun went on to have a key role in leading the Google Self-Driving Car Project, which has since become the $45bn+ company Waymo (right).

Curating corner cases is the current bottleneck in autonomous driving

Building self-driving cars involves the same challenge as building general robotics models: gathering and curating large, representative datasets. Today, autonomous vehicle companies have an abundance of real-world data that dwarfs language and text datasets. For a long time, self-driving technology needed a brute-force data collection approach to improve performance.^[53] By today, Waymo has driven 20mn autonomous miles on public roads, which we estimate amounts to over 130 petabytes of data—about nine Common Crawls. Its Jaguar I-Pace cars generate more than a terabyte of log data per hour—the size of Open X-Embodiment in a single day of driving. Tesla has similarly collected data from over a billion miles driven by five million vehicles.^[8]

Now, autonomous car companies are focused on curating representative data, not simply collecting more of it. They throw out most of the data they collect in order to filter for rare, critical situations among the noise of monotonous, everyday driving. Waymo now deletes data it deems repetitive and prioritizes snow, emergency vehicles, and cyclists.^[54] In fact, Waymo is planning several ‘road trip’ deployments to places like Tokyo, where cars will be driven manually to expose the self-driving system to new environments and weather conditions.^[55] Like Waymo, self-driving truck companies Aurora and TuSimple also discard most of the data they collect, save for examples of unusual situations, like debris on the freeway.^[54] Tesla’s autonomous driving work also hinges on curating the right balance between common and rare events.^[57] Only about 1/10,000 of Tesla’s miles driven are useful for training models.^[56]

Robotic cars took 20 years and $200bn. Will other verticals be the same? 

The road from inaugural data collection efforts to commercial products in autonomous driving took 20 years and $200bn. It really began in 2004 with DARPA’s first Grand Challenge, a $1mn prize to autonomously navigate 150 miles of the Mojave desert. In the first year, no vehicle accomplished more than 5% of the course. But by 2007, Grand Challenge teams were reliably completing both all-terrain and urban races.^[7]

After DARPA threw in the first few million dollars of prize money, investors and car companies contributed nearly $200bn of additional funding over the next two decades. During this time, both automotive and tech executives made a swath of predictions for the arrival of self-driving, anytime from 2017 to 2030.^[58] Finally in 2020, Waymo opened its fully driverless service to the general public in Phoenix, and then in San Francisco in 2024. 

Autonomous cars have been technically capable of navigating urban streets since the early 2000s, and have been carrying people on point-to-point trips since the mid 2010s. But it took several more years for them to become safe and performant enough to launch a generally available service for consumers. We think that today’s AI for robotic physical intelligence is like autonomous driving in the early 2010s. It’s likely that much more time, money, and grit is necessary to make general robots into commercial products.

There are both reasons to be optimistic and hesitant that the data problem in robotics can be solved more quickly, and cheaply, than it was for autonomous driving.^[23] We have wildly better model architectures and machine learning infrastructure today. But lack of historical, internet-scale data collection creates meaningful friction. If other applications of robotics follow the exact same path, then we estimate it could take 7–10 years, and tens of billions of dollars, to bring them into our homes and everyday lives.

Section 3. Pilot Testing Does Not Guarantee a De-Risked Robotics Business

As we’ve shown, investors can assume that most robots are satisfactory hardware held back by unadaptable software. Investors should also understand that the symptoms of these software-driven limitations bring distinct risks. In particular, a robotics company reaching the phase of pilot testing should not signal the same de-risking point that it might for investors familiar with SaaS.

For most hard tech companies, launching a revenue-generating pilot is a critical de-risking point for investors. This is not the case for robots. Robotics companies that find early pilot customers quickly may still struggle to scale at the rate venture capital typically requires. Starship, a relatively successful sidewalk delivery robotics company, took 10 years and $218mn of VC funding to grow to approximately $20mn in revenue. Even robots launched by tech incumbents with distribution, funding, and an existing customer base have run up against the post-pilot commercialization barrier. Amazon has incubated two consumer-facing robot products in the last five years—the home robot Astro and the sidewalk delivery robot Scout—both of which have been dropped after years of small deployments (Astro is still available by invite only). 

Pilot testing retains more risk for robots, relative to other technologies, because piloting is part of product development. No robot can be prepared for real-world situations in a pristine lab. Robotics companies must pilot early to collect representative data about how robots should handle a variety of realistic situations, not just to evaluate ROI.

One might have naively expected low-hazard robots to be an easier problem than other hard tech. But investors should not underestimate the way the extreme unpredictability of unstructured places can erode a robot’s commercial horizon.

Why did sidewalk delivery robots fail to capture a $10bn opportunity in the pandemic?

A pertinent story about the riskiness of robot pilot testing is why sidewalk delivery robots didn’t take off during the pandemic. In the 2010s, it was expected that sidewalk delivery robots could capture material share in last-mile food delivery by 2025.^{[59] [60] [61]} A common bet was that even if sidewalk robots didn’t completely revolutionize the last mile, they would certainly arrive before robotaxis. It wasn’t irrational for investors to expect that small sidewalk robots could outpace full-size autonomous vehicles: sidewalk robots have many lower technical hurdles, cheaper inputs, lower safety requirements, and less fear or resistance from users.^[62]

By 2019, a cohort of over a dozen startups had raised $1bn—all with similar ‘cooler on wheels’ robots. They were joined by three corporate projects: the Amazon Scout, FedEx Roxo, and Postmates Serve robots. Then, March 2020 brought the tailwind of the century for these companies. Sidewalks emptied of crowds and demand for short-haul food delivery skyrocketed.

UberEats and DoorDash both sustained 100%+ revenue growth for the first two years of the pandemic, and grew their user base by over 20mn people.^[74] From 2019 to 2023, the food delivery market grew from $3bn to $13bn—a $10bn opportunity.

Pexels

Sidewalk delivery robots made by Starship Technologies.

But sidewalk delivery robots captured essentially none of this $10bn prize. All sidewalk delivery robots completed less than 0.1% of the total food deliveries since the beginning of the pandemic. Many of the dozen-plus original sidewalk delivery robotics startups have folded or pivoted. And this wasn’t for lack of capital, as these startups doubled their funding during the pandemic. Of the corporate projects, both the FedEx Roxo and Amazon Scout robots were canceled after several years of pilot testing.^[63] Postmates’ robotics division was spun out as Serve Robotics after the Uber acquisition and is operating with ~$2mn of revenue from its small fleet of 48 robots. 

The startups that still operate today have remained at modest scale in limited geographies. The most successful to date is Starship Technologies—which was founded in 2014 and rapidly reached its first pilot in the UK within a year. Starship’s US growth strategy has been to focus on college campuses—slightly more structured environments where sidewalks are more orderly and predictable than urban downtowns. In 2019, Starship announced its plan to have robots deployed on 100 US college campuses within two years.^[64] It raised $178mn towards this goal but only reached an additional 16 deployments in that time period. Today, it operates on 55 campuses, just 1% of US colleges.

Sidewalks are open worlds

The fundamental challenge faced by sidewalk delivery robot companies is that sidewalks are open worlds. Compared to roads, sidewalks are the Wild West. Roads offer a controlled, standardized environment with traffic rules, consistent signage, and reliable maintenance. In contrast, sidewalks include wheelchairs, strollers, children, dogs, narrow turns, first responders, and sometimes cars turning right on red—all of which causes serious trouble for delivery robots.^{[65] [66] [67]} It’s true that sidewalks have a lower safety barrier than roads. But their frequency of corner cases is much higher, even if the stakes are lower. This is why companies have fallen back on the simplest, most structured sidewalks they can find: walkable European towns in the summer, arid US cities with little pedestrian traffic, and well-groomed college campuses.

The story of sidewalk delivery robots should warn investors that even with plenty of funding and market-transforming tailwinds, the unrelenting complexity of open worlds makes robots risky. Delivery robot companies didn’t have the data to prepare their robots for all the disorderly situations they encountered, so they couldn’t break through the post-pilot commercialization barrier, even with plenty of demand. They show that pilot testing is only the beginning of product development for robots meant to coexist with people.

Figure 9 Taken together, all sidewalk delivery robots completed less than 0.1% of total food deliveries since 2020. Source: ^[74], Positive Sum analysis.

Commercializing humanoid robots

So far, we’ve outlined our framework for how software defines the robotics frontier and plays a key role in robots’ ability to break through pilot testing and succeed in new markets. Investors can use this framework to form expectations about new opportunities. 

Perhaps the hottest space today where it’s important to have realistic expectations is humanoids robotics. Humanoids are buzzing and well funded. Two startups—Apptronik and Figure AI—have already achieved unicorn status. Figure has raised the most money of any humanoid company, with $845mn of funding that amounts to 30% of US humanoid funding in the last five years. The company is currently in talks to raise an additional $1.5bn at a nearly $40bn valuation, which is a 15x increase in value since February 2024. 

It can be hard to discern the line between innovation and spectacle with humanoid robots. They naturally capture our attention because they look like us. When a humanoid robot completes a task, people often make generous assumptions about how much it mirrors our underlying abilities. A demo video of a humanoid opening a door might seem to imply that the robot could open any other door, or the same door with a different handle, or a refrigerator, but this likely isn’t the case.^[68] Humanoids are notorious for stagecraft demo videos with sped-up footage, remote control, or pre-scripted movements that exaggerate their capabilities. 

But humanoids aren’t all stagecraft. Within the past year, several humanoid robotics companies began their first deployments with paying customers. Figure and Sanctuary AI both completed pilots in automotive manufacturing within the last year. Figure has reportedly engaged a new customer, but hasn’t yet disclosed their identity.^[69] Logistics company GXO has begun pilots with fleets of Apptronik’s Apollo and Agility Robotics’ Digit, where the Digit fleet earns $30 per hour, per robot.^[70] 

This milestone means humanoid companies have just begun an important and early phase of their product development.^[71] In pilots, companies begin the process of learning lessons from how their robots function in the real places that they could never encounter in a pristine office, or even in simulation. As in other robotics applications, these pilots will be critical to generate the real-world data required to drive software improvements.

Lack of open-world AI is an especially acute bottleneck for humanoid robots because they are envisioned as generalists—able to accomplish a human-like variety of tasks in everyday environments. For a humanoid to be as good at doing dishes as it is at moving pallets around a warehouse, it will need incredibly reliable open-world reasoning. Building generalist AI capable of powering humanoids as a horizontal product will require data spanning a vast range of contexts and situations.^[73] Humanoid companies are increasing their focus on this problem. Apptronik announced a strategic partnership with the Google DeepMind robotics team, alongside the release of its first Gemini Robotics models^{[79] [80]} Figure AI had a similar partnership with OpenAI, which it has since ended in favor of building its own model.^[15]

Outside of the overarching software challenge, there are still many challenges to address through early deployments of humanoids.^[72] One thing to tackle is safety around human proximity—humanoid pilot tests today are taking place without people nearby. Another thing is speed and balance. Even the fastest humanoids run at only 20% human speed. To get faster, humanoids would benefit from better control and balance systems for things like slippery ground, carrying unbalanced boxes, or even swapping out their own batteries.^{[73] [75]} These are simple examples in a long list of tricky problems that humanoid robotics companies will need to address as they continue pilot testing.

Humanoids are the Formula One of robots

Developing humanoid robots will be a long road so investors should be cautious about expecting rapid commercialization. Most investors should not jump on humanoid robotics companies unless they’re comfortable investing in a product with a long research and development horizon. However, there may be a silver lining for the robotics industry as a whole, as humanoids could yield dividends for more readily commercializable technologies. 

Humanoid robotics companies are like the Formula One teams of robotics—they’re prioritizing a unique form factor at the expense of cost. Formula One teams allow car companies to ideate with a big budget and room to experiment. The goal isn’t to make these cars everyone’s daily driver. But plenty of Formula One tech has made it from racetracks to regular cars. This role as a research incubator is already held by robotics company Boston Dynamics. Famous for its sensational YouTube demos, it has been a subsidiary of Hyundai since 2021. Boston Dynamics develops the Atlas humanoid, which it sees as a research project intended to test the limits of what’s possible for other robots.^[76]

Ruthless pursuit of humanoids could change the landscape of robotics enabling technologies for the better. Humanoids push the boundaries of robots’ motion control, balance, obstacle avoidance, battery life, and safe human collaboration. Building actuators, sensors, and algorithms that improve these things for humanoids will make it much easier to develop new robots in general. It may become simpler for more smart people to start robotics companies and work to commercialize robots in a wider variety of places.

Figure 10 Several humanoid robotics companies have begun pilot deployments with early customers. Source: Positive Sum analysis.

Section 4. Find Compelling Software Strategies

Our framework for the robotics frontier is that the evolution of software will drive robots from their current applications in structured environments to new applications in open, unpredictable places. What makes robots exciting today is just how little value has been captured in these markets, yet how transformative robots would be if they reached their full potential. 

As robots gain market share in areas where they already have a foothold today—like in manufacturing, warehouses, or autonomous driving—they open up opportunities worth hundreds of billions of dollars. Beyond this, similarly vast opportunities exist in markets where robots have barely made inroads: working in homes and other populated places. Altogether, these current and frontier opportunities represent over $1tn. As robotic technology matures in these markets, it won’t just generate massive economic value—it will fundamentally reshape industries and redefine the technology that inhabits our homes and everyday lives. 

Investors should be clear on why these opportunities have remained untapped for so long. We conclude that software is the key bottleneck preventing robots from delivering real value in existing markets, and breaking into open worlds. Robots require open-world reasoning to navigate variable environments and respond to dynamic situations. This breakthrough in software, not hardware, is what constrains the robotics frontier.

To a great extent, the hardware for frontier robotics already exists, and it’s getting cheaper. The first humanoid research robot for the home, the wheel-based Willow Garage PR2, cost $400k and required $5k of sensors to operate. But within five years of its development, the cost of similar mobile robots with arms fell 10x to less than $40k. And today, some humanoids like Tesla’s Optimus are projected to cost even less, despite the added complexity of legs. Beyond humanoids, we have the hardware for a variety of robots to grasp objects and navigate spaces. If hardware had been the only bottleneck, it’s probable that hospital delivery robots like the HelpMate—first introduced in 1991—would be in every US hospital already. Neither the fundamental design nor business model of hospital delivery robots has changed much in 30 years, but commercializing them has remained a challenge. 

In contrast, the software for frontier robotics does not exist today. Current approaches to AI in robotics fall short of physical intelligence—the ability to understand variations in the physical world and adapt to a variety of physical tasks. But better open-world software for generalized reasoning could enable robots to operate in new environments and capture huge value in untapped markets.

These advances in software are what will ultimately break the commercialization barrier. Today, most robotics startups face a critical challenge: they can get their robots out of the lab and into early-stage pilots, but the software isn’t strong enough to scale them beyond that. Only two types of robotics businesses—those making industrial arms for manufacturing and robot vacuums—have successfully broken through this barrier. 

Our conclusion is that investors in robotics should focus on software. This doesn’t necessarily mean only focusing on software companies; rather, investors should look for robotics businesses with a compelling software strategy. We think there are three such approaches that could yield investable opportunities. First, investors looking to make visionary bets on the robotics frontier should seek out companies looking to crack the robotics data problem in open worlds. These companies are the ones developing next-generation AI models for robots, the infrastructure to train and support them, or generalist robots themselves. Investors looking to make more near-term bets should focus on how software pushes boundaries in structured worlds where robots have some traction today—like warehouses. In these markets, incremental software advancements can lead to ROI more quickly. Lastly, investors might consider companies that sidestep software limitations entirely by developing tele-operated (remote controlled) robots designed to rely on human operators.

Three robotics software strategies

1. Crack the robotics data problem

A cohort of companies are currently aiming to build AI models for physical intelligence, and the infrastructure to develop them. This software layer could become a keystone enabling technology for robots to enter into new markets and take on more flexible tasks in existing ones.

Physical Intelligence. Founded in 2024 by a team of researchers to build robotics foundation models.

Skild AI. Founded in 2023 by two CMU professors to build robotics AI models that can jump between different robot form factors.

Google DeepMind. Working extensively on AI for robots. Most notably, it runs the Open X-Embodiment Project to consolidate real-world robotics data.

NVIDIA Isaac. A robotics development platform product designed to provide everything needed to build and refine robot software in simulation.

2. Push the boundaries of structured world robots

Companies building robots for structured environments can operate where the growing pains of robots’ limited adaptability are the least felt. This space can hold investable opportunities today because these applications will be among the first to turn incremental AI improvements for robots into ROI.

Mytra. Mytra’s material flow system uses a 3D grid and a fleet of robots that carry pallets in any direction, enabling flexible and data-rich material handling.

Robust.AI. Robust.AI’s warehouse robots use touch-sensitive handles and a human-in-the-loop UI to enable flexible autonomy and sophisticated collaboration with people.

Diligent Robotics. Diligent Robotics makes the Moxi hospital courier robot. Moxi is designed to be polite and predictable in crowded hallways and elevators.

3. Forego autonomy

Companies can sidestep some of the challenges of scaling robots by developing a value proposition that’s not dependent on autonomy. This approach may hold investable opportunities because companies can design a GTM strategy that doesn’t buckle under software limitations. Many robotics applications rely on teleoperated robots in cases where customers don’t expect automation. For example, drones and other reconnaissance robots can be teleoperated in defense or search and rescue applications. Surgical robots, like Intuitive Surgical’s da Vinci surgical system, are operated by surgeons. Teleoperated robotics companies in big, regulated markets like healthcare and defense can differentiate themselves without requiring a moat based on data or novel AI.

Intuitive Surgical. Develops robotic systems for assisting in minimally invasive surgery. It has installed nearly 10k robots in hospitals worldwide.

Teledyne FLIR. Makes a variety of reconnaissance technology for defense and industrial applications. It currently sells the PackBot (originally developed by Roomba-maker iRobot), a small tank-like rugged reconnaissance robot.

Good questions to ask robotics companies, and good answers to look for 

What is your software approach for atypical events or flexible tasks?

For most robotics companies outside of manufacturing, commercializing robots hinges on enabling them to be flexible in real-world situations. There are many potential strategies to address this challenge. Robotics companies can focus on highly structured environments where robots’ required adaptability is low, design systems for humans to intervene easily when robots fail, or build teleoperated robots meant to be controlled by people.

What do you hope to learn from pilot testing?

Pilot testing robots is a key part of product development. Through pilots, robotics companies learn about how their robot operates in realistic situations and gather data about real-world corner cases. Companies should pick intentional design partners and pay close attention to unanticipated aspects of their robots’ performance and user experience.

How did you choose your price point?

To be accepted by users long term, a robot should deliver (or over-deliver) on the expectations inspired by its price point. For example, expensive home robots must not disappoint consumers, especially as they are a splurge purchase for most households. In enterprise robotics, customers like warehouses and hospitals must also evaluate opex savings against the risk of installing new systems. So robotics companies should keep price in mind from the very earliest stages of design. 

Why is a robot uniquely appropriate to solve the problem you’re working on?

Hammers looking for nails are abundant in robotics. Companies should have a strong reason why a robot is the right solution to the problem they care about, compared to non-robotic alternatives.

Click here to subscribe to print for your office or home.

Click here to subscribe to print for your office or home.

Each issue, we showcase someone doing the thing they seemed destined to do. We call these people Pathfinders because, beyond the direct impact of their creations, they set an example by identifying their life’s work—and relentlessly pursuing it.

Blake Hall’s path to founding ID.me began with his experience commanding an elite platoon of scouts and snipers in Iraq. When he returned home, he discovered a problem: veterans suffered from identity theft at nearly twice the rate of other Americans, while simultaneously struggling to prove their service status to access hard-earned benefits. Since its founding in 2010, ID.me has grown into America’s largest digital identity network, with over 140 million members, as it pursues a bold vision: a single secure login for everyone’s online identity.

Colossus Review: You were 23 when you were deployed to Iraq, in charge of 32 scouts and snipers in your platoon. How do you reflect on the experience of that responsibility at such a young age? 

Blake Hall: Carrying the lives of 32 people, and the hopes of their families, is an overwhelming responsibility. It’s like looking down at the ground from a great height—the enormity of what’s at stake can incapacitate you if you confront the reality of the situation directly. To do my job, I needed to let go of the scenario I feared the most, losing my men, to focus on the tasks that would reduce risk for my team.

There’s a paradox that you often have to let go of what you want the most in order to get it. In reality, I had to push what I wanted so deep down into my subconscious that it didn’t impact my performance. That mindset allowed me to influence the variables I controlled, while admitting I had no control over the external variables that might not go my way. I knew that if one of my men died under my command, I would only be able to live with myself if I had done everything I could to be the best leader for them. 

I’m so grateful that each of them came home to their families. After we got back, on the last night I led the platoon, two of my snipers pulled me aside. What they told me about how the platoon felt about me made me cry. I’m so glad that I didn’t let them down.

You’ve said that “the whole point of leadership is to be inversely correlated with context.” Can you explain what you mean by that, and how to do it? 

Most humans don’t like to work. On a day-to-day basis, most prefer what’s easy over what’s hard. But if you ask someone to look back on their life, many wish they’d lived a life that made a difference, versus an easy life. 

Leaders help teams believe in a mission that unfolds over a long period of time. By helping the team understand what their current work means for their future self and for others, they can shape the team’s perspective to care more about the totality of their life than about their current pain and suffering. To me, this is strategic leadership. Strategic leaders wake people up to their true potential. They animate the world. 

By helping the team understand what their current work means for their future self and for others, they can shape the team’s perspective to care more about the totality of their life than about their current pain and suffering.

–Blake Hall 

Along the way, there are ebbs and flows as the organization and teams fare better or worse over time. When things go well, recruiting and fundraising are easy, but teams get complacent and distracted. When things aren’t going well, some people lose faith and leave, so recruiting and fundraising get harder. 

Operational leaders are like gyroscopes. They acknowledge the context, communicate the big ideas, and orient the team to a better way of working together. When we were in combat and got the briefing about ‘God mode’, the risk-taking psychological state soldiers enter around month six of combat, my immediate thought was, “All gas, no brakes. Not good. Someone had better put on the brakes.” That someone is the leader. 

Hall with his sniper teams in the Green Zone in Baghdad, Iraq.

The military is renowned for building effective leaders. Where does the corporate world tend to go wrong? 

The military has a positive selection bias because it attracts people willing to give their life for others. For that reason, any organization with a worthy mission, based on service and purpose, will have the advantage of attracting good and capable people to it. 

The military does an exceptional job training leaders and establishing core competence in domains. Now, training isn’t always an efficient use of time in the military. However, the amount of time invested to train leaders stands out from corporate America. 

At the same time, our peacetime military suffers from many of the same issues that plague corporations. The best politicians, not the most skilled generals, tend to accrue power when times are soft. I think the same is true for many large organizations. 

When times are hard, the most skilled generals, like General Petraeus, rise to the top out of necessity. There are lessons in that for corporate America. If you can ensure your most skilled leaders are in charge at all times, you will produce outsized value. 

You said war “infused lightning in your bones”. Can you talk about how the experience shaped your sense of purpose in life when you finished your 15-month mission in Iraq? 

I am definitely an addict. Adrenaline is a powerful drug. Every mission brought a different kind of high. But I knew if I kept playing that game in combat, eventually I would die. 

The great gift combat gave me was to show me that life is short. The real risk isn’t founding a company with an ambitious mission that will likely fail; it’s ending up as a cog in the machinery of a huge bureaucracy, and getting stuck there. 

I thought a lot about my friends who gave their lives for America. They didn’t have the opportunity to go for it. I wanted to make my life count for them. 

You can’t take anything with you when you die. You can only make a positive impact on the lives of other people. That’s what will endure. And I think that happiness comes from helping others. So, that’s what I’m doing with every fiber of my being. 

When did it become obvious that you could, and would, devote the next phase of your life to the idea behind ID.me? 

When does a grain of sand become a pile? It’s like the sorites paradox—I knew I had something burning inside me by the second year of business school, after I got home from combat. I didn’t know it was ID.me specifically until late 2012. But I knew I would make a positive change in the world, or spend my life trying to effect that change. 

What are the most effective ways to instill a sense of mission across a large number of people in a corporate setting? How do you do it at ID.me? 

Humans connect with stories, not numbers. The most impactful way to reinforce the mission is to share stories of us living it. It’s about celebrating our teammates who have lived our cultural values to help a specific customer or person. 

How do you measure success in life? 

I measure success in business by coming through for people who have trusted me. 

The best phone call I ever made as CEO was to our first VP of Product. We hadn’t talked in a few years when I called him in 2021 to tell him about our Series C, and that we’d become a unicorn. He was driving home at the time, having been to see a financial counselor with his wife, to discuss how they would pay for their triplets to go to college. 

I told him he would be just fine because there was a secondary offer tied to the Series C and he was a multimillionaire. He took a deep breath, paused, and then asked me for a few seconds to pull the minivan over to the side of the road so he could cry. 

In my personal life, I want my children to want to come home when they’re older, to know that my wife and I love them unconditionally, to know right from wrong, and to be brave enough to chase their dreams. 

You’re allowed to invite four people (dead or alive) to dinner at your house. Who’s coming and why? 

Aristotle, Alexander the Great, Thomas Jefferson, and Benjamin Franklin. I love to learn. Aristotle mentored Alexander the Great. Jefferson and Franklin breathed the greatest experiment in the history of mankind into life: America. Civilizations, culture, humans, leadership, war, innovation—I’d love to ask questions and just listen. 

If you were making a study guide for a young entrepreneur, what three pieces of writing would you recommend they study? 

Steve Blank’s The Four Steps to the Epiphany explains the four stages of building a company by using the scientific method to validate graduation to the next phase. Geoffrey Moore’s Crossing the Chasm applies this method to targeting customer segments over time, according to individual and corporate psychographics. Clayton Christenson’s Disruptive Innovation theory explains market entry and how to unseat large incumbents. 

If a founder has deep expertise and differentiated skill, the wisdom in these three will give them all they need to win given enough time and grit—and assuming they don’t run out of money. 

Snipers in Dora, a neighborhood in Baghdad. Hall said: “If there’s hell on Earth, it was probably Dora.”

What unexpected plot twist in your life looked bad at first but turned out to be a gift? 

The decision to decline McKinsey’s full-time offer and start ID.me. I signed a $48,000 lease for a Washington DC townhome, where our technical team and customer support team could work, when I only had $3,000 in my checking account, $5,000 in our business account, and maxed-out credit cards I had used to fund the business. I sold my dream car to free up cash to finance the business. Things did not look good. 

You were heavily influenced by the examples set by your father and grandfather, who was a World War II hero. As you raise your own children, how do you think about passing down values while letting them find their own way? 

When they’re young, they need to be pushed to try things because humans have an aversion to trying new and different activities. Once they find something they love, then my job is to nurture that interest so they can chase their passions without regard for my own. 

My dad was a West Point grad. I got into West Point and Vanderbilt. He sat me down and said, “If you don’t know that you want to go to West Point, you’re going to hate the next four years of your life.” I picked Vanderbilt, and that was one of the best decisions I ever made in my life. I’m so grateful for that conversation, and the example he set. 

The real risk isn’t founding a company with an ambitious mission that will likely fail; it’s ending up as a cog in the machinery of a huge bureaucracy, and getting stuck there.

–Blake Hall

Do you have a morning routine? What do you protect most fiercely in your daily schedule? 

I do my best thinking in the morning. I wake up, make coffee, play with my kids a little, and then try to get to work as quickly as possible to get into flow on hard problems. 

What do most people not understand about identity and what you’re trying to achieve? 

We’re building Visa in 3D because identity should be everywhere you want to be, to make life safer and easier. Identity is so much bigger than just payments. 

Imagine going back to the early 1950s before Visa existed. You might know what the world should look like with Visa to streamline payments. But how would you explain Visa to merchants and everyday people in a way they would understand? 

At various stages of ID.me, people would say, “Oh, you provide a Single Sign On—but just for the military right?” That was at 10 million users. Now, they say, “Oh you provide a Single Sign On—but just for government websites right?” This is at 140 million users. 

In the spirit of ‘show, don’t tell’, Americans will begin to understand ID.me when they experience life events like retirement or taxes or buying a home; when they see ID.me’s green button makes it safer and easier to access the things they need. 

Among all your achievements, which one would you most like people to understand deeply, and why? 

I’m most proud that the men I led in combat respect me. To me, this is the ultimate test of leadership and far more important than any medal or award. Earning informal credibility from a group of people who are deeply skilled at their jobs gives me more intrinsic satisfaction than anything else because it indicates true skill in reality. And I believe the leader of any group should be the most skilled at leading that group. 

I’ve carried that mindset with me every day leading ID.me. I try to learn as much as I can, as fast as I can, where it matters the most—to help the team and earn my title. 

What product, service, or experience do you recommend more than any other? (Aside from ID.me) 

I used Starlink last week and it was so fast. So, if you’re reading, major airlines and Amtrak, please do all of us travelers a favor and get high-speed internet going everywhere!

JARED SOARES

Click here to subscribe to print for your office or home.

Click here to subscribe to print for your office or home.

“Sometimes I feel like I’m running the X-Men Academy,” says Matt Huang, describing his $12 billion crypto investment firm Paradigm as a place for brilliant mutants who possess unusual powers.

Take Charlie Noyes, the firm’s first hire. A 19-year-old MIT dropout who couldn’t operate a calendar, he rolled into his first 10am meeting five hours late, completely unapologetic; he’s now a general partner. Or Georgios Konstantopoulos, Paradigm’s CTO, who transformed from a World of Warcraft addict into one of crypto’s most prolific engineers. Or the developer known only by his X handle, transmissions11, whom Paradigm discovered on a Discord server while the anon was still in high school.

“They create an absurd amount of chaos sometimes and you want to pull your hair out,” Huang says. “But then you see what they can do and it’s like, holy crap. Nobody else in the world could do that.”

On the chilly morning I visited Paradigm in San Francisco, two of Huang’s team were working on a mechanism that could reshape how hundreds of billions in digital dollars move through the financial system. In a penthouse conference room, curved like the whispering gallery of a cathedral, Partner Dan Robinson tapped his Paradigm-green Nike Air Force 1s on the floor while explaining their latest breakthrough at the speed of a high-frequency trade. Research Partner Dave White, in hexagonal spectacles and a scraggly beard, was hunched over his laptop, pausing occasionally to discuss equations behind the idea he invented. Huang listened intently, his athletic frame still in a plain black Japanese sweater, his manner carrying the quiet authority of someone who’s always been ahead of the curve.

“Everything he touched was good,” says Doug Leone, who ran Sequoia during Huang’s time there from 2014 to 2018. “He’s super smart and incredibly humble. It’s tough not to meet Matt and come away with the impression that he is super special.”

Through two large arched windows high above San Francisco’s Union Square, traditional finance’s concrete towers loom to the east, while the startup sprawl of SoMa spreads south. It’s a fitting view for a firm bridging traditional finance and bleeding-edge technology—and for Huang, whose career has been defined by spotting revolutionary potential.

In 2012, while on a week-long vacation to Beijing, he visited a startup operating from two apartment units. The Founder, Zhang Yiming, was building a personalized news app—an idea Huang thought was destined to fail. But sitting at an old Ikea table near a dusty refrigerator, watching Yiming speak through a translator, Huang noticed something that transcended language: “I remember having this very deep sense that this is an extremely competent, obsessed, aggressive person who was somehow balanced and wouldn’t blow himself up. He had extreme clarity on what he was trying to build and an aggressive take-over-the-world ambition.” Yiming was the most impressive person he’d ever spent time with—so impressive that Huang had to invest. That company became ByteDance, creator of TikTok, and Huang’s share of it has turned into a nine-or-ten-figure sum—he doesn’t keep spreadsheets, so he can’t tell you exactly. 

That instinct for spotting talent is central to Paradigm, which began in 2018 when Coinbase Co-founder, Fred Ehrsam, approached Huang at Sequoia Capital with a vision for a different kind of investment company. They started as equal partners, but Ehrsam has since stepped back to split his time between crypto and his new brain-computer interface startup, recognizing that Huang was, in his words, “born and bred to run Paradigm.”

The son of one of the world’s most accomplished financial theorists and a pioneering computer science professor, Huang grew up at the nexus of math, economics, and technology. In six years, his firm has grown from managing $400 million to over $12 billion by making early, concentrated bets on foundational crypto projects, while also building significant parts of crypto’s core infrastructure. Paradigm’s researchers—who double as investors—develop foundational innovations, then open-source them for the entire industry to use. It’s an unusual approach for a financial firm, but Paradigm isn’t a typical investment company. It’s more like a research lab crossed with an engineering outfit, wrapped in the sophistication of a West Coast Wall Street.

He’s super smart and incredibly humble. It’s tough not to meet Matt and come away with the impression that he is super special.

–Doug Leone, Sequoia Capital

Back in the penthouse conference room, Robinson and White are working on ‘bullseye liquidity’, a breakthrough that could transform how stablecoins—digital tokens pegged to the US dollar—are traded. Stablecoins have become a crucial part of crypto’s financial plumbing, yet the infrastructure for trading them remains primitive, with each pair requiring its own pool of capital. Their innovation would unite these fragmented markets in a single efficient system. Though it could give Paradigm’s portfolio companies, like Uniswap and Noble, a significant edge, they plan to publish the research openly on their blog. “If someone else implemented this and it made things better for crypto overall, we’d be fine with that,” Robinson says.

White pauses his work with OpenAI’s o1 Pro, where he’s verifying mathematical proofs, to refine a point about n-dimensional space. On the screen, Robinson projects a visualization of the math that looks like a quarter slice of Captain America’s shield. Huang mostly listens—he always prefers to listen—but when he does speak, it’s clear he’s fully absorbed the complexity of what they’re presenting.

Years ago, when they were kids, Robinson recalls, their friend group would debate endlessly until Matt spoke. “He didn’t say much,” Robinson says, “but we always ended up doing what he suggested.” Those who know Huang best describe someone whose quiet demeanor masks extraordinary capabilities. “The insight per minute with Matt is very high, even if there are many minutes where he says nothing,” observes Stripe Co-founder Patrick Collison, who added Huang to his board in 2021. His attention to detail extends to everything he touches—from the speed of Paradigm’s website to the obscure Japanese streetwear he favors, to the people he hires. “He has a high bar for excellence,” says Coinbase CEO Brian Armstrong. “There’s no tolerance for mediocrity.”

Yet beneath this intensity lies a disarming humility. As Leone puts it: “He has a good sense of humor but it’s hidden because there’s so much goodness in the man that you can’t help but put your serious game on.” Perhaps most telling is that these insights come from others—Huang is the type of person whose greatest achievements are whispered about rather than trumpeted, whose impact is felt more than advertised. Collison adds: “Not every great investor or great leader is a great person. In all the integrity checks of ‘could this person be a godfather to your kid,’ he passes emphatically with flying colors.”

Paradigm isn’t a typical investment company. It’s more like a research lab crossed with an engineering outfit, wrapped in the sophistication of a West Coast Wall Street. 

This combination of technical excellence and quiet integrity has helped establish Paradigm as one of crypto’s most important institutions. In an industry that’s grown from zero to $3 trillion through waves of speculation and collapse, the firm’s open-source tools now power 90% of smart contract development. Its innovations help hundreds of billions in digital dollars move more efficiently. And its investments have earned the trust of the world’s most sophisticated investors, including Harvard, Stanford, Sequoia and Yale.

CAROLYN FONG

Waking up

One of Matt Huang’s earliest memories places him alone on the streets of Tokyo, a nine-year-old navigating the world’s largest city. Each morning, he would thread his way through narrow side streets and across bustling thoroughfares on an hour-long round trip to school. This early independence shaped his worldview. “Once you have an N equals two,” he said of comparing Tokyo to New York, “that changes how you relate to everything.”

The family had moved to Japan in 1997 when his father, Chi-fu Huang, was tasked with establishing the Asia office of Long-Term Capital Management (LTCM). Before the move, the elder Huang had run LTCM’s Asian trading operation from Greenwich, working 4pm to 3am to match market hours. The only son among four sisters in Taiwan, Chi-fu’s parents had invested their modest savings to send him to America alone. From there, he worked his way to an economics professorship at MIT, then to Goldman Sachs where he built and led Fixed Income Derivatives Research, and finally to LTCM—a firm staffed with Nobel laureates that represented the perfect fusion of academic theory and market practice.

Huang’s mother, Marina Chen, had carved her own path in academia after immigrating from Taiwan. At Caltech, she pioneered parallel computing research under legendary technologist Carver Mead, developing techniques still used in modern processors. Despite her position as one of Yale’s first female computer science professors, seemingly destined for a distinguished academic career, Chen chose to leave academia to raise their three boys, channeling her intellectual intensity into their education.

Dinner at the Huang household ran like an investment committee. Each evening, the boys would hear the garage door opening and scramble to complete their father’s assigned readings—carefully selected, age-adjusted articles ranging from economic principles to Scientific American. At dinner, they were expected to field questions about their allocated topics. Each brother developed different coping mechanisms to their parents’ intensity. According to Matt, the eldest, his instinct was to push back.

Their Tokyo chapter ended abruptly in 1998 when LTCM’s models failed during the Russian financial crisis. The collapse wiped out the Huang family’s savings. Yet Chi-fu Huang emerged from the wreckage to found Platinum Grove Asset Management in 1999 with his LTCM colleague, Nobel Prize-winner Myron Scholes. The firm grew from $45 million to $6 billion in less than nine years, becoming one of the world’s largest fixed-income hedge funds leading into the 2008 financial crisis. This pattern—seeing opportunity in chaos, building something new from system-wide collapse—would become familiar to his son.

The New York suburb of Scarsdale became Huang’s fourth home in 11 years after Tokyo. As one of only three Asian students in his predominantly Jewish school, these constant relocations and cultural adaptations honed his ability to read social dynamics and connect with diverse personalities.

In class, Huang couldn’t sit still. His restlessness led to expulsion from Chinese weekend school after repeatedly disrupting other students. “Uncontrollable,” his parents would later describe him at his wedding. Yet when engaged on his own terms, he displayed an ability to focus. With his “uncool, but academically inclined” circle of friends, Huang directed amateur films, debated libertarian philosophy, and mastered games. His approach to StarCraft—competing semi-professionally on international servers—foreshadowed the attention to detail that would become his signature, extending even to his current obsession with handstands.

Yiming spoke through a translator, but Huang found himself captivated by the founder’s nonverbal cues— his gestures, expressions, and intensity painting a picture he didn’t need words to understand. 

Everything changed when he discovered mathematics. The math club revealed a natural affinity for the subject, and while he wasn’t a top competitor in national contests, it showed his parents their uncontrollable son could excel at school when appropriately challenged. Poker and chess became other outlets for his analytical mind.

The reformed student worked his way to MIT where, in 2006, he found himself among “one of the highest concentrations of strange people on the planet.” He studied mathematics and, at one point, took a semester off to play online poker, working eight tables simultaneously. But the defining moment (besides meeting his future wife) came when a close friend, Albert Ni, announced he was dropping out to join a small startup called Dropbox, as its sixth employee. For someone raised to become a PhD, dropping out of a BSc seemed unthinkable. Yet Ni wasn’t failing—he was one of the most capable people Huang knew, making a deliberate choice to build something new. It led him to read all of Paul Graham’s essays, through which Huang discovered Silicon Valley and the temptation of ultimate rebellion: forging his own path.

With his roommates, Huang applied to Y Combinator in his final year of MIT, initially facing rejection. Graham told them: “We love you guys. We couldn’t hate your idea more.” Six months later, with a working prototype, they got in. The MIT graduates drove cross-country to San Francisco in six days. At YC, they built what Huang now calls “a terrible idea”—a TV guide website for the streaming era called Hotspots. This “failed startup” chapter that lasted two years nonetheless gave him deep empathy for founders and led to an acquisition by Twitter, where he observed a “terribly run” company during its pre-IPO period.

By 2012, Huang was ready to carve a new path. In his mind, Silicon Valley had become too obvious; the consumer space too predictable for interesting work and compelling returns. During a week-long vacation from Twitter, with thoughts of founding a tech company in China, he visited Beijing to meet six founders. One was Zhang Yiming, who was building what seemed like another doomed consumer app. Yiming spoke through a translator, but Huang found himself captivated by the founder’s nonverbal cues—his gestures, expressions, and intensity painting a picture he didn’t need words to understand. Huang left the apartment thinking, “I’ve got to figure out some way to back this person.”

He wrote a check into ByteDance at $20 million and $30 million valuations, marking his largest personal investment at the time. With ByteDance now worth $300 billion, his investment has multiplied roughly 10,000x—turning an illustrative $50,000 into $500 million. He still holds most of his position, and while he’s “increasingly zen about it,” he admits “it definitely messes with your head that it’s probably the best investment I’ll ever make.” That same year, in San Francisco, he made seed investments in YC companies Instacart, Benchling, PlanGrid, and Amplitude—all now billion-dollar companies.

In 2014, when a Sequoia recruiter’s email appeared in his inbox while at Twitter, Huang initially dismissed it as spam. Despite his impressive track record, he had no intention of becoming an investor. But curiosity prevailed, and his interview assignment—a one-pager on a company Sequoia should invest in—led him to write about Coinbase, then just seven employees.

At Sequoia, Huang found what he calls “the highest-standards place I’ve ever experienced.” When Facebook acquired WhatsApp for $19 billion on his second day, partners gathered briefly in the lobby. Champagne was poured but left untouched. Within five minutes, everyone had returned to work. Eleven-figure exits faded into the shadows of the firm’s legendary portfolio that includes trillion-dollar businesses like Apple, Google, and Nvidia. The culture demanded excellence in ways that elevated his already considerable ambition.

“You start to see how far the axes extend, what a great founder looks like,” Huang says of his four years there. “Without being exposed to that, your whole perception of what’s possible is missing the dynamic range of the top end.” Sequoia also showed him that excellence takes many forms. Working alongside investors with widely different styles but consistency in key dimensions gave him confidence to develop his own approach: “The realization that I can do it my own way was very liberating.”

Sequoia, for its part, got something valuable in return. “Sequoia US was getting its ass kicked by Sequoia China in their poker tournament every year,” Leone remembers. “And he went and he won it for Sequoia US. We brought back the Don Valentine crazy-colored jacket thanks to Matt Huang.” You won’t hear this story from Huang, who never boasts about his achievements. Like many things about him, you have to learn it from others, or know exactly what to ask.

Leaving home

Huang first encountered Bitcoin in 2010 while at MIT, immediately drawn to its elegant fusion of mathematics, economics, computer science, and game theory. 

“I internalized it as just a really beautiful idea,” Huang recalls. But in those early days, it seemed more of an intellectual curiosity than an investment opportunity. It wasn’t until 2012 that he bought some Bitcoin on Mt. Gox, the dominant exchange at the time, and rode his first major bubble. “You almost need to lose money the first time,” he reflects. “Then you kind of give up on it, write it off as dead. And when you see it come back and it’s not dead, then you start to wonder.”

Multiple sources reported that legendary investor Michael Moritz called Huang “the only regrettable loss in Sequoia’s history.” Leone said, “he was the first person in my career that had left Sequoia under his own volition.” 

At Sequoia, Huang found few colleagues who shared his growing conviction about crypto’s importance. The firm was supportive of his interest—he led several crypto investments on its behalf—but he increasingly sought conversation partners outside its walls. He began attending monthly dinners in San Francisco with six to eight other crypto-curious investors, exploring ideas at the frontier of the emerging technology.

It was during this period in 2017 that Fred Ehrsam, who had recently stepped down as President of Coinbase, wrote a blog post arguing that crypto was the metaverse. Huang, still at Sequoia, reached out to discuss the idea. “I know I’m not going to build a company around this,” Ehrsam thought, “but it would be hilarious to pitch Sequoia on the idea for fun.”

What began as an intellectual exercise evolved into a 40-email thread between the two, diving deep into crypto’s possibilities. Their backgrounds were perfectly complementary: Ehrsam had co-founded and operated crypto’s most important company, while Huang brought elite investing experience.

“Nothing felt quite right until I met Matt,” says Ehrsam, who had explored starting a crypto-focused fund with several other potential partners. Over six months, they methodically explored working together, testing alignment on everything from investment philosophy to fund structure. They were particularly focused on ensuring true partnership—everything would be split 50/50, a principle that “drove some people nuts” but felt essential to both of them.

Leaving Sequoia was wrenching for Huang. It was the first place where he truly felt he belonged: “It felt like somewhere I could retire from, if they would have me.” Multiple sources reported that legendary investor Michael Moritz called Huang “the only regrettable loss in Sequoia’s history.” Leone said, “he was the first person in my career that had left Sequoia under his own volition.” But Huang had become convinced that crypto would be one of the most important technological trends of the coming decades. “When he told me he thought it was the opportunity of a lifetime, it was very easy. Follow your dreams, go after it,” Leone said, with a tinge of regret. “I am very mad at myself because he had been talking about Bitcoin and I usually have a good sniffer. If I were really smart, I would have followed that and created an opportunity for him to do the fund within Sequoia.”

CAROLYN FONG

Inverting the status quo: Huang and Ehrsam

Huang and Ehrsam founded Paradigm in June 2018 with two theses: first, that crypto would be one of the most important technical and economic shifts of the coming decades; and second, that the space lacked the kind of investor they themselves would have wanted as entrepreneurs—one that was deeply ‘crypto-native’.

Graham Duncan, Founder of East Rock Capital and an advisor to Paradigm, whom Huang credits as the most helpful person to the firm in its early days, was struck by their conviction from the outset. “They were consistently planning for what could happen from a size perspective in a way that seemed almost absurd to me,” Duncan said. “It blew my mind, but it was not cockiness at all. Their time horizon was just so different, and what they planned for ended up happening.”

They raised their first fund in late 2018, securing $400 million from elite institutions including Harvard, Stanford, and Yale—the universities’ first significant crypto investments—as well as Sequoia. The fund structure was novel: open-ended with no fixed timeline for returning capital, allowing them to hold both public crypto assets and private investments. Then they made an even bolder move: rather than calling capital gradually like most venture firms, they quickly asked for the full $400 million and began averaging into Bitcoin and Ethereum, with those positions comprising about 90% of the fund to begin with. The average purchase price for Bitcoin was around $4,000 to $5,000 a coin—a massive bet that the crypto winter, which had sent Bitcoin down over 70% in 2018, would eventually thaw.

Their first three hires embodied different aspects of their vision. Charlie Noyes, whom Huang met in a Telegram chat discussing Bitcoin Cash forks, became their first employee. “From his messages, I thought he was a 40-year-old with a beard, very cynical and craggy,” Huang remembers. “When he showed up to dinner, I was really surprised he was 19.”

Noyes had been immersed in crypto from age 12 when he discovered Bitcoin through gaming circles. He’d already published research papers on crypto applications and won Intel’s science competition twice, before dropping out of school to attend MIT, and then dropping out of there to join Paradigm. His initial adjustment to office life was rocky—he thought “opining on pitch decks over email and coming to the office once a week” was normal. When he showed up late on his first day, Huang sat him down to explain professional expectations. The patience paid off. 

Today, Noyes is 25 and already one of Paradigm’s general partners. Huang likens him to an artist, capable of making large intuitive leaps by condensing vast amounts of disparate information and coming out the other side with a crystal-clear investment thesis. Like when he identified MEV as a pivotal blockchain issue in 2020 and became the lead investor in Flashbots, whose infrastructure now touches nearly every transaction on Ethereum and has established critical market rules for the $450 billion ecosystem.

Dan Robinson, Huang’s middle school friend and “the smartest person I knew growing up,” embodied the technical depth needed to advance crypto’s frontier. After Harvard Law School left him disillusioned, Robinson switched to programming and explored crypto while working at Stellar, a blockchain company. Huang and Ehrsam crafted a unique role split between investing, research, and helping portfolio companies build. What began as a compromise became the template for Paradigm’s research-driven approach, with Robinson going on to invent key mechanisms for Uniswap, crypto’s leading decentralized exchange.

She’s been Matt and Fred’s third partner, totally building the firm.

–Graham Duncan, East Rock Capital

CAROLYN FONG

Paradigm’s Managing Partners: Huang and Palmedo

Alana Palmedo, joining four weeks into Paradigm’s existence when it still rented office space by the week, brought the institutional rigor needed to bridge crypto and traditional finance. Though not “deeply into crypto,” her experience managing complex operations at Boston University’s endowment and Bill Gates’ investment office during the 2008 financial crisis proved invaluable. Initially skeptical, she was won over by Huang and Ehrsam’s thoughtful approach to building an institutional-grade firm and her value investor’s intuition that with Bitcoin down so far, “this must be the low.”

“She’s been Matt and Fred’s third partner, totally building the firm,” says Duncan, who helped interview Palmedo. She initially managed everything from trade settlements to finance to compliance before recruiting specialists for each function, freeing the investment team to focus on deals. Now, as managing partner, she has architected Paradigm’s high-performance culture, where radical transparency and daily self-reflection are expected from everyone, regardless of role. “Every single person must be in the top 1% of their domain,” Palmedo insists.

By mid-2019, with crypto prices beginning to recover but most investors still wary of the sector, Paradigm went back to market. Its initial investor base committed an additional $360 million. The timing was characteristic of Huang’s approach: raising capital when others were skeptical, from partners who shared their conviction that crypto would fundamentally reshape finance.

Though crypto has yet to fulfill its transformative promise, Paradigm’s investments have already yielded exceptional returns. According to public filings, its first flagship fund grew from $760 million to $8.3 billion by the end of 2024. Speaking with sources close to the company, Paradigm has also returned all of its Limited Partners’ initial capital, paying out more than $1 billion from the fund to date. 

The long view

In spite of Paradigm’s early success, it’s hard not to wonder why Huang, someone for whom money ceased to be a concern long ago and who seemingly had the perfect job at Sequoia, would bother with the wild world of crypto at all.

Brian Armstrong, CEO and Co-founder of Coinbase, pondered a similar question—”Who leaves a job like that at Sequoia, right?”—before answering, “He’s a silent killer. Our industry needs more people like him who have high integrity and are in it for the long term and the right reasons. He had such conviction to go off the beaten path.”

For Huang, the answer is simple: “I suppose I’ve always had somewhat of a skepticism towards authority, and so when I see authority exerting itself, it does make me wonder: is this the way we want the world to work?”

“Everyone in the US looks at China and says, that looks dystopian,” he says. “I don’t think they fully realize the same thing is happening in the West.”

Take the Obama administration’s Operation Choke Point, in which the US Department of Justice sought to restrict certain industries’ access to banking services. Operation Choke Point 1.0, which lasted from 2013–2017, targeted industries deemed ‘high-risk’ like payday lenders and firearms dealers. Under the Biden administration, Operation Choke Point 2.0 focused on ‘debanking’ crypto. Even individuals like Uniswap Founder Hayden Adams and Gemini Co-founder Tyler Winklevoss found their personal bank accounts suddenly closed without explanation.

Huang sees crypto evolving in three critical stages: first as money, then as a financial system, and finally as an internet platform. Each phase builds upon and enables the next. “Money is upstream of a lot of the rest of crypto,” he explains. “Buying your first Bitcoin or setting up your first wallet is often the first step towards using other crypto applications. It’s like getting your AOL account and connecting to the internet for the first time.”

The money stage has already produced staggering results. Bitcoin has grown from a whitepaper in 2008 to an asset worth nearly $2 trillion today, making it the most valuable startup created since it was published. Remarkably, nation-states—even the US—are beginning to adopt it. 

Institutions that scoffed at the industry in 2018 (like BlackRock’s CEO Larry Fink, who called Bitcoin “an index of money laundering”) now embrace the technology. In 2024, BlackRock’s Bitcoin ETF gathered $50 billion in just 11 months—the fastest-growing ETF launch in history. Traditional portfolio models are also shifting, with Fidelity now recommending 1–3% crypto exposure. The classic 60/40 portfolio is becoming ‘59/39/2’ as institutions carve out dedicated allocations to crypto assets.

The second stage—building an entirely new financial system—is accelerating. While traditional finance operates through layers of intermediaries, crypto can enable near-instant transactions, 24/7 markets, and novel financial instruments, all built on permissionless rails. The rise of stablecoins—blockchain-based digital currencies pegged to stable assets like US dollars—demonstrates this potential, growing from $500 million to over $200 billion in circulation since Paradigm’s founding.

The third stage—as an internet platform—remains the most nascent and least legible. Unlike today’s internet where platforms control user data and online assets, crypto could enable true digital ownership and direct user-to-user interactions without intermediaries. High transaction costs have held back everyday applications like social media and gaming, but Huang sees this changing as new scaling technologies drive costs down dramatically. The same infrastructure that today supports NFTs and meme coins will eventually enable more serious applications, much like YouTube evolved from cat videos to become one of the world’s most important platforms.

CAROLYN FONG

Huang, Ehrsam, and Palmedo outside the penthouse floor of their office in San Francisco

Of course, like every new technology, crypto also has a dark side. Scams and hacks are common, meme coins promote short-term thinking over building what’s truly needed, token prices are volatile, projects implode, and the whole thing often looks more like a casino than the future of finance.

Yet Huang takes the long view. Like the early internet attracting brilliant researchers alongside scammers and fraudsters, crypto’s open frontier enables both innovation and bad behavior. Each new wave, including speculative bubbles that can look irrational from the outside, brings in fresh talent and capital to build critical infrastructure.

Stablecoins are a perfect example. The 2017 ICO (Initial Coin Offering) bubble brought mainstream attention to crypto and minted a new generation of crypto-rich investors. Some of that capital flowed into developing stablecoins, leading to dramatic improvements in their infrastructure. On Ethereum, the cost to send USDC (a popular dollar-pegged stablecoin) has fallen from $12 in 2021 to $1 today. On Coinbase’s popular Layer 2 network Base, the same transaction costs less than a cent. In turn, circulation has grown exponentially, up 400-fold since the bubble popped, and real-world use cases have emerged.

SpaceX uses stablecoins to repatriate revenue from emerging markets, converting local currencies to digital dollars that transfer instantly. Scale AI pays its global network of data labelers through stablecoin rails, eliminating cross-border friction and costs. Corporate treasury teams at companies like Ramp have found another advantage: while savings accounts pay a fraction of a percent interest, stablecoins backed by Treasury bills can capture most of the yield that banks typically keep for themselves.

The numbers confirm the narrative. Transaction volumes have grown 120% annually for five straight years. In 2024, stablecoins processed $5.6 trillion in payments, nearly half of Visa’s $13.2 trillion. This momentum led Stripe to acquire Bridge, a stablecoin payments platform, in October 2024. “Stablecoins are room-temperature superconductors for financial services,” wrote Stripe Co-founder Patrick Collison.”Thanks to stablecoins, businesses around the world will benefit from significant speed, coverage, and cost improvements in the coming years.”

Matt’s disposition is distinctive. He’s calm, rigorous, and patient—uniquely well-suited traits to any complicated technology whose consequence is back-weighted, like crypto.” 

–Patrick Collison, Stripe

This adoption mirrors crypto’s broader evolution: Bitcoin launched in 2009, reaching its first million users in 2011. Ethereum followed in 2015, hitting the same milestone in 2017. Then came stablecoins in 2019, DeFi in 2021, NFTs in 2022, and social applications in 2023. 

Detractors often highlight the lack of impact crypto has had on day-to-day business. Huang believes stablecoins are the next killer app but also distinguishes between ‘single-player’ technologies like AI that provide immediate utility, and ‘multiplayer’ technologies like crypto that require coordinated adoption. “It’s like adopting a new language or settling a new city,” he explains. “It’s useless if you do it alone.” He points to email as an instructive parallel. Early critics called it “technologically intriguing but economically naive,” much like crypto skepticism today.

When speaking with Huang, it’s notable how poised he is on crypto as a whole. Patrick Collison, who added Huang to Stripe’s board both for his crypto expertise and his broader business acumen, said, “Matt’s disposition is distinctive. He’s calm, rigorous, and patient—uniquely well-suited traits to any complicated technology whose consequence is back-weighted, like crypto.”

What sets him apart is his ability to hold both sides of the investment thesis. “He can handle bear cases, which are typically much more specific than the bull case,” says Collison. “And then he understands the possibilities of technology to see how small nascent things can become a very big deal in the future.”

Recently, artificial intelligence has emerged as tech’s new frontier, with clear and immediate applications that have captured the world’s imagination. Huang and his team at Paradigm even considered expanding their focus to include AI. But they have maintained their commitment to cryptocurrency, with Huang explaining: “AI is going to be fine with or without us. Crypto is a very important technology that needs to coexist with AI, but there aren’t a lot of great champions. We believe it’s important for us to work on it, to make sure that it actually succeeds.”

Invention

That commitment to ensuring crypto’s success has led Paradigm to develop an unusual approach to investing. While most venture firms wait to back winners, Paradigm helps create the conditions that make winning possible. This means doing more than analyzing trends or writing checks—it’s solving fundamental technical problems that expand what the entire industry can achieve.

The firm’s research-driven style emerged almost by accident. When Huang hired his middle school friend and best man, Dan Robinson, it wasn’t obvious how a former lawyer turned self-taught programmer would fit into an investment firm. “We wanted Dan on the team because he was the smartest person I knew,” Huang says, “but he wasn’t the most commercial, and we weren’t sure how he’d input into the investment process.” Mostly as a concession to bring him on board, they created a novel role that included time for Robinson to work on open-source projects, which they mutually termed “exploratory research.”

“It turns out that particular genre is an incredibly important thing to do in crypto,” Robinson explains. “Most investment research is about gathering and analyzing existing information. We’re trying to invent new things.” The research team’s breakthroughs often come from exploring theoretical questions before companies even recognize they need answers, like bullseye liquidity.

What makes crypto somewhat unique is how mathematical mechanisms can create enormous leverage. A traditional exchange might need thousands of servers and hundreds of employees to match buyers and sellers. But when Vitalik Buterin proposed a simple equation (x*y=k) on Reddit in 2016, he created the foundation for trillion-dollar markets to run autonomously on a blockchain. The challenge was that this elegant solution, while computationally efficient, wasted massive amounts of capital by spreading liquidity across all possible prices.

Robinson knew Hayden Adams, who had built Buterin’s concept into Uniswap, from the early Ethereum research community. Within weeks of joining Paradigm, he wrote a memo on Uniswap, which led to a seed investment, and began actively working on improvements. His contributions to Uniswap v2 enabled trading between any Ethereum-based tokens, helping the protocol expand from $2 billion in volume to over $1 trillion.

But Robinson and Adams spent most of 2019 searching for an even more fundamental breakthrough. Through mathematical exploration, the team discovered a way to efficiently concentrate liquidity within specific price ranges—allowing traders to focus capital where it was actually needed. This innovation became Uniswap v3, improving capital efficiency by up to 4,000x. A $5 million position could now provide the same trading depth as $2 billion spread across all possible prices. By October 2022, Uniswap was valued at $1.7 billion.

When they go up against other firms, Paradigm can actually help you build a crypto company. They have experts on staff in protocol design, security, legal, even policy.” 

–Brian Armstrong, Coinbase

This pattern of research leading to breakthrough products repeats across Paradigm’s portfolio. Last year, when Blur approached the firm about adding margin trading, the team faced a fundamental challenge: how do you safely lend against illiquid NFTs whose values are hard to determine? The research team spent four months developing an entirely new lending protocol called Blend. “If you can solve lending against NFTs,” Robinson notes, “you can potentially solve lending against any illiquid asset.” Within months of launch, Blend had created and dominated an entirely new category of lending.

Unlike traditional venture firms that separate technical resources from investment decisions, Paradigm’s researchers are core to every investment. They attend every pitch meeting and help make every decision. This integration means they often identify opportunities others miss because they’re already working on similar technical challenges. When algorithmic stablecoins like Terra became popular, Paradigm stayed away—years of trying to design better stablecoin mechanisms had taught the firm that these projects hadn’t solved the fundamental problems.

This deep technical work creates a powerful competitive advantage in sourcing and closing deals, as well as hiring talent. “When they go up against other firms,” Coinbase’s Armstrong explains, “Paradigm can actually help you build a crypto company. They have experts on staff in protocol design, security, legal, even policy.”

“The biggest part of our process is figuring out what’s actually the most important problem,” Robinson explains. This requires staying close to crypto’s rapidly evolving frontier. “The internet has these very short generations,” Huang notes, drawing a parallel to Sherlock Holmes’s network of street urchins who provided vital intelligence from London’s streets. “Even two years can make a difference in understanding crypto’s culture.”

This insight led to the Paradigm Fellowship, which identifies exceptional young developers while they’re still in school. The program grew partly from the firm’s experience with transmissions11, who the team discovered on Discord while he was still in high school. When he once dialed into a pitch meeting from a school assembly, it crystallized both the challenge and opportunity of working with crypto’s next generation of innovators.

CAROLYN FONG

Sketching their next breakthrough: Konstantopoulos, Noyes, and Robinson

Crypto crypto crypto

On the last Thursday of May 2023, reporters at crypto news firm The Block made a discovery that sparked industry-wide controversy. Using the Internet Archive’s Wayback Machine, they noticed Paradigm had quietly removed all mentions of ‘crypto’ from its homepage and social media, rebranding as a ‘research-driven technology investment firm’. The apparent revelation—despite the fact that the change was a month old—prompted immediate backlash. In an industry where loyalty runs deep and apostasy is harshly punished, it felt like betrayal.

“We don’t want to work for you anymore,” tweeted one portfolio company, referencing both the rebranding and Paradigm’s investment in FTX, which “became a scar on our entire industry”. The criticism stung but was characteristic of crypto’s brutal honesty. Paradigm not only restored the word ‘crypto’ but doubled down, adding flashing neon tickers to the homepage that read: ‘crypto crypto crypto’.

The reality behind the change was more mundane. Two researchers on the team had complained that potential AI collaborators weren’t returning their emails after seeing ‘crypto’ on Paradigm’s homepage. “We thought, okay, all the crypto people already know us, they never go to our homepage. They go to our blog, and every portfolio company and blog post is about crypto. So what’s the big deal?” Huang explains. “In hindsight, it was clearly a mistake. People take the website as a collective statement about what you’re proud of.” (Patrick Collison, for one, notes that the Paradigm website is “probably the fastest you have used this year.”)

The episode revealed deeper tensions, however. By November 2022, Bitcoin had plunged 75% from its peak in 2021 to below $16,000. Ethereum was down 80%. That same month, ChatGPT launched, sparking an AI boom that made crypto seem like yesterday’s frontier. Major venture firms were already pivoting their attention and capital toward AI.

For Paradigm, the website controversy capped a humbling period. Just 18 months earlier, the firm seemed untouchable. Its Bitcoin position had multiplied 15x. One of its earliest investments, Coinbase, had gone public at an $85 billion valuation. It had raised a $2.5 billion venture fund. Yet the euphoria of 2021 would test even crypto’s most disciplined investors.

Fred Ehrsam recognized the warning signs. In March 2021, he sent a letter to portfolio companies titled ‘Surviving Crypto Cycles’. After noting that prices had doubled in just two months, with Bitcoin crossing $1 trillion and “pixelated crypto art regularly selling for millions,” he warned that “Senators even have lasers for eyes! Euphoria abounds.” Drawing on his Coinbase experience, where a third of employees left during the 2014–2017 downturn, he outlined specific preparations: stress test systems for 10–100x usage spikes, consider fundraising while capital is available, and caution new hires about crypto’s brutal cycles.

Down years are easier than up years. Signal to noise is very high and okay, prices are down, but with a long-term perspective, that doesn’t bother us.” 

–Matt Huang, Paradigm

His warning proved prophetic but insufficient. “We made a lot of mistakes during that period,” Huang reflects. “When you over-focus on a rival, you become more like the rival.” He explains how watching competitor a16z raise massive funds made them question whether they needed to match that scale. Paradigm grew from 18 to 62 people. “We definitely let the quality bar slip,” he admits. “I remember instances of making that compromise, feeling like if we don’t do this, or hire that person, we’re going to fall behind. In hindsight, those were all wrong decisions.”

Huang is not one for spreadsheets and doesn’t remember the firm’s largest peak-to-trough drawdown, but one moment is etched into his memory: FTX. Paradigm had invested $278 million in the exchange, making it one of the largest investments in the firm’s history. By 2022, FTX had become crypto’s public face, with founder Sam Bankman-Fried speaking at conferences, testifying before Congress, and appearing on magazine covers. That October, he gave the keynote at Paradigm’s LP meeting. Weeks later, FTX collapsed amid charges of fraud and revelations that customer funds had been misappropriated.

The investment failure was total, but the betrayal cut deeper. During due diligence, Paradigm had identified the key risk: the relationship between FTX and Bankman-Fried’s trading firm, Alameda Research. The team asked direct questions and received false assurances. When Huang later testified in court during Bankman-Fried’s criminal trial, the experience solidified an important lesson about founder alignment.

“It was pretty clear, even at the time, that he didn’t share our vision for making crypto better,” Huang says. “For him, it was a way to make a ton of money and then give it away.” The disconnect became apparent around policy discussions, where Bankman-Fried advocated for compromises that Paradigm believed would damage crypto’s core promise.

FTX wasn’t Paradigm’s only misstep. The firm had co-led OpenSea’s $300 million Series C at a $13.3 billion valuation at the peak of the NFT craze. Since then, the NFT marketplace’s trading volume has dropped 98%. BlockFi, another portfolio company, went bankrupt due to FTX exposure.

“In venture capital, there are going to be investments that don’t turn out the way you hope,” Huang says. “That’s always an opportunity to reflect, and we’ve done a lot of that.” He maintains that bearish periods actually provide clearer signals than bull markets. “Down years are easier than up years. Signal to noise is very high and okay, prices are down, but with a long-term perspective, that doesn’t bother us.”

The firm emerged from this period smaller but more focused. The investing and research team, which had grown to 20 people in 2021, was cut back to 11. Standards tightened and an explicit filter was added to new investments: the founder must align with Paradigm’s mission to advance the frontier of crypto.

The website episode proved instructive in another way. The swift negative reaction showed how much the crypto community had come to view Paradigm as more than just another investment company. It was a standard bearer for the industry.

Writing the future

A miniature electric guitar rests behind Georgios Konstantopoulos’s desk, occasionally deployed during impromptu meetings. The image of the firm’s Chief Technology Officer breaking into riffs when discussing blockchain architectures captures something essential about his approach: technical virtuosity combined with an intuitive feel for what works.

In 2019, Konstantopoulos was a sought-after researcher and software engineer, known in crypto circles for his development skills. His technical work was so thorough that Paradigm’s portfolio companies kept mentioning his name.

When Huang first met him at a conference, Konstantopoulos was weighing whether to expand his consulting practice or join a startup. But Huang, with his characteristic ability to spot unconventional talent, saw a different path. He proposed creating a role similar to Robinson’s, where Konstantopoulos could combine technical research with investment evaluation.

The role evolved in unexpected ways. In 2020, while helping portfolio company Optimism implement its research, he noticed how many projects struggled with the same fundamental problems. The challenges weren’t in the ideas, but in the tools needed to build them. Rather than support companies one at a time, Konstantopoulos wondered if he could build open-source infrastructure that would advance the entire industry.

Things people tell you are hard usually aren’t. They’re only hard because you don’t have agency over your tools.

–Georgios Konstantopoulos, Paradigm

That philosophy led to Foundry, his first major contribution. Konstantopoulos spent a weekend building a tool that made writing secure, smart contracts dramatically simpler. Think of it like spell-check for blockchain code—except instead of catching typos, it prevents multimillion-dollar bugs. Within months, it had become the industry standard, and now has 90% market penetration and over $100 billion in smart contracts secured (so far).

But Foundry’s success highlighted a deeper challenge. Ethereum, the platform powering most of crypto’s innovation, ran on inefficient software that made scaling impossible. It was like trying to stream 4K video over a dial-up connection. Konstantopoulos proposed an audacious solution: rebuilding Ethereum’s core node software from scratch.

“You’re crazy,” his team said. “This will take five years.” But Konstantopoulos had earned their trust, and he understood their capabilities better than anyone through his unique hiring approach. Rather than traditional interviews, he found his engineers through their contributions to open-source projects. “The code doesn’t lie,” he says. “I want to see how people think through real problems.”

The resulting project, Reth, took just 18 months to complete. While its function seems simple—downloading transactions, executing them locally, and writing them to a database—its impact has been profound. By optimizing this fundamental process, Reth runs 80% smaller and 10 times faster than alternatives. Major platforms like Coinbase’s Base, WorldCoin, and Optimism (last valued at $1.65 billion in 2022) already rely on its superior performance (it launched in June 2024).

These technical contributions create a virtuous cycle. Paradigm’s researchers identify problems while evaluating investments. They build open-source solutions that become industry standards. These tools attract the best developers, who either join portfolio companies or become founders themselves or, in some instances, join Paradigm.

The strategy culminated last October in Ithaca, spun out from Paradigm with a $20 million investment. As CEO (while maintaining his Paradigm CTO role), Konstantopoulos aims to commercialize what his team has built. “What other teams need 20–30 engineers, seed funding, and two years for,” he notes, “we can do in weeks.”

His confidence comes from having built every layer of the stack, from low-level cryptography to user interfaces with Reth and Foundry in between. “Things people tell you are hard usually aren’t,” he argues. “They’re only hard because you don’t have agency over your tools.” This philosophy of radical self-reliance—building whatever tools the industry needs—has transformed Paradigm’s role in crypto.

As for his own transformation, Konstantopoulos describes it in characteristically Greek terms: “Matt is the only mentor I haven’t been able to eclipse.” Most mentors are eventually outgrown, but in Huang, Konstantopoulos found a leader who evolves alongside his team. While most engineers of his caliber would have left to start their own companies, he and others stay at Paradigm because Huang keeps growing with them. “They push me to be a better version of myself every day.” Huang says, “I don’t want to be eclipsed.”

On the back of Huang’s MacBook, a telling detail appears: three stickers form a perfectly aligned row, with the logos of Foundry and Reth flanking Paradigm’s mark. It’s both a reflection of his careful attention to detail and a window into his evolving vision for venture capital. “What if Sequoia had not just backed Google, but had founded Google?” he increasingly wonders. The question points to a future where the line between investor and builder blurs.

In Konstantopoulos’s journey from gaming addict to architect of crypto’s core infrastructure is the fulfillment of Huang and Ehrsam’s original thesis: that crypto required a different kind of investor. Not just brilliant misfits who could evaluate technology, but builders whose code could shape the future of finance. In an industry inherently hostile to central authority, Paradigm has become one of crypto’s most trusted institutions by focusing on creation rather than control. Huang and his team aren’t just investing in the future—line by line, they’re writing it.

CAROLYN FONG

Dom Cooke is the managing editor of Colossus

Click here to subscribe to print for your office or home.

Click here to subscribe to print for your office or home.

Picture him in the middle of nowhere, in Guangzhou, which back then was still a torment of factory exhaust and steel. It is 2007, and Neil Mehta is 24. He is tall, about six foot two, but looks tiny beneath the only other thing standing in the vast empty space, a giant apartment building, which is going for about $35 million. Mehta is here to purchase it at auction. 

He is living in Hong Kong, in the heady expat district of Lan Kwai Fong, but he’s been dispatched here to Guangzhou by a subsidiary of D.E. Shaw, the hedge fund, to bid on this big apartment building in the middle of nowhere. He’s coming in with real heft at his back, Goldman and the like, so Mehta shows up to the auction representing the good offices of D.E. Shaw in a smart jacket and slacks, which is uncomfortable because it’s humid and hot, and he had to take a train and then a cab which followed some guy on a motorcycle who claimed to know where this building in the middle of nowhere was but couldn’t find it until the third try.

So imagine Mehta’s surprise when he gets to the auction, which is just a bunch of lawn chairs creaking wanly on the empty ground beneath the building. In front of the chairs is a wooden podium and an auctioneer handing out paddles. Mehta takes a paddle and sits next to his competition for the $35 million building, a farmer with a paunch in a wife beater smoking a cigarette. The tendrils of smoke curl slowly in the humid air, and the lawn chairs creak, and in the weird little Wes Anderson movie his life has become, Neil Mehta looks up at the big building in the middle of nowhere and smiles, probably.

Later that night he is in Shenzhen, to visit the offices of BYD. In 2007 BYD was not yet a $600 billion conglomerate vaporizing the German auto industry. It was a seemingly fraudulent goat rodeo of a company making batteries for frivolous little ‘electric cars’, in a dank warehouse in a city that was still in the process of transforming from a fishing village. When Mehta walks into BYD he notices liquid dripping down from the ceiling onto the floor, which isn’t the kind of thing you prefer to see in a place manufacturing lithium iron phosphate batteries.

The place is just a mess, but it doesn’t really bother Mehta. He loves it, in fact. He’s obsessed with this idea that electric vehicles are going to be a big thing, and he loves the energy of the place. He especially likes talking to the people there about the company’s Founder, Wang Chuanfu. Mehta loves hearing about Wang. It’s been all well and good learning about Mongolian commodities and Australian housing and apartment developments in Guangzhou at D.E. Shaw in Hong Kong, but Mehta realizes he’s got this thing about founders. He really, really likes them.

He likes them a little too much, maybe. At one point he meets the founders of this animation studio called Imagi making a Pixar-style movie about Astro Boy, the postwar Japanese manga, and he loves spending time with the founders so much he raises $3 million to keep Imagi afloat when it’s approaching bankruptcy at about 17 cents a share. The 24-year-old Mehta saves Astro Boy from certain death and when it comes out, the critical consensus is that it’s something like the worst animation movie ever made. But it’s got Nicolas Cage and Kristen Bell doing voices and enough people see it that Imagi’s stock price climbs back up, and in a few months Mehta makes over three times his money on a company staring at bankruptcy.

Which isn’t even really the point. The point is that he loved spending time with the founders and being their closest partner at the lowest point in their lives and seeing it through until their heads were back above water. It’s the kind of thing he starts doing more and more, on nights and weekends, on the margins of his job investing in distressed real estate, buying distressed banks, flying through torrential Hong Kong downpours in a single-engine helicopter to Macau to negotiate financing for luxury hotels, and all the other stuff a young associate in the special situations group at D.E. Shaw has to do. Mehta gets this idea into his nut that what he should really be doing with his life is being the best partner a founder’s ever had.

Now it’s August 2008, the Olympics, and Mehta’s in Beijing to watch Michael Phelps compete in the 4×100-meter medley relay. Mehta is sitting in the stands in the Beijing Water Cube and the lights go down, the meet is about to start.

He was here in this same spot just a week earlier, watching Phelps swim a different meet, and when the lights went down that time everyone just watched the meet. This time, though, the lights go down, and Mehta, clutching his BlackBerry Messenger, notices that the faces of everyone in the stands are illuminated and looking down at their laps, as if they’re boycotting Phelps’ dominance by refusing to look at it.

By now it’s a familiar sight—the lights go down in a movie theater, and everyone’s face is lit up blue by their iPhone—but BlackBerries didn’t do that, and back in the summer of 2008 the iPhone was only a year old, so it was still a weird thing to see. He asks a friend next to him what the hell everybody’s doing, and the guy says they’re all on QQ. Everyone in the stands has some version of an iPhone knockoff HTC smartphone, and they’re all on QQ, the precursor to WeChat. 

Mehta goes back to his apartment in Hong Kong after the meet and looks it up. QQ is going from near zero to racking up more than 30 million users a quarter, the growth curve is pointing to 100% penetration. Insane! Mehta calls Benny Peretz, his friend at D.E. Shaw in New York and his little brother’s old college roommate, whom he first bonded with a few years earlier at the UPenn Spring Fling, which they spent discussing the arcana of European cruise ship financing. “This is it,” Mehta says to Benny about his Water Cube epiphany. “We don’t get to be around for many of these in life. This is the kind of thing we should be doing.”

By this point Bear Stearns had already crashed, and when Mehta went to the brass at D.E. Shaw to explain why they needed to be investing in private technology companies right now, Lehman Brothers collapsed a few days later. Among other things, there’d been too much credit going around, which is why provincial farmers were bidding on $35 million apartment buildings. Every TV at D.E. Shaw was plastered with footage of men with ties slung over their shoulders wandering around Wall Street holding their life in a box. In other words, Mehta’s superiors were not interested in hearing from a 24-year-old associate just then about how these mobile smartphone apps have zero marginal cost, you just write lines of code, you deploy it on a platform for free, there’s a network effect—it’s all so sick!

By 27 Mehta’s seen enough, and he’s left D.E. Shaw to start a firm with his friend Benny. On a whim he calls it Greenoaks, after the street in Atherton, California where he grew up. In the first deck for Greenoaks there’s a slide called ‘Finding Value in Unusual Places’, which at the time meant the internet. The idea was that these internet companies, these unbelievably cheap businesses, were going to replace a large percentage of the S&P 500. In 2012, he bets 40% of his first fund on Coupang, an ecommerce retailer in South Korea, led by a founder, Bom Kim, whom Mehta’s fallen head-over-heels in love with. Greenoaks soon owns upwards of 15% of the company, and eventually returns about $8 billion from that one investment.

Today Mehta is 40, and over its first 13 years, Greenoaks has played a legendary part in the rise of Coupang, Figma, Wiz, Carvana, Stripe, Discord, Rippling, Toast, Robinhood, and other unicorns led by N of 1 founders, generating over $13 billion in gross profits, a 33% total net internal rate of return, and only about a point of principal impairment. The firm is unusually small and concentrated: five funds of 55 core companies across nearly $15 billion of assets managed by nine investment professionals—with Mehta himself as one of the largest LPs in each fund. Henry Kravis, one of the first LPs, told me that at the top of the market between 2020–2022, Greenoaks probably returned more money to investors than anyone else. “Neil’s extremely disciplined, he’s gone against the tide many times, and he’s had exceptional timing,” Kravis said. “He’s the real deal.”

The firm has now achieved icon status among founders and investors—but casual observers more familiar with names like Sequoia and Andreessen Horowitz might have come across Greenoaks for the first time only recently. In February 2025, Bloomberg reported that Ilya Sutskever, one of the founding fathers of modern AI, raised over $1 billion for his startup, Safe Superintelligence (SSI), at a valuation of more than $30 billion—six times what it was less than six months before. Leading the deal, according to reports that neither SSI nor Greenoaks have confirmed or even acknowledged, is purportedly Mehta’s shop, with an investment of over $500 million.

SSI operates in complete secrecy, and says it doesn’t plan to release a product until it develops artificial superintelligence—the term for a machine that would far surpass human intellectual capabilities in virtually every domain—as distinct from the artificial general intelligence (which would merely match or slightly exceed human-level cognition) being developed by foundation model companies like OpenAI, Anthropic, and xAI. But Sutskever, the former Chief Scientist and Co-founder of OpenAI, recently made the tantalizing claim that he’s identified a “different mountain to climb” from the model companies.

Over several weeks of interviews, in which he opened up to a reporter for the first time, Mehta declined to comment on SSI. But a potential clue into the matter, and into our future as a species, might be gleaned by peering into the life and work of Mehta himself, who over a dozen founders and investors interviewed for this story variously compared to Warren Buffett, Masayoshi Son, and Elon Musk—often in highly emotional tones.

Bom Kim, who is normally the Bill Belichick of tech when it comes to interacting with reporters, agreed to speak at length with Colossus Review about Mehta. “From the day we met,” he said, “I’ve believed that Neil is going to be a household name.”

CAROLYN FONG

“I’m best friends with my parents,” Mehta told me on a miserably wet morning as we walked through his Pacific Heights home, where an agreeable playlist of Bob Dylan and Amy Winehouse played quietly on surround sound. His favorite part of the house are the three kids’ bedrooms, which are vestibuled like a passenger train, with double sliding doors connecting each room to the other. “I’ve never had a bad day with my parents.”

As yours might have done, my eyes rolled back into my skull when I first heard Mehta say things like this, which in this case he did as ‘Visions of Johanna’ played from the small white speakers in the ceiling of his living room. Yet as I eventually sensed, such sentiments are frank and deadly serious, and part of the key to unlocking Mehta’s unusual genius for both business and partnership—and to dissolving the cynicism one initially feels when they observe the apparently unaffected bounce in his step.

Mehta’s parents, Nitin and Meena, grew up in the slum town of Palanpur, in the state of Gujarat, and later came to America in the first great wave of Indian migration that began after the Immigration and Nationality Act of 1965 abolished the national-origins quota system established in the 1920s, replacing it with a skills-based preference system. At the time, Gujaratis had mostly dominated the hotel, restaurant, and retail trade sectors, while IT was mainly the preserve of Telugus and Tamils. But like many Indian immigrants to America of the time, Nitin was an engineer, and came to Rapid City in 1967 to attend the South Dakota School of Mines & Technology, before spending a long career at McKinsey.

Many of Mehta’s earliest memories are of obsessively discussing business with his father, like the time they returned home from a fifth-grade classmate’s mini-golf birthday party at the Malibu Grand Prix in Redwood City. For most children, the outcome of having fun at mini-golf is that they want to play more mini-golf, perhaps even golf. But the fifth-grade Mehta had so much fun that he wanted to “get into the mini-golf business,” and so he and his dad modeled it out on Excel. “I can’t remember a day when I wasn’t talking about businesses with them,” he said.

There was also the time they visited Niagara Falls, when Mehta was 11. They woke up at the crack of dawn, and his dad piled Neil into several coats before boarding the Maid of the Mist for a freezing cold boat tour. It turned out the Maid of the Mist was basically a motorized sardine can that rides out to the Falls for a quick photo and then turns around and comes back. Aghast, Mehta checked the tour tickets and noticed they were 15 bucks a pop (this is the 90s, mind you). So the boat docks and one of the deckhands is just standing there, letting all the frozen begoretexed sardines off the tour, when a small Indian child obscured by multiple layers of down walks up to him and demands to know why no one is competing with them for prices. The apparently forbearing deckhand informed Mehta that Maid of the Mist has an exclusive concession over the port that grants them a monopoly.

The experience led Mehta to an interest in the idea of cornered resources, and he started looking into aquariums, Six Flags Magic Mountain, and Madame Tussauds (“No one’s going to build a second wax museum in London,” he remembers thinking). “I loved sports, but I had a terrible jump shot, I wasn’t that good at soccer, I’m not that fast,” he said. “I was captain of my debate team, but I wasn’t that proud of it. What motivated me was investing in companies, even back then.”

In high school, Mehta was a shiftless student who skipped class and soccer practice, and was kicked out of Spanish for refusing to speak anything other than English. He made money on the side as a top-ranked door-to-door seller of Cutco steak knives, cleaning up from stay-at-home moms with discretionary income in suburban Silicon Valley. He also tried launching a company called Zymst, an online high school yearbook catalog where students could edit their written comments on someone else’s photo from the previous year, based on whether they still liked them or not. When he graduated, he was given a card, signed by his teachers, which read, “Remember, Neil: In real life, detention is called jail.”

Mehta wasn’t motivated by money, per se. “It was more like a puzzle, like a mathematical equation,” he said. “It was deducing things down when there’s all these various opinions. And when you’re young, so much of your life is like, you’re being told, ‘this is the way the world works’. So this was one of the only ways that I knew how to express myself, that maybe I had differential insight or opinions, and maybe the world doesn’t work exactly like they say. And then when you’re right, it feels really rewarding, especially when you’re young.”

“Not exactly,” he said, when asked if his interest in founders in particular only came later. “A lot of my view of the world was formed from my grandfather,” a Jain from Ahmedabad, about a three-hour drive from Palanpur. “He was very simple, grew up dirt poor, and he owned a little gun shop. He sold old pistols and rifles from the Maharajahs. He loved art, he loved music, and guns to him were the equivalent of architectural masterpieces. He’d spend hours explaining to me a scene crafted on the husk of a gun, telling me the story of why it was crafted. It would be like an hour-long story, and I’d be captivated.” Starting when Mehta was nine, his grandfather would set up balloons and bottles outside the garage of his home in Mumbai and teach Neil and his younger brother to shoot them, to their mother’s horror.

“He cared a lot about quality, a lot about craftsmanship. He cared a lot about heritage. When you’re young, these aren’t concepts that you immediately understand. The finishings of a pistol, or the colors that were put into a rifle 120 years ago, and the relevance of why these colors were picked … I think it was imbued in me fairly early from my grandfather that producing things of value, producing craftsmanship, caring about the quality of finishings, caring about the artistic beauty of something—those are things that could bring out enormous joy in life.”

“Then if you pair that with my dad’s influence,” he continued, “I became a big believer in capitalism, really company formation, as the primary source of human progress. I’m a fervently religious believer in capitalism … So the reason I do what I do is because I want to be a small part of being in service to the people creating those companies in the capitalist structure that allows humans to progress. And the quality and the beauty and the finishings of how they do that, and how they deliver it to humanity, are really important to me. I think it all comes from my grandfather, and some of it from my dad.”

Mehta pointed me to an internal Greenoaks document called ‘Our Soul’, in which he explains the firm’s relationship to founders. “I never really made the connection until now,” he said, as we sat in a conference room at the Presidio offices of Greenoaks. “I’ve never said this out loud. My wife doesn’t even know most of this. But it does feel kind of obvious as I say it: There’s a part in ‘Our Soul’ that’s directly attributable to my grandfather. Which is that I talk about founders as artists. ‘Each one’s painting their own painting.’”

“In the document, I tell the team: It’s easy to have opinions and spend other people’s money. But that’s not our job. We’re in the business of understanding what founders are doing, and being humble and curious and empathetic about it. Our job is to figure out how and why they’re painting their painting. That’s it.”

“It’s so easy to be an art critic,” he said. “Understanding what a painter’s actually trying to do—that’s the hard part.”

Agony can be hard to spot in the bright life of Neil Mehta, but an early bout arrived during his freshman year of college, when he abruptly left California for Mumbai, where his grandfather was dying.

Mehta was 19, and the Asha Parekh public hospital in Mumbai was about as rough as you’d think. “I had a hard time digesting it,” he said. “Really, he raised me. My parents were amazing. But my dad traveled a lot, and my grandfather was the closest person. It felt like I had this person who was also my fan and always there for me, and no matter what happened in the world, I’d be able to talk to him and have his support and love … In Mumbai, that was the first time I was like, Oh. It’s just going to be me.”

To alleviate the loneliness of his grandfather’s hospital room, Mehta tried calling his girlfriend, whom he’d been dating since high school. It was 2002, and the only phone available was one of those ridiculous white hospital landlines with hundreds of buttons. They’d agreed on a time to speak, but Mehta called her something like 10 times with no answer. After several hours he finally got through to her, and while Mehta stood in the dingy hovel where he would shortly lose the most important person in his life, his girlfriend said to him, “I want you to know something, Neil. I don’t love you anymore.” And then she hung up the phone.

“Never talked to her again,” he said, like a boy whose dog ran away but is trying to be brave about it. “And I was like, well. This fucking sucks.”

It wasn’t the Siege of Leningrad, exactly. But Mehta didn’t handle it well, at least by his standards. He went home to Atherton and hung around Santa Clara University, going to class less and less, opting instead to swim ponderously at the YMCA pool, smoke a lot of weed, and read Schopenhauer and Victor Frankl, as is a heartbroken young man’s balm. “My mom was really worried about me,” he said. “My dad was like, he’ll figure it out. I would just roam around. But I enjoyed the wandering. My parents never rushed me to get out of it. They were awesome about it, in retrospect. I owe them a thank you for that. For just letting me have space to wander.”

‘Wandering’ for Mehta took all of a semester, at which point he looked in a mirror, he said, and told himself that it was the last bad day he’d ever have, as if it was his choice. “I’ve worked really hard to make every day better than the previous one since then. My grandfather would tell me that all the time. I try to live by that.”

Mehta with his father and grandfather, 1986

Mehta finished college at the London School of Economics (LSE), where due to an administrative error he was assigned to live in student accommodations with four girls—an otherwise nightmarish experience redeemed by the fact that one of his roommates, Jash, became the love of his life, and later his wife and the mother of his three daughters.

After LSE, Mehta took a job as an associate at the investment firm Kayne Anderson, where one day he uncharacteristically had to be physically separated from a partner and sent home after they nearly came to blows over an investment that Mehta, who was six months out of college, thought was shit. The fact that he wasn’t fired—he was promoted, in fact, after wringing an admission from his boss that he was right about the quality of the investment—gave him confidence in his budding value as an investor, and his ability to think for himself.

That continuity between what he calls his “internal and external voice” evolved at Kayne Anderson, but only really took off once he landed the job with D.E. Shaw. “It was the most formative experience for me,” he said of his time in Hong Kong. “It took going there until I felt really alive. It woke me up in a way. At Kayne it was already happening, but I was still part of the system. In Hong Kong, the shackles came off.”

“I didn’t have anything else in Hong Kong. I had no other responsibilities. I had no obligations. I had nobody I needed to see for dinner. I couldn’t watch sports, the time zones were all screwed up. And I didn’t care to go out. I had nothing. And honestly, I turned into someone else.”

“Then when I came back here and started to build Greenoaks with Benny,” he reflected, “that’s when it really crystallized. I knew then my source of joy was going to be this. I didn’t need to complicate it. By now I’ve had enough feedback from the successes and failures of working with founders—and in the last 13 years I think we’ve captured a disproportionate percentage of great founders—that I’m like, ‘No, this is it. This is what my purpose is in life.’”

“There’s something remarkable about Neil that is so unique to the way his brain works compared to anyone else I’ve met,” I was told by a Silicon Valley source intimately familiar with Greenoaks. “Which is that Neil has an extremely high internal locus of control”—meaning the degree to which someone believes he has control over the circumstances of his life. “When you consider that everything in the world was first a thought in someone’s head, that’s an expression of how humans can impact our environment. And Neil is a force in this way. I have mental models of probably a thousand people in my brain, and Neil has the strongest internal locus of control that I’ve ever encountered. I can’t emphasize this enough. He can break reality.”

CAROLYN FONG

Bom Kim, the Founder and CEO of Coupang, contrasted this quality of Mehta’s with most other investors, who will often say to founders with crazy-seeming ideas, “‘Look, I actually think you’re right. But if all the sailboats are moving in one direction, and I’m the only sailboat moving in the other direction, if I’m right, I get promoted; if I’m wrong, I get fired.’ So think about the competitive advantage someone like Neil has when he’s not beholden to that framework. Which takes courage, by the way. So it’s not only the first principles, clear thinking, the conviction-driven thinking that he operates with. It’s also having the guts. I think from Greenoaks’ early days, Neil’s always told LPs, ‘This is who I am. This is how we operate. We don’t diversify. We concentrate, and we drive on X.’”

“Neil’s got an incredible IQ. He’s also got incredible EQ, which wins people,” Kim said. “But there are a lot of people with EQ-IQ combinations. Not that many, it’s still special. But it’s not what makes Neil rare, which is those plus conviction and courage.”

“He tries to kind of shape or bend the world, rather than bending to it.”

Mehta and Benny Peretz started Greenoaks in 2012, in a foul, borrowed backroom office of an insurance broker in San Francisco. “We started with a very clear mission,” Mehta said, “and it hasn’t changed. Find great founders building great businesses, become their single most important resource, work tirelessly to invest in them, and then become the most important partner they have. And do it over the course of decades, not years.” 

It turned out that back in 2012, the world was open for business. Mehta and Peretz cold emailed hundreds of people, everyone they could think of that they wanted to meet, and pretty much everyone responded, including Patrick and John Collison, Jack Dorsey, Travis Kalanick, and Joe Lonsdale. Tom Hardy, an early Greenoaks hire, remembers some of the biggest hedge fund and private equity figures in the world gravitating toward Mehta. “It was kind of a game-recognizes-game situation,” Hardy said, “where these people who were objectively much more successful, much further on in their careers, gravitated toward Neil almost more than he did to them. Some of them practically stalked him.”

We’re only focused on two things in life: great business models and great founders. When you find them in the same situation at the same time, we go all in.

–Neil Mehta

Greenoaks’ first official investment was a $3 million secondary transaction in Palantir, which Lonsdale permitted the unknown but promising Mehta on the condition that he got his shit together, which presumably meant moving out of the broker’s office. Other early investments included Flipkart, now the largest ecommerce company in India, and OYO Rooms, another Indian unicorn.

But the investment that put Greenoaks on the map was Coupang, which technically began a couple of years earlier. Mehta came up with the name Greenoaks, in fact, in order to sound legit when he approached Bom Kim, whom he’d met through a friend, to inquire about backing him. Kim was then in the process of dropping out of business school to start the Groupon of South Korea, selling daily deals.

“What’s Greenoaks?” Kim asked. “Well,” said Mehta, “we invest in companies like yours.” “Uh, okay,” Mehta remembers Kim saying.

“I just remember meeting him and thinking to myself, this guy is a genius. He’s clearly going to be the Michael Jordan in this space,” Mehta said. “I knew I wanted to play with him for the rest of his career. The ability for him to articulate what is most jugular, go attack it and solve it, and then demonstrate how he solved it, that feedback loop, you see it once in a generation.”

Today Mehta is the Lead Independent Director of the $45 billion company, and Greenoaks has led five of its eight rounds, investing nearly $1 billion in Coupang across a decade. But while Mehta speaks often of his good fortune to have met Kim so early, the story of how they got to where they are now belies the idea of luck.

At a certain point early on, Kim realized that Coupang wasn’t going to be Korea’s Groupon but a global ecommerce juggernaut. By the time of the company’s Series B, Mehta had essentially bet all of Greenoaks on Coupang, and was de facto involved in the fundraising. Kim and Mehta invited all the biggest Bay Area and New York investment firms to visit the office and meet the team. “It was like an emperor has no clothes moment,” said Mehta. “These are all the most lauded firms I’m supposed to respect, and they’re all asking the wrong questions.”

For example, at the time Coupang had about 8% gross margins, and the firms didn’t like that. “Amazon’s gross margins are 30%,” they said. “How are you going to get to 30%?” What they should have done, Mehta explained, was look at Coupang by category. An ecommerce retailer adds product to its website in stages, and starts out with very unfavorable agreements with its wholesalers and manufacturers, which it then renegotiates over time as it builds scale. Criticizing Coupang’s margins at this juncture missed the whole point, said Mehta, which was that Kim was willing to take significant losses at the start while he built out a Jaw Dropping Customer Experience (JDCE)—which included the best overall delivery experience, and pricing Coupang’s goods to match the lowest prices in the market—understanding that with time, scale, innovation, supply chain optimization, and increasing operational efficiency, momentum in customer adoption would accelerate, and margins would eventually expand.

“Today, it’s a 30% margin business,” said Mehta. “But nobody asked about any of that stuff at the beginning. They came in and they just saw the low margin number and they walked out. They didn’t see it because they wanted the answer to be easy, rather than understanding that to build something great, it takes breaking tradeoffs that might not be obvious at the start. You have to go talk to the people involved in the business, the suppliers, the customers.”

“Back then, you could go up and down apartment buildings to talk to customers about Coupang, and they’d be crying, literally crying, from happiness. I remember watching with Bom as we saw one mother crying because the diapers she ordered were showing up at her door first thing in the morning—which meant she didn’t have to carry these giant boxes back home from the store. She was like, ‘If you took this away from me, I don’t know what I would do. Please don’t take this away from me.’ You can’t see that kind of affinity in a retention curve.”

“We had this huge transition where we had just turned profitable as a third-party marketplace, we were planning to go public,” Kim said. “And it was at this moment where a lot of people could cash out with very, very high IRRs. We were basically right before the final step of going public, and had spent months and months preparing, and I pulled the plug. Because I was convinced that there was this huge opportunity if we made a very big bet, which was to invest in not only a fulfillment network, but also a logistics network from scratch”—a process that included building a cutting-edge warehouse management system, football field-sized warehouses, delivery camps, localized distribution points, custom packaging, etc. “And we had lots of investors, even board members who were very familiar with Amazon. And they said, ‘Amazon couldn’t make that successful, therefore you can’t.’”

“That’s not clear, first principles thinking,” Kim explained. “That’s like saying, well, a bird flaps its wings, therefore an airplane has to have wings that flap, and metal doesn’t flap, therefore you can’t fly. Whereas Neil, who instead of seeking safety, really tried to get to conviction. That meant he spent hours and hours and hours with this kind of voracious hunger to understand the details and new concepts. And as much as we would open up, he was willing to spend time to get to conviction on either a yes or a no. And it hasn’t always been a yes.”

Coupang is also the investment that first gave Greenoaks its enduring reputation for being a founder’s go-to partner in an emergency. The reader may recall that in the fall of 2017, there was a wee chilly moment when the nuclear-armed totalitarian dictator of North Korea used the 14th-century, late-Middle English term ‘dotard’ not once but twice to refer to the President of the United States, himself the commander of 5,000 nuclear warheads and no great paragon of rhetorical restraint. Kim had been in the process of fundraising for months when spooked US investors pushed Korean equities off a cliff. After several rounds of discussions about how to move forward, Mehta and the Greenoaks team flew to South Korea and injected half a billion dollars into Coupang shortly thereafter. That investment returned more than tenfold in a few years.

A lot of them talk about how the secret to being a great investor is leaning in when others lean out, and vice versa. It’s very easy to say, but really terrifying to do in practice.

–Parker Conrad, Rippling

“Nobody else wanted to do it, we were the only ones,” said Mehta. “All of our alpha comes from ignoring the memetic vibes of Silicon Valley and New York and focusing on the fundamentals of a business and the founder. That’s it.”

“It takes a lot of stamina, doing that kind of deep dive, it’s physically grueling,” Kim concluded. “We have a phrase in our company, ‘You’re still on the helicopter’. You need to land the helicopter, get on the ground, you should be on your knees getting dirt in your fingernails. Most people bring the helicopter down from 10,000 feet to 8,000 feet and think they’re diving deep, but they can’t get off the helicopter. Is it because they’re not smart? No. It’s because it takes too long to get to the core of the issue when there’s a lot of ambiguity and too much information.”

“How did Greenoaks become a machine? It’s because Neil gets on the ground, gets dirt in his fingernails. And it’s unpleasant, it’s tiring. It’s physically and mentally grueling. But I’ve seen him get off planes, lock the door, and spend six hours with me or other team members, just going through all the details. How many investors do you think just send due diligence teams to do CYA [cover your ass], versus getting to conviction one way or another? That’s the thing with Neil: one way or another.”

Such virtuosic stories of being early and right, and of making more on a single investment than the returns of most other investors’ entire funds or careers, are legion at Greenoaks—including of having predicted the Silicon Valley Bank run of 2023 in a letter that saved portfolio companies with money there; and of Wiz, the Israeli agentless cloud security company, which Mehta and his partners didn’t know existed until they sussed it while diligencing the company’s competitors—at which point they called the Founder, Assaf Rappaport, talked to him for 30 minutes, dropped the competitors, and a week later led a round at a $1.7 billion valuation when the company was doing only $2 million in revenue. (In March 2025, Google acquired Wiz for $32 billion.)

There was also Rippling, which had hundreds of millions of dollars’ worth of customer payroll potentially stuck in SVB as the bank was collapsing. The company’s Founder and CEO, Parker Conrad, called Mehta on that Friday to float the idea of backstopping its customers’ payroll using the company’s balance sheet. “And I sort of said, ‘Look, this only works if you can wire us the money on Monday morning,’” Conrad told me. “So we would need to negotiate the deal, sign the term sheet, and wire the funds by then … And we were like, you have to understand: We’d take this money, and there’s a decent chance we’d send it right out the door to customers to cover payroll. That was a possibility, and it was also not clear how Rippling would be viewed after this. We were talking with Greenoaks about, what if 30% of our customers leave us next month because of this issue? It was entirely possible he was putting money into a sinking business.”

“And Neil was like, ‘Great, we’re in’—right on the phone,” Conrad remembered. “And we signed a term sheet at 9pm Friday night. So this was a $500 million investment round that went from a handshake to a term sheet in 12 hours, on a day when the financial world was melting down … There were no other investors doing that. A lot of them talk about how the secret to being a great investor is leaning in when others lean out, and vice versa. It’s very easy to say, but really terrifying to do in practice.”

CAROLYN FONG

Then there was Carvana. Like the Imagi team, like the Collison brothers and Kim and Rappaport and Conrad, Mehta fell for Ernie Garcia. Garcia founded and built Carvana, the online used car retailer, in the private market without taking a dollar of venture capital, and not for lack of trying. When Garcia went out to Silicon Valley to raise money for his company, nobody would give it to him.

Carvana went public just as Greenoaks had started to get going. The price was high, but Carvana was the acme of the JDCE, which by that point was becoming the Greenoaks mantra. “We didn’t know Ernie yet,” said Mehta. “But I knew it was a Jaw Dropping Customer Experience for anyone looking to buy or sell a used car, that it was irresponsible for anyone looking to do that to go anywhere besides Carvana. It was just clearly what Amazon did for books.”

As the pandemic unfolded, Carvana’s price collapsed, so naturally Mehta wanted to buy. Several Greenoaks investors went ballistic, figuring that once COVID eased, the era of buying used cars online would be over. “This is a terrible trade, Neil, don’t invest, this company is going to go bankrupt,” that kind of thing. And they were right, in a way: Garcia did a $2.2 billion acquisition of ADESA, a physical auction business, but then the cycle turned, the country went from low to higher rates, growth slowed, people stopped buying cars, and Carvana underwent an infamous peak-to-trough drop.

Not many companies go from $60 billion to less than $1 billion in market cap unless they’re frauds, Mehta pointed out, and that’s the headline the media ran with. Every news story in those days was about Carvana approaching bankruptcy, about how Garcia had a secret deal with debt holders, that he was going to inject capital and wipe out the equity. Yet once the stock price dipped below $20, Mehta and Greenoaks got serious.

“There was an insane amount of noise,” Mehta said. “So we just rolled up our sleeves and tried to understand what was happening from first principles. We spent a lot of time talking to customers of the product, a lot of time talking to competitors of the company, a lot of time understanding the cost structure of the business—an inordinate amount of time understanding the capital structure of the business. And we started buying stock.”

It was unusual for a mostly private investment firm to buy stock, especially as Carvana went from the $20s down to the $10s and eventually to $5. By that point Greenoaks had returned something like $11 billion to investors, but the phones were still ringing off the hook with LPs saying things like, “What the hell do you know about Carvana? You don’t know anything about Carvana! How do you know more than Apollo [the lenders]?” But Greenoaks kept buying more stock until it bought about 4% of the company, which had its own recurring ‘suicide watch’ features in the business press, counting down the clock to bankruptcy.

“It was hard on all of us,” Mehta said. “If you haven’t been through something like that, which many of our younger colleagues hadn’t, it can be demoralizing. Terrifying. You start asking yourself, what’s the point of doing all of this work? We’re screwed.”

“In other times, we’ve had differentiated insight and it’s been obvious internally that we had it. That’s when you see something in the numbers that nobody else sees, and we believe we’re right and everybody else is wrong. That was happening with Carvana, too. But for some reason, because it was a public company, seeing a price you buy out go down by half the next day, and you’ve put a lot of your fund into it, and then immediately lose money on it, there’s some psychological effect. It was easy to worry about our LPs firing us.”

“But the thing is, we’re not playing the game of, ‘Is the stock going to go up or down tomorrow?’ The game we’re playing is, ‘Will this company delight consumers in the tens of millions over the fullness of time, produce real gross profit per unit, and be a free cash flow machine run by a great founder that could compound for the next 10 years? Is Ernie building a generational company that’s going to be a meaningful part of the S&P 500 over time, or isn’t he?’”

“So there was a moment where I was like, look, maybe we just need to go understand this. We have to stop talking to Ernie on Zoom; we have to go spend time with him.”

Mehta flew out to Arizona to see Garcia alone, and spent four or five hours with him over dinner. It was at the peak of rumors that Garcia had a secret debt deal with Apollo and was waiting to bankrupt the company. Mehta remembers Garcia with his soul between his teeth, “not because of what was happening, or because of the character assassination in the headlines,” he said, “but because of how he had to manage his employees. He was going through a way worse version of what I was going through with some of our own team members who were anti-Carvana. Ernie was having to explain to people that worked for him that he wasn’t a criminal, including spouses worried about their families who had until recently been so proud to work at this extremely high-performing company. He had tears in his eyes talking about how he had to explain to the families of people that worked for him that Carvana was going to be okay.”

“I remember thinking, ‘Either he’s a psychopath, or he’s going to be just fine.’ Because you can’t manufacture that level of empathy. So we just talked about the company, we talked about the strategy, how he was going to cut costs, how he was going to manage through this. It was all rational. It all made sense.”

Mehta acknowledged that Carvana was a very good investment for Greenoaks—the stock it holds is currently worth around $1.2 billion—but insisted he’s proudest of the other rewards. “First, I’m proud of the differential insight we had around the founder and the business. Ernie and his team did all the hard work, but it was rewarding being in their corner. Second, there are lots of firms that would have sold their investment the moment they were proven right about the company staving off bankruptcy. I’m proud that we were able to see through some of the worst times to understand how the company could grow into a category leader and the cognitive referent in the industry. So it’s not only getting the bankruptcy trade right and being mathematically sound—and being foundationally better than other people at the underwriting. It’s also about allowing yourself to be partners to the company as it goes and attacks something big. What I’m really proud of is that we’re the kind of firm that could do both.”

“I went to dinner with one of the most successful Silicon Valley VCs right after we had purchased the stock,” Mehta remembered, “and he was like, ‘Are you fucking insane? You’re playing the lotto. You don’t need to look at the numbers, just look at the stock price. It’s over.’ And I realized then that he didn’t actually know any of the numbers. He had just read the articles.”

Mehta confesses to some big stumbles over the first 13 years of Greenoaks, whose raison d’être, as he put it, is to capture “the very small number of the world’s founders who are going to produce a significant proportion of the value that humans enjoy.” When asked which founders he’s missed, Mehta responded, without missing a beat, “Tony Xu, Brian Armstrong, David Vélez … but no one bigger than Elon Musk.”

Mehta and Peretz had known about SpaceX early on, but some of their industry mentors warned them away from Musk. “He fires people quickly, he micromanages people like crazy, he disappears for large swaths of time, comes back in and changes everything,” Mehta remembers being told. “And we were like, well, I guess we can’t back him then.”

“We didn’t do the primary work ourselves,” Mehta ruminated, venting his spleen. “We outsourced that work. I cost our investors billions of dollars because of that. We later became small investors [in SpaceX], but I’ve had to look our investors in the eye and tell them we’ve lost them tens of billions of dollars because of the mistake we made as a firm on that. And I told myself then: I’ll never not do the work myself again.”

We can understand from the outside-in better than almost anybody else on Earth. There’s no need to explain the 101 or go through remedial background. For a founder, that makes a first meeting with us feel like a fourth or fifth.

–Neil Mehta

The machine Greenoaks has fine-tuned ever since is purpose-built to correct for that early error. “If there have been 100 billion people that have lived on Earth, and between 10,000 to 100,000 who have affected the technological progress of humankind,” he explained, “our job is to find the few hundred living now that could join the pantheon of great humans that have driven humanity forward. So first, we’re focused squarely on that. Second, where we’re differentially great partners, where we generate alpha, is by having a deep understanding of the business model.”

The process looks something like this: Greenoaks selects for people who believe that it’s really hard in the world of capitalism—which is filled with “me-too products swimming in a river of beta,” in Mehta’s phrase—to build something that delights human beings at a differential rate. The world therefore moves forward only through the products and companies built by generational founders who break tradeoffs and build competitive moats by creating a Jaw Dropping Customer Experience, which “usually starts with doing something perceived to be impossible, either technically or operationally, that gives competitors nightmares. Everything else is just a shell game,” Mehta believes.

Each year, Greenoaks identifies about 10–15 people who might be like this, and to whom the firm could be a uniquely close partner. Before any meetings, they set about preparing. “I don’t mean go on the website and use the product a little bit,” Mehta said. “We’ll talk to their customers, examine exactly what competitors are doing, understand the product in a granular way, study the underlying technology and how it’s evolving. There’s a series of things we’re testing for, depending on the company. We can understand from the outside-in better than almost anybody else on Earth. There’s no need to explain the 101 or go through remedial background. For a founder, that makes a first meeting with us feel like a fourth or fifth.”

Then there is the pace. “I think what we’ve become much better at,” he said, “is increasing the speed and velocity in our information asymmetry—getting much more information and being able to generate differential insight that matters to long-term enterprise value, and putting those three things together with a small team and building a flywheel for doing it over and over again every day. That’s been a sea change in the last couple of years.”

After preparing, Mehta and Peretz spend several hours discussing each company. While talking, they’ll call various partners, friends, and colleagues to join the conversation for an hour or two. Then they go home, put their kids to bed, and talk again on the phone, often from 9pm until 1am, nearly every night. The overwhelming majority of these conversations end in a decision not to pursue. “And then we just drop it and let go,” said Mehta. “Do you know how infuriating that would be to most people? But we love it. So we ‘wasted’ four hours. Who cares? We’re trying to find someone we want to partner with for decades who’s going to create art that changes the world. They’re painting their painting, and we want to find them. It has to be perfect.”

When it’s a yes, Greenoaks does not lard the process with advisors, large investment committees, or long, asynchronous diligence sessions. Mehta personally leads every first meeting with a founder, typically on site at the company rather than in the Greenoaks offices. “I want to be very explicit about this,” Peretz told me. “Neil joins every single investment process from the very first meeting with the founder.”

“If I walk out of a meeting and I don’t want to spend time on something that has five of the best firms in the world on the cap table,” Mehta said, “I feel totally comfortable killing it. I won’t lose a wink of sleep. And the opposite is true, too. If I find something incredibly interesting, but it’s a little crazy and a little out there, which a lot of the things we get excited about are, I’m totally comfortable pursuing an investment of $500 million or more in the next 36 hours. And I can marshal the resources and feel comfortable that if I’m wrong, we’ll live to fight another day, as long as we did the work.”

Monday mornings are for the Greenoaks pipeline meeting, and for spending time as a team talking about what they want to prosecute, what they want to drop, and how they want to organize time. Peretz in particular plays the role of extending the firm’s time horizon, according to Mehta. “Sometimes when things are moving really fast and you’re in the fog of war, you find yourself trying to make decisions that are optimal in the short term because you can only see that corner,” he said, with the gooey eyes he gets when discussing his partner. “Benny has an astonishing ability to step back and remind everyone, including me, what we’re trying to optimize for over the long term.” On Monday afternoons, Mehta occupies the ‘War Room’ (it’s just a room that connects Mehta and Peretz’s offices), where he talks to people about companies until around 8pm, at which point the team goes to dinner—the only night of the week he doesn’t see his daughters.

Tuesdays, Wednesdays, and Thursdays are for calls and company meetings. Fridays are for “deep work,” which usually means Mehta sitting alone in an office writing about a company or studying it, or reading through a pile of papers on Bitcoin, childhood gaming apps, or AI labs. On his desk, Mehta also keeps a list of the S&P 500, “and I just try to figure out what companies are not on that list today that will be on that list tomorrow, and how do I become the single most important partner they have?”

CAROLYN FONG

“This is controversial,” Mehta replied, when asked if the Greenoaks machine has identified an ideal type, “but I do believe there’s an archetype for a great founder. And I think that once you see it and learn it, it’s a repeatable process. We’re looking for remarkable intellect, extreme focus, an obsession with the customer, unreasonable determination, especially in the face of adversity, clear and credible ambition, and usually a bit of divergence—people who don’t feel the need to be liked by everybody.”

“A lot of people in our industry say they want these things, but we try to be more systematic and rigorous than anyone else about how we look for it. Every business is the cumulation of a million small decisions made by the founder, and when you look really closely at a company, it’s kind of a prism that shows you much more about them than any conversation does on its own. It’s like, Buffett says you could sell 20 IQ points and still be great; our process would probably allow you to sell more than that. It’s not that complicated. It’s the discipline of only looking for those types of businesses and those types of founders. There is no secret sauce. It’s just the consistency of doing this again and again across thousands of companies.”

When told that as a writer, and not an investor or founder, I find it hard to square the brute force of Greenoaks’ research process with his analogy to “perusing an art market, looking for the future Kerry James Marshall, or David Hockney, or Rembrandt,” Mehta replied:

“If anyone sits with Benny and me for a day, I think the thing they’re probably most surprised by is how much we talk about beauty. We love the analytical work, we love spending time on a P&L. But the whole point of that kind of work is to help us find the greatest founders building the most beautiful businesses that truly delight their customers. We love beautiful relationships, we care about beauty in the world … And understanding a founder, deeply understanding a business, that’s just the process of discovering beauty. That’s why we invest in the companies we invest in.”

Which brings us back to Ilya Sutskever, and reports that Greenoaks may be investing more than $500 million in Safe Superintelligence.

Although neither Greenoaks nor SSI have acknowledged the story, and Mehta did not respond to requests for comment from Colossus Review, it is worth a modest attempt at speculation, as the 40-year-old Mehta enters what he hopes will be the beginning of Greenoaks’ prime.

The first thing to note is that Greenoaks has remained conspicuously removed from the $1 trillion stampede of investment into the companies chasing artificial general intelligence (AGI) since the release of ChatGPT at the end of 2022. Greenoaks is invested in Scale AI and Databricks, but has been only a small investor in OpenAI, and has not invested in Anthropic, Cohere, xAI, Mistral, or any of the other foundation model companies.

What Mehta was willing to discuss is why Greenoaks has largely stayed out of the model war, in which each player insists, almost daily, that AGI is around the corner. “They may evolve to become great businesses, like ChatGPT is, but in their first incarnation they are all kind of bad business models,” he said. “Huge capital investments up front to create this asset, the asset is worth some amount of money, which then depreciates over the course of 12 months, so you have to reinvest again 12 months later. It’s like the airline business in the 1980s; you invest in the best fleet, but then 12 months later the other airline has the newer models, and you don’t pay back the cost of your initial capital investment because the unit economics don’t work. That’s the AI model companies. They have no competitive advantage. If you create a brand like ChatGPT, or if you achieve so much scale that you capture all the capital and no one else can compete, maybe you can escape that. But it’s not obvious that everyone does.”

Sources sympathetic to Mehta’s viewpoint who requested anonymity explained that the business limitations are inherent in the technological ones. In their telling, the model companies are all doing more or less the same thing, which is scaling transformer models: building massive data centers, using ever more GPUs, buying huge amounts of label data to unhobble the models, and then training the models with more compute and more data, over and over and over again. In a word, they’re merely scaling the horsepower of transformer models, rather than innovating on the underlying model. Insofar as this is a fair characterization, it would explain Mehta’s skepticism of the potential for returns, competitive advantage, or anything else of classical business interest, which powers the Greenoaks machine.

It might also go some way to answering the question posed by Dwarkesh Patel in his August 2023 interview of Anthropic CEO Dario Amodei, which did not appear to receive a satisfactory answer. Patel asked why the foundation models, which have memorized the entire corpus of human knowledge, and can easily reference it within microseconds, have not been able to make a single new connection that has led to a discovery—whereas even a moderately intelligent person with that much knowledge memorized could make a connection that leads to, for example, a medical cure.

The answer, or part of it, is that the foundation models are essentially colossal word replication and regurgitation machines. This technology is in fact likely good enough to reach AGI, which could lead to trillions of dollars in productivity gains over the course of time—no small achievement, to put it mildly. But they are unlikely to ever match or exceed the possibilities of human cognition. For that, one would need to build a model that could learn a lot from very limited amounts of data, and generalize from first principles. Such is the threshold for superintelligence, which could make the kinds of connections and discoveries that would elude AGI.

What Sutskever’s claim to have found a “different mountain to climb” suggests, and what the sextupling of SSI’s valuation in a little over five months appears to imply, is that the company’s progress toward superintelligence is going quite well. At the very least, if the reports of Greenoaks’ investment are accurate, one could reasonably infer that Mehta was shown something that met every one of Greenoaks’ criteria described above—which the foundation model companies largely have not. “We’re simple people,” Mehta told me in a different context, when discussing Parker Conrad and Rippling, to which he wrote a $500 million check overnight during the SVB crisis. “We’re only focused on two things in life: great business models and great founders. When you find them in the same situation at the same time, we go all in.”

Like Conrad, like Musk—like the archetype of a founder, in other words, that Mehta says is repeatable, and which Greenoaks pursues with unusual ferocity, and is anxious never to miss out on again—Sutskever underwent a painful and controversial exit from his previous venture, and now leads a company he’s said is “very personally meaningful to him.” There is also the matter of Sutskever’s raw intellectual firepower. He authored the 2012 AlexNet paper, which catalyzed the modern AI boom, developed Recurrent Neural Networks and Sequence-to-Sequence Models at Google Brain, co-founded OpenAI, and served as Chief Scientist for GPT 1-4 and the company’s reasoning models and alignment research. His murky, convoluted, and protracted departure from OpenAI, moreover, appeared to stem from an authentic intellectual break with CEO Sam Altman, over which Sutskever was prepared to cut his umbilical connection to the company he co-founded, which at the time was soaring past a valuation of $150 billion. Taken together, Sutskever evidently possesses the type of N of 1 conviction and sedulousness that Mehta believes makes “every great founder look approximately the same.”

It is likewise reasonable to deduce the nature, if reports are accurate, of Sutskever’s interest in Mehta. Given the total secrecy in which SSI operates, and will continue to operate until the very day, it says, it unleashes superintelligence upon the world, it would likely want a lead investor who does not employ a privy council of investment committees; who does not delegate research to subordinates; who is known for hiding his plumage; who has the implicit trust of his LPs; who is among the largest LPs in his own funds; who is highly concentrated; who pooh-poohs “venture math” and “rivers of beta”; whose firm, on principle, has never invested in China; who can understand a company from the outside in better than anyone else; whose first meeting often feels to the founder like the fourth or fifth; who is “the epitome of non-zero-sum thinking,” a combination of “warm, jubilant energy and high ambition,” and “the same dude across the board,” as Figma founder Dylan Field put it; who has “lost his taste” for any other type of work; and who—once he believes he’s understood a business better than anyone else—can write a $500 million check on the spot.

Perhaps it would also be of interest to Sutskever, who may or may not be within reach of a product that could change the species, that Mehta would view him as an artist; as a painter painting his life’s work; as a creator of beauty and innovations that multiply the supply of beauty in the world; as one of the 10,000 people who’ve ever breathed air who could affect the progress of humankind; as someone that Mehta would get off the helicopter for, onto the ground, with dirt in his fingernails, as his closest partner not for years but for decades, to build a Jaw Dropping Customer Experience that will delight human beings in the millions or billions, and produce a free cash flow machine that compounds ad infinitum.

In contemplating the possibility of such a partnership, one is reminded of Neil Mehta as a tiny figure in a vast and alien place, standing beneath an enormous structure towering above him, staring into the gorge of the future—and smiling.

Jeremy Stern is the editor-in-chief of Colossus.

Click here to subscribe to print for your office or home.