The Humanoid Takeover: $50T Market, Figure's Full Body Autonomy & Robots in Dorms

Peter Diamandis interviews Brett Adcock, CEO of Figure Robotics. Detailed Summary.

Section 1. The State of Figure Robotics.

Figure is about three and a half years old and Brett believes they've cracked the recipe for what a general purpose humanoid robot architecture looks like. They have four buildings on campus and are shipping Figure 3 robots off their production lines right now.

The company has a facility called "The Grid" which Brett describes as his favourite place. It's one of their buildings outfitted to run hundreds of robots 24/7. It has a second-story mission command post, like a "007 situation room", where operators can see every robot. The robots' own vision transmits back to mission control, so it's like watching through the eyes of soldiers on a battlefield. They plan to have 250 to 300 robots running simultaneously in this space, doing both home and commercial workforce tasks. The Grid just opened this week and they'll start shipping Figure 3s into it this month.

Brett confirmed they are sold out for about 3 years into the future. They can't make robots fast enough to keep up with demand. BMW is a customer, along with other industrial use cases.

Section 2. Helix and the Neural Net Revolution.

This is where things get really interesting. Figure's robot control has shifted entirely to neural networks with their system called Helix. Previously, robots relied on layers of traditional engineering: battery systems, electric motors, control software, embedded systems, sensors, structures. Now with Helix, the highest level behaviour telling the rest of the stack what to do comes from a neural net.

Brett gave a powerful example: in their kitchen demonstration, the neural net is making the planning decisions. It's telling the robot what to do next, knowing to pull the rack out of the dishwasher, to grab the water cups and not the coffee cups. No human in the loop. No pre-programmed sequences. Just a neural network making real-time decisions.

The critical insight Brett shared: "If you can tele-operate it, you can learn it." Meaning if the mechanical hardware can physically perform a task when a human controls it remotely, then the neural net can learn to do it autonomously. This is huge because it means the hardware is no longer the bottleneck. It's purely a data problem now.

Quote from Brett: "The only difference between why I can do logistics and now I can learn towel folding or dishes or manufacturing, literally it's just data. Data goes in the neural net, now I can do this work. The robot hardware doesn't need any updates. I just use new neural net weights on board."

Section 3. HARK, Brett's Vision of AGI.

Brett revealed something fascinating about his broader AI ambitions. He's building a system called HARK, which goes beyond what the major frontier labs are doing. His criticism of current AI is blunt: "When I talk to AI today or use it, I just feel like it's so dumb. It feels like you're starting a new chat, basically asking for knowledge retrieval. It's like an advanced Google search engine."

His vision of AGI is fundamentally different from the frontier labs. He wants to build Jarvis. Not abstract reasoning benchmarks, but a synthetic human that can reason, talk, have perfect memory, touch the world both digitally and physically, and be general purpose.

HARK can already do remarkable things. Brett asked it to build a monster truck for his son in CAD. From a single prompt, HARK went out, found a CAD package, installed it, opened it up, learned how to build CAD with the right parameters for monster trucks, and produced the design. End to end in under an hour from a single prompt.

Brett's frustration with the AI industry: "Everybody's copying the other frontier lab that's copying their frontier lab. Nobody's building true multimodal systems that really can reason and understand and have persistent memory and can go out and touch the world and do things."

He also praised OpenClaw, the AI assistant platform, saying it demonstrates how simple harnessing with markdown files and MCP tools on top of models like Opus can do "magical things for the world." His point: the capability overhang is enormous. The models can already do far more than products are letting them.

Section 4. The Road to the Home.

Brett is personally passionate about getting robots into homes, particularly for elder care. He grew up on a farm in Miko, Illinois, a town of 1,800 people with no traffic lights and no fast food. His parents own and operate senior housing facilities in the Midwest, so he grew up around senior care.

The timeline for home robots: By end of 2026, Brett aims to put a robot into an unseen home that can do fairly long horizon work. The metric they'll measure is human interventions, whether it's once an hour, once a day, once a week, or once a month. By 2027, they'd be on a path to ship to users' homes and ensure they work well.

Brett emphasised this won't be a big bang launch. It's iterative: work well at 1 home, then 10, then 100, then 1,000, then 10,000, then 100,000, then 10 million. An exponential growth curve.

He's already had robots in his own home with his kids, but they're monitored. His personal bar for readiness: when he can put a robot in his home fully autonomously, free reign, around all his kids, with no supervision. That's when it's ready for everyone.

Section 5. Safety, Privacy, and Asimov's Laws.

Safety is Brett's number one concern for getting robots into homes at scale. There are multiple layers. Semantic safety: understanding that if there's a candle lit and you knock it over, or a boiling pot of water, those are dangerous. Intrinsic safety: can the robot physically be safe around humans, animals, and pets? They believe they can build intrinsically safe robots, potentially safer than humans by a large margin, just like autonomous cars are safer than human drivers.

On privacy: these robots will be in your home with cameras everywhere, seeing everything. Brett says being upfront about what data they collect, where it goes, and how it's encrypted is super important. They have an entire team on cyber security, both on the product side and corporate side.

On Asimov's Laws: Brett has thought about this extensively. They want to put fundamental safety rules into the non-volatile memory at the chip level of the robot. He wouldn't share their specific version publicly, but acknowledged Asimov got a lot right with the three foundational laws.

Brett also compared their approach to Apple's philosophy: build the best out-of-box experience, don't make mistakes, build that reputation. If Figure gets the safety reputation right, people will choose Figure robots the same way people trust the Apple brand.

He explicitly stated they will NOT license out their neural net to other hardware manufacturers. Quote: "We have a fiduciary duty to our civilisation to build really safe humanoid robots at scale. Just giving this AI system or even hardware to anybody that would want it is not something we will entertain." Safety critical systems require vertical integration.

Section 6. Scale, Economics, and the Trillion Dollar Opportunity.

The numbers Brett discussed are staggering. Human labour represents roughly half of global GDP, about $50 trillion. Every human should have a humanoid robot, plus 5 to 10 billion in the commercial workforce. That's tens of billions of humanoids on the planet.

Current pricing trajectory: robots will get down to $10,000 to $20,000 each. At $20,000, you'd lease one for about $300 a month, $10 a day, 40 cents an hour. At that price, the question isn't whether you'd get one, it's how many.

Brett compared it to cell phones rather than cars: "We make a billion or more cell phones a year." He even raised an interesting question: if your robot breaks, do you want a brand new refurbished one, or do you want YOUR old robot back because you've known it, it has a personality, it has scratches you recognise? It's going to be personal.

The manufacturing challenge is immense. Even if you have a billion in demand, you need massive working capital. A billion robots at $20,000 each is $20 trillion in working capital. But Brett pointed out that financing markets for this exist: credit card receivables and car leasing are already trillion-dollar markets.

Their current facility can support about 4 lines, each doing 12,000 units per year, just under 50,000 total. But the real plan is to have robots building other robots. Brett hopes that within 24 months, all the robots will build all the robots. The key is solving two things: general purpose neural nets that transfer across tasks, and robots in the loop building other robots.

Dave raised a brilliant point about how this could transform developing economies. You could literally ship a box to Kenya that opens up and starts building a Figure manufacturing plant out of thin air. All the IP, the neural net, stays centralized. But the local manufacturing capacity could 100x the GDP of that region. There's latent capital all over the world waiting to build something productive.

Section 7. The Competitive Landscape.

Brett was surprisingly candid about competition. He sees China as Figure's main competitive threat, not any Western company. He praised China's explosion of talent and entrepreneurial work ethic. However, he noted a critical gap: "We've not seen any closed-loop AI control from these systems at all. We've seen a huge lack thereof. They're basically doing open loop, hand-controlled operations."

On whether other tech giants will enter: Brett has been in conversations with every major tech company in the world in the last 12 months. Apple is rumoured to be pivoting from their cancelled car project to humanoids. Meta, Amazon, Nvidia, Google, even Sam Altman are all making noises.

But Brett thinks it's going to be incredibly hard for newcomers. He compared the difficulty to rocket design: "If you know Meta is building rockets, you'd think that'd be crazy." Humanoid robotics is probably harder than the electric aircraft he built at Archer, which itself involved 6,000-pound aircraft with 12 motors, six independent battery systems, and custom control stacks.

On using off-the-shelf hardware from companies like Unitree: "It doesn't really work at scale. It works for hobby-grade demonstrations, but if you really want to do robotics at scale, you're going to have to go design yourself."

His competitive moat is clear: "The work we showed today and the work we showed two years ago has never been done in our mind by any other human or company in history. If that's the marker, whoever can do the Keurig test for a couple minutes with uncut film, closed loop, that's two years away from where we're at."

Section 8. The Figure 3 Hardware Tour.

Peter got an up-close tour of all three Figure generations.

Figure 1: Designed in-house, prioritised function over form. They incorporated the company and walked the robot in under one year, believed to be one of the fastest times in history. All aluminium, CNC machined, about 130 to 140 pounds. Not pretty, but it served its purpose, unlocking AI and control scheme development. They did the famous Keurig coffee cup demo with this robot and their first neural network deployment.

Figure 2: Massive cost reduction, about 90% cheaper than Figure 1. Better looking, David the design lead made it "much more presentable." Double or triple the compute, double the battery capacity, second-generation actuators, third-generation hands. All wires moved internal. Exoskeleton structure like an aircraft where the outer shell takes all the loads. Added knee padding and safety features so if you got your finger stuck, it would hurt but wouldn't cause serious injury. About 150 pounds.

Figure 3: The current production model and a major leap. Key specs: about 135 pounds, lighter and skinnier but kept all the same speeds and torques. So just as powerful and fast, but in a smaller package. 20 kilogram carrying capacity in each hand. Completely redesigned hands with gloves, tactile sensors, compliant material for better grasping, and palm cameras. Very few pinch points anywhere on the robot. Most of the body is soft-wrapped with squishiness for safety. New feet with a passive toe that helps with walking and getting down on the ground. Massively reduced cost from Figure 2.

The head houses three screens, a main screen and two side screens, along with cameras and sensors. The brain, meaning all the major computation, is in the head where heat can be vented easily. Brett joked you could put a silicone face and hair on it if you wanted. They also have backward-facing cameras, cameras in the torso pointing down to see feet and objects below, and different outfit configurations for different work environments, including cut-resistant jackets for industrial use.

The robots still have 3 to 5 times more speed headroom in their actuators than what's currently being used. The software just hasn't caught up yet. Brett also mentioned they just brought online a cluster of 3,000 Nvidia B200 GPUs for pre-training, with an even larger GPU deployment planned.

Section 9. Surgery, Healthcare, and the Personal Robot.

Brett made a bold prediction: from a hardware perspective, by end of 2026, Figure robots will be capable of performing surgery. The hardware, the physical dexterity and precision, will be there. You could tele-operate it to perform real surgery. Then it's just a matter of getting the AI brain good enough to do it autonomously.

On the healthcare vision more broadly: Peter described a future where you're constantly monitored for blood biochemistries, protein levels, vitamin levels, and that data is uploaded to your Figure robot in the kitchen, which cooks meals ideally suited for what your body needs in that moment.

Brett shared that he went through Fountain Life's comprehensive health screening, spending 5 hours getting 200 gigabytes of data about his body. He was so impressed he bought the same package for his parents. The intersection of comprehensive health monitoring and home robots cooking personalised meals is the future.

Section 10. Mining, Space, and the Bigger Vision.

Peter raised the topic of mining Greenland, which was a hot topic at Davos. Europeans said it's impossible to extract minerals from frozen tundra. Brett's response was essentially: give a million robots the task and they'll figure it out. The demand for $50 trillion worth of humanoid robots will create the drive to find engineering solutions for resource extraction.

Alex Weissner-Gross, who joined for the ending, pushed even further: we'll disassemble the moon and the asteroid belt for materials. Brett agreed 100 percent. They also discussed operating in zero gravity. Brett said Figure robots could operate in space with a better training set and that they'd love to run in space. Someone needs to assemble those orbital data centers.

Brett's parting message to the audience was simple and urgent: "General purpose robots are coming. It feels very close. I don't think people have a clue how fast this transition is going to be. At some point, probably in San Francisco first, you'll walk out and see more humanoids than humans. And I think that'll be an amazing day."