Every decade or so, a new computing platform emerges that fundamentally restructures the technology industry — and with it, the entire economy. The mainframe gave way to the personal computer. The PC gave way to the internet. The internet gave way to mobile. Each transition created a fresh set of trillion-dollar companies, destroyed incumbents who failed to adapt, and rewired the relationship between humans and machines in ways that seemed fantastical just years before.
We are in the early innings of the next platform shift. This time, the platform is not a chip or a network or a screen. It is a body. The humanoid robot — a general-purpose physical agent capable of operating in environments designed for humans — is crossing the threshold from science fiction into commercial reality at a pace that is outrunning most institutional investors' frameworks for evaluation. The capital flooding into the sector reflects this: in 2024 and 2025 alone, the leading humanoid robotics companies collectively raised well over a billion dollars in disclosed funding. And this is before any of the major consumer or industrial deployments have reached meaningful scale.
At Estes Capital, we have spent two years developing a thesis on robotics as a platform investment. This piece lays out why we believe the timing is right, what distinguishes the companies we think will define the platform, and how we are thinking about the seed-stage opportunity in the broader robotics stack.
Why Now? The Convergence of Three Forces
The idea of a humanoid robot is not new. Researchers have been working on bipedal locomotion since the 1960s. Boston Dynamics, now a subsidiary of Hyundai, has been building increasingly capable robots for decades — Atlas, their hydraulic humanoid, first walked publicly in 2013. And yet, for most of that history, the machines remained impressive demonstrations rather than commercially deployable products. What has changed?
The answer is the convergence of three forces that did not coexist at sufficient scale until recently. First, foundation models. The same transformer architectures that produced GPT-4 and Claude are now being applied to robotics policy learning, enabling robots to acquire complex manipulation skills from a combination of simulation data, video demonstration, and human feedback — dramatically reducing the cost and time required to teach a robot a new task. Second, hardware maturity. Actuator, sensor, and battery technology has reached a point where a humanoid robot can operate for meaningful durations in real-world environments without constant maintenance. The hardware is no longer the primary bottleneck. Third, and perhaps most importantly, the emergence of a commercial pull signal. Labor markets in the United States, Japan, Germany, and South Korea are structurally constrained in ways that were not true in prior decades. Aging populations, declining labor force participation in physically demanding roles, and the increasing cost of human labor in warehouse, logistics, and light manufacturing have created a genuine industrial customer base willing to pay for robotic automation at a price point that was previously unavailable.
Figure AI: Building the Platform Layer
No company better illustrates the maturation of the humanoid robotics sector than Figure AI. Founded in 2022 by Brett Adcock, the company has executed one of the fastest capital formation trajectories in recent technology history. In February 2024, Figure closed a $675 million Series B round at a $2.6 billion valuation — a financing that included OpenAI, Microsoft, NVIDIA, Amazon, Intel Capital, and Jeff Bezos's personal investment vehicle. The round's investor composition is itself significant: it represents a convergence of software platform incumbents, chip manufacturers, hyperscale cloud providers, and generalist technology investors — all placing a coordinated bet on humanoid robots as a foundational technology layer.
What distinguishes Figure's technical approach is its early commitment to a vertically integrated AI stack. Rather than acquiring robot-agnostic manipulation skills from third-party AI providers, Figure has built its own foundation model for robotic control — a system called Figure Neural Network, or FNN — designed from the ground up to handle the multimodal sensory input and real-time control loops that physical embodiment demands. This matters because the company is not just building a robot; it is building the operating system for robots. The commercial partnership announced with BMW to deploy Figure 01 units in automotive manufacturing represents the first major industrial customer validation of this approach.
For seed-stage investors, Figure itself is no longer a seed-stage investment. But what the company's trajectory reveals is the category of companies that are: the suppliers, component makers, data infrastructure providers, and tooling companies that will define the robotics stack beneath the platform layer. We will return to this point.
Physical Intelligence: The Software-First Bet
If Figure represents the vertically integrated hardware-software platform play, Physical Intelligence — founded in late 2023 by former Google Brain and Stanford researchers including Sergey Levine and Karol Hausman — represents a different and equally compelling thesis: that the core value in robotics will ultimately accrue to the software layer, specifically to the general-purpose robot policy that can operate across hardware form factors.
In November 2023, Physical Intelligence announced a $70 million seed round — itself an extraordinary number for a seed financing — led by Khosla Ventures and Lux Capital. By late 2024, the company had raised a total of $400 million at a valuation reported to exceed $2 billion, making it one of the most heavily capitalized seed-to-Series A progressions in recent technology history. The capital reflects the market's belief that a horizontal AI software stack for robotics — analogous to what Android provided for mobile or what Linux provided for servers — represents a winner-take-most opportunity.
Physical Intelligence's technical work centers on pi0, their foundational robot policy model. Early demonstrations showed the model controlling a diverse range of robot hardware — from industrial arms to mobile manipulators — on tasks including laundry folding, package sorting, and tabletop assembly, without task-specific fine-tuning. This is the key proof point: generalization across tasks and hardware, rather than specialized performance on a narrow use case. If the model continues to improve at its current rate, the long-term commercial implication is a robotics industry that looks more like the smartphone industry than the automotive industry: one platform, many manufacturers, winner-take-most economics on the software and AI layer.
1X Technologies: The Longer View
The third company worth examining in detail is 1X Technologies, the Norwegian robotics company founded by Bernt Øivind Børnich and backed since 2017 by Halodi Robotics before its rebranding. In 2023, 1X raised $100 million in a Series B led by EQT Ventures, with participation from OpenAI. The 1X story is important not because of its scale — it is smaller than Figure or Physical Intelligence — but because of its vision and its execution timeline.
1X's commercial product, EVE, is a wheeled humanoid designed specifically for facility security and logistics applications. Unlike Figure's focus on automotive manufacturing or Physical Intelligence's horizontal policy ambitions, 1X has identified a specific wedge market — security and logistics in commercial buildings — where the economics of robotic deployment are already favorable. The company's wheeled form factor reflects a pragmatic judgment: bipedal locomotion is not necessary for most commercial deployments today, and the additional complexity it introduces raises hardware costs without proportionate commercial benefit. This is the kind of considered trade-off that distinguishes world-class product founders from researchers building toward an abstract technical ideal.
The OpenAI involvement in 1X's financing is worth noting. OpenAI's investment thesis in robotics — which also includes their support for Figure AI — reflects a strategic view that language and reasoning models trained on text will eventually need embodied agents to fully actualize their capabilities. A robot that can understand natural language instructions and translate them into physical action is the consumer-facing endpoint of the AI stack that OpenAI is building. The question is not whether this product will exist; it is which company will build it first.
Boston Dynamics: The Standard-Bearer and the Cautionary Tale
No discussion of robotics as a platform would be complete without acknowledging Boston Dynamics, the company that has defined popular perception of what a capable robot looks like for two decades. Spot, the quadrupedal robot dog, is deployed across industrial inspection, construction, and public safety applications worldwide. Atlas, the bipedal humanoid, has been featured in viral demonstration videos that have accumulated hundreds of millions of views and played a meaningful role in establishing public credibility for the entire sector.
Boston Dynamics is also a cautionary tale about the gap between technical capability and commercial scale. Despite decades of world-leading research and hardware development, the company has struggled to achieve the kind of revenue growth that justifies its repeated acquisition premiums — first by Google (SoftBank acquisition of Google Robotics), then by SoftBank, then by Hyundai for $1.1 billion in 2021. The challenge has not been the robots; it has been the business model. Industrial customers require substantial integration work, extended proof-of-concept periods, and ongoing support that a hardware manufacturer without deep software and services capabilities cannot efficiently provide at scale.
This is precisely the lesson that the current generation of humanoid robotics companies has internalized. Figure, Physical Intelligence, and 1X are all building software-first or software-heavy approaches. The hardware is necessary; the software is the moat.
The Seed-Stage Opportunity: The Robotics Stack
For seed-stage investors, the primary opportunity is not in funding the next Figure AI or Physical Intelligence — both of which are now well-capitalized growth-stage companies. The opportunity is in the infrastructure, tooling, and enabling technology that the platform companies will need to scale.
Consider the analogies from prior platform transitions. When the iPhone launched in 2007, the multi-billion-dollar opportunities were not only in building competing smartphones. They were in building app development tools, mobile payment infrastructure, location services, mobile analytics, and the dozens of other software and services categories that the mobile platform required. The robotics platform will create an analogous set of opportunities, and most of them are currently at seed stage or have not yet been founded.
The categories we find most compelling at the seed level include: simulation infrastructure for robot training data generation; sensor fusion and perception software that can operate across multiple hardware form factors; industrial deployment tooling that helps enterprise customers integrate robotic systems into existing workflows; teleoperation and human-in-the-loop systems for training data collection; and safety certification and compliance infrastructure for regulated deployment environments such as healthcare, food processing, and pharmaceutical manufacturing.
Each of these categories represents a genuine wedge: a problem that is specific enough to be solvable at the seed stage, large enough to build a venture-scale company around, and strategic enough that the platform companies will either want to acquire it or partner with it.
What We Look for in Robotics Seed Investments
At Estes Capital, we apply a consistent set of criteria when evaluating seed-stage robotics investments. The first is genuine technical depth at the founding team level. Robotics is not a domain where smart generalists can acquire sufficient expertise quickly enough to outrun well-funded incumbents. We want to see founders with direct experience in robotic systems, control theory, machine learning for physical systems, or relevant hardware domains.
The second is a clear and defensible wedge. The robotics market is enormous in aggregate, but the path to capturing it runs through specific applications with specific customers. We are skeptical of companies that lead with a general platform vision without a clear first customer and a clear first use case. The companies that have succeeded in adjacent hardware markets — from semiconductors to industrial automation — have almost always won by being the best in the world at one specific thing before expanding.
The third is a considered view on the hardware-software split. Given the capital intensity of hardware development, we are generally more interested in companies that have a credible path to software-led unit economics than those whose moat depends primarily on proprietary hardware. This is not an absolute rule — there are cases where hardware innovation is genuinely defensible — but it is a strong prior.
The fourth, and most important, is founder conviction and time horizon. Building the robotics stack is a ten-year project, not a three-year project. The founders who will define this platform are those who are genuinely committed to the long arc of the technology — who understand the science deeply enough to know what is hard and what is merely unsolved, and who have the psychological resilience to execute through the inevitable setbacks that come with building at the frontier of physical and software systems simultaneously.
The Investment Thesis
We are in the first act of a platform transition that will take fifteen to twenty years to fully play out. The humanoid robot is not a product; it is a general-purpose computing platform with a physical form factor. The companies that define this platform — at the hardware, software, AI, and infrastructure layers — will be among the most valuable technology companies in the world by 2040.
The seed-stage window for the enabling technology layer is open now, and it will not be open indefinitely. As Figure, Physical Intelligence, 1X, and their successors scale, the problems they need solved will become large enough and visible enough to attract later-stage capital. The time to invest in the robotics infrastructure stack is before the platform companies have defined which problems they are outsourcing and which they are solving in-house.
This is the bet we are making. If you are building in robotics infrastructure, simulation, perception, deployment tooling, or safety systems, we want to hear from you. Get in touch.
