Is Robotics Having Its ChatGPT Moment? The Capital Says Yes, the Data Says Not Yet

On October 22, 2026, Anthony Pompliano's Pomp Letter asked whether robotics is living through its own version of November 2022, the month OpenAI released ChatGPT and the AI trade began. The letter's answer is that it is not too late to get exposure, but the best opportunities will take some creativity. That framing matters. Almost four years after ChatGPT reached an estimated 100 million users in about two months, investors are looking for the next platform shift. Many have decided it will have arms, legs, and a battery pack. The stakes are large. If robotics follows the AI curve, the winners will be set in the next 24 to 36 months. If it does not, a lot of capital is about to learn that atoms are harder than bits.

The ChatGPT Comparison

The comparison is useful because it is specific. ChatGPT did not invent large language models. The transformer architecture dates to a 2017 Google paper. GPT-3 shipped in 2020. What changed in late 2022 was distribution. A free chat window put a capable model in front of ordinary people, and the public reaction pulled capital, talent, and compute toward the sector almost overnight. Nvidia's data center revenue went from roughly $15 billion in fiscal 2023 to more than $115 billion in fiscal 2025. That is what a moment looks like in the numbers.

Robotics has its own long prehistory. Unimate, the first industrial robot arm, went to work at a General Motors plant in 1961. The International Federation of Robotics counts more than 4 million industrial robots operating in factories worldwide, with annual installations above 500,000 units in recent years and China taking roughly half of them. None of that is new. The new element is the idea of a general-purpose robot. That means a machine that learns tasks from data, not from hand-written code, and that can move from a warehouse to a kitchen without a team of integrators rebuilding it.

Nvidia chief executive Jensen Huang made the comparison explicit at CES in January 2025, when he said the ChatGPT moment for general robotics was just around the corner. Nvidia has backed that claim with product. Its Isaac simulation stack, its GR00T foundation models for humanoids, and its Jetson Thor onboard computer are built to sell picks and shovels to every robot company at once. When the largest company in the AI supply chain tells you where it thinks demand is going next, it is worth taking seriously. It is also worth remembering that Nvidia sells to whoever shows up.

Where the Money Has Gone

Capital has already moved. Figure AI raised $675 million in February 2024 at a $2.6 billion valuation, with Microsoft, OpenAI, Nvidia, and Jeff Bezos among the backers. In September 2025 the company announced a round of more than $1 billion at a reported $39 billion post-money valuation. That is a 15x markup in about 19 months for a company with limited commercial revenue. Physical Intelligence, a San Francisco startup building general robot brains instead of robot bodies, raised $400 million in late 2024 and then a larger round in 2025 at a multibillion-dollar valuation. Skild AI, Apptronik, and Agility Robotics have all raised nine-figure rounds.

On the public side, Tesla has told shareholders that Optimus could eventually be worth more than its car business. Elon Musk has set and moved production targets for the humanoid several times. Tesla's 2025 compensation package for Musk tied part of the payout to delivering 1 million robots. Hyundai owns Boston Dynamics, which retired its hydraulic Atlas in 2024 and replaced it with an electric version aimed at Hyundai's own plants. In China, Unitree sells its G1 humanoid starting around $16,000. The company has pursued a listing on Shanghai's STAR Market. UBTech is already public in Hong Kong. Beijing has named embodied intelligence as a strategic priority and backed the sector with state funds and local subsidies.

Consumer products are arriving too. Norway's 1X opened preorders in October 2025 for Neo, a home humanoid priced at $20,000 or $499 a month. The company was candid that early units would rely on remote human operators for difficult tasks. That candor is a data point. The robot in the living room is, for now, partly a person in a control room.

Banks have published large forecasts to match. Goldman Sachs projected a $38 billion humanoid market by 2035. Morgan Stanley has floated multi-trillion-dollar figures for 2050. Treat those as directional. Nobody has a reliable model for a market that does not yet exist.

The Bull Case

The optimists have real arguments, and they are not only about hype.

First, the cost curve is falling. Actuators, sensors, and batteries ride the same manufacturing scale that electric vehicles and phones built. A humanoid that cost several hundred thousand dollars to build a few years ago can now be produced for a fraction of that, and Chinese suppliers are pushing component prices down fast. When Unitree prices a humanoid like a used car, the experimentation budget for every university lab and small factory changes.

Second, the software approach has changed. Robots used to be programmed step by step. Now companies train vision-language-action models that take in camera feeds and instructions and output motor commands. Google DeepMind's RT-2 and Gemini Robotics work, Physical Intelligence's pi-zero models, and Nvidia's GR00T all point the same way. Models trained on diverse data generalize better than code written for one task. That is the same lesson the language model labs learned.

Third, demographics create demand. Japan, South Korea, China, and much of Europe have shrinking working-age populations. US manufacturers report hundreds of thousands of unfilled jobs. Warehouse operators like Amazon, which has deployed more than 1 million robots across its fulfillment network, have shown that automation pays when labor is scarce and volumes are high. A general-purpose machine that fits into human workspaces without retooling the building attacks the biggest cost in automation, which is integration.

The bull case on investing follows from that. Pomp's point about creativity is fair. The best-known private names, Figure and Physical Intelligence, are closed to most investors. The public pure plays are thin. So exposure has to come sideways. That means through component makers, chip suppliers, motion-control firms like Harmonic Drive and Nabtesco, sensor companies, and the large incumbents with robotics divisions. That is how many retail investors captured the AI trade: through Nvidia, not OpenAI.

The Bear Case

The skeptics have arguments that are just as concrete.

The biggest is data. Language models trained on the internet, trillions of tokens of text that humans had already written. Robots have no equivalent archive. Every hour of useful training data for a robot hand folding laundry has to be collected, usually through teleoperation, simulation, or expensive fleets. Simulation helps, but the gap between simulated physics and a real kitchen remains wide. This is the core reason the ChatGPT comparison may break. ChatGPT was a release of capability that already existed. Robotics still has to build the dataset.

The second is reliability. A chatbot that is right 90 percent of the time is useful. A robot that drops one in ten dishes is a liability. Physical errors break things and injure people. That raises the bar for deployment far above what software products face, and it brings in regulators, insurers, and safety standards bodies. The EU's Machinery Regulation, which applies from January 2027, and its AI Act both add compliance work for autonomous machines. In the US, the Occupational Safety and Health Administration has no humanoid-specific rules, but liability law will fill that gap through the courts.

The third is unit economics. A warehouse worker in the US costs roughly $40,000 to $60,000 a year fully loaded. A humanoid priced at $20,000 to $100,000 sounds cheap until you add maintenance, downtime, battery replacement, software subscriptions, and the human supervisors who still step in. Pilot programs at automakers and logistics firms have been announced with fanfare. Few have published throughput figures that prove the robot beats either a person or a purpose-built machine. A fixed robotic arm that does one thing perfectly often wins against a humanoid that does many things adequately.

The fourth is valuation. A $39 billion price on a pre-revenue hardware company assumes a near-perfect path. Hardware does not scale like software. Every additional robot requires metal, motors, and a supply chain. Margins that look like Apple's are a hope, not a plan. The 2021 cycle in electric vehicle SPACs is a recent reminder of what happens when capital prices in a decade of execution before the first year has gone well.

There is also a geopolitical angle that cuts against the US bull case. China installs more industrial robots than the rest of the world combined, controls much of the supply chain for rare earth magnets used in motors, and has state backing for its champions. American investors betting on a US-led robotics boom should ask whose factories will actually build the robots.

Capital, Money, and Who Captures the Gains

There is a monetary story under the robotics story, and it deserves a direct answer. Robots are a deflationary technology. They lower the cost of producing goods and services. In a sound money system, that productivity would show up as falling prices, and savers would share in the gains simply by holding money. In a fiat system run by central banks targeting 2 percent inflation, the Federal Reserve and its peers expand the money supply to offset those falling prices. The gains flow first to those closest to new money and to those who own the productive assets. Wage earners who hold cash fall behind. That is why every new technology boom now feels like a race to buy assets before they reprice. People are not only chasing returns. They are running from the currency. Accredited investor rules then lock most of them out of the private rounds where the largest gains happen. Bitcoin is the cleanest exit from that trap. A fixed supply of 21 million coins means productivity gains show up as more purchasing power for the holder, not as a subsidy to whoever borrows newly created money first. If robots make the world radically cheaper to run, the asset that cannot be diluted should capture a large share of that abundance. We think it will.

There is a practical link too. Machines that work on their own will need to pay for things. Charging, data, compute, parts, and services between devices all need settlement. A robot cannot open a bank account, pass a know-your-customer check, or wait three business days for a wire. Bitcoin over the Lightning Network already supports small, instant, permissionless payments between software agents. Several developers are building exactly that for AI agents today. Whether robots settle in bitcoin or in stablecoins issued by regulated firms is an open fight. The answer will shape who controls the machine economy.

What to Watch

The next 12 to 18 months will separate the ChatGPT comparison from a more expensive version of the 2010s self-driving hype cycle. Here is what will decide it.

Published deployment metrics. Expect at least one major humanoid company, most likely Figure, Agility, or Tesla, to publish hours worked, task success rates, or cost per unit of output from a paying customer by late 2027. If those numbers stay vague, assume the economics are not there yet.

Tesla's Optimus schedule. Musk has promised meaningful production several times. Watch Tesla's quarterly filings, not keynote demos. A credible external customer outside Tesla's own factories would be the real signal. We expect continued delays and a revised timeline in 2027 guidance.

Chinese price pressure. Unitree and its peers will keep driving humanoid prices down. A sub-$10,000 capable humanoid from a Chinese maker within two years is likely. That will compress margins for Western competitors and force a debate in Washington about tariffs and export controls on robot components, similar to the fight over drones and DJI.

Data partnerships. The companies that win will be the ones that secure large real-world training data. Watch for deals between robot developers and logistics, retail, and manufacturing firms that trade deployment access for data. Those contracts matter more than funding rounds.

A down round or failure. At current valuations, at least one well-funded humanoid startup will raise at a lower price or be absorbed by a larger company before the end of 2027. That will not end the sector. It will reset expectations, and it will create the entry points that patient investors want.

Public market access. More robotics listings, including Chinese firms on the STAR Market and Hong Kong exchange, and possibly a US IPO from a well-known private name, will give ordinary investors direct exposure. Expect the SEC's accredited investor rules to stay mostly unchanged, which keeps the earliest and largest gains in private hands.

Machine payments. Watch for the first commercial robotics or AI agent platform that settles device-to-device payments natively over Lightning or a stablecoin rail. The choice will tell you whether the machine economy runs on open money or permissioned money.

Our view: robotics is not having a ChatGPT moment today. It is having a GPT-3 moment. The capability is real, the capital is in, and the public breakthrough is still ahead. That means the creative bets Pomp describes are worth making, with a hard eye on unit economics and a clear view that the money you invest with matters as much as the companies you invest in.

---

Go deeper: The Lightning Network

Source: Pomp Letter