The ChatGPT Moment for Robotics

For a while now, Silicon Valley has claimed that “the ChatGPT moment is here” for robotics, or at least on its way. But when will it come?

In January of 2026, Jensen Huang, founder and CEO of NVIDIA, a chip manufacturing company that powers massive amounts of artificial intelligence infrastructure, claimed in their press release that “the ChatGPT moment for robotics is here. Breakthroughs in physical AI — models that understand the real world, reason, and plan actions — are unlocking entirely new applications” (NVIDIA, 2026). During the keynote speech itself, Huang’s wording was more restrained, saying it was “nearly here” instead (Goldman, 2026). However, in January of last year, NVIDIA’s press release quoted him saying “the ChatGPT moment for robotics is coming” (NVIDIA, 2025). Yet after all this time, it is still unclear if we are heading towards a point at which physical AI technology will grow its capabilities and deployment exponentially, akin to how ChatGPT has developed since its release. The phrase moved from “is coming” to “nearly here” to “here” in twelve months, and managed to move faster than the robots did.

Money has followed the phrase. According to Crunchbase, “in the first half of 2026, global venture funding in the space totaled $47.4B across 521 deals…up significantly — by nearly 80% — from the $26.4B raised across 436 deals in the first half of 2025.” For reference, that is more than investors put into the sector across 2022, 2023, and 2024 combined (Azevedo, 2026). Moreover, that funding is extremely consolidated among top companies. Waymo’s $16B Series D at a valuation of $126B accounted for a third of the capital being invested into the physical AI space during the first half of the year (Waymo, 2026). Public markets have also shown that the sentiment can turn incredibly quickly. Unitree, China’s leading humanoid robotics company, closed its Shanghai debut 460% above its IPO price (Bao et al., 2026). Within less than two weeks, the company had lost nearly half its value (Tong, 2026). The issue is that while the robots’ physical abilities kept improving, they still lacked the knowledge to do work that could create value.

Revisiting November of 2022, ChatGPT was not a scientific breakthrough (a chat interface and some instruction-tuning placed on top of models which OpenAI had been scaling since 2020) but rather a distribution breakthrough: ChatGPT hit an estimated 100 million monthly users two months after launch (Milmo, 2023). It was free, it ran on devices people already owned, and its mistakes were cheap because a person could always verify the output before acting on it. None of that holds for robots.

There are three main reasons. The first is data. UC Berkeley’s Ken Goldberg estimates the internet-scale data behind today’s vision-language models amounts to roughly 100,000 years’ worth, and robots have nothing comparable (Goldberg, 2025). The second is the body. The human hand has about 17,000 specialized tough receptors, which no robot comes close to matching (Johansson & Vallbo, 1979). Videos of people working capture motion, not touch. The third is reliability. A chatbot with an 80% accuracy rate can be a useful product; a robot that drops one plate in five is a liability.

Nevertheless, skepticism about a sudden moment should not translate into skepticism about progress, because the progress is undoubtedly real.

Physical Intelligence, a physical AI research lab and company founded by Stanford professor Chelsea Finn, released π0.7 in April, which can fold laundry on a robot that had never seen the task and run an express machine about as well as specialized models trained for that job. It still has limits: success on unfamiliar tasks runs around 60-80%, compared with over 90% on tasks it was trained on (Physical Intelligence, 2026). While short of what a commercial customer would look for, that sort of generalization barely existed a year ago.

The data problem is also starting to give way. Chatbots learned from decades of text already sitting on the internet, but there is no equivalent archive showing a robot how to fold a shirt or load a dishwasher, so that archive is now being built by hand. Companies are paying people to wear head-mounted cameras while they do everyday chores, and to remotely guide robots through tasks so that each movement can be recorded and used for training. Tesla, for one, has shifted its humanoid robot program toward workers wearing camera-equipped helmets (Kay, 2025). As recently as last year, the major open robotics datasets combined held only about 5,000 hours of interaction data (Levin, 2025). Scale AI, the data-labeling company that built its business supplying self-driving programs, now reports that more than 1,000 hours of robot demonstration data are uploaded to its platform each day (Choghari et al., 2026).

Deployments are compounding as well. By March, Waymo was delivering 500,000 paid rides a week across 10 U.S. cities, a greater than tenfold increase in less than two years (Korosec, 2026). What truly stands out is how unglamorous the expansion has been. Each new city starts with only dozens of cars. Waymo maps the streets, tests with safety drivers, removes them, opens rides to a small group, and only then widens access. Even this pace may fall short of the company’s goals: one independent forecast expects Waymo to end the year closer to 775,000 weekly rides than a million (Schwarz, 2026). That is not a ChatGPT-style explosion. It is a technology proving itself one city at a time, more than five years after Waymo first put riders with no one behind.

So what can we expect? The arrival of this moment everybody is waiting for will likely appear more as one vertical and one site at a time, gated by reliability rather than demos. Robots that are capable of doing narrow jobs well will get out first, and the data they collect will train the general models that follow. It will likely also look incredibly boring. The world’s largest deployments right now are still autonomous surveyors, heavy material transporters, and consumer robotic vacuums. The robots that eventually change the economy will be the ones that we stop noticing.

The risk in framing this evolution as a “ChatGPT moment” is not even that it’s too optimistic about the technology, but more so that it promises a sudden consumer event when the evidence is clearly pointing towards a steady industrial buildout. Investors who price in the first will be disappointed by the second, even though the second is remarkable in its own right.

Works Cited

Azevedo, M. A. (2026, August 18). VCs pour billions into physical AI as the next wave of AI investing takes shape. Crunchbase News. https://news.crunchbase.com/venture/physical-ai-funding-startups-robotics-aerospace-h1-2026/

Bao, A., Reid, J., & Lee, J. (2026, August 20). Humanoid robots' 'ChatGPT moment' could be 10 years away, Unitree founder says. CNBC. https://www.cnbc.com/2026/08/20/unitree-humanoid-robots-chatgpt-moment.html

Choghari, J., Sansone, A., Pasqualis, N., Mader, C., Tiupikov, A., & Sivapurapu, M. (2026, May 19). The path to large scale dense video captioning. Scale AI. https://labs.scale.com/blog/path-to-large-scale-dense-video-captioning

Goldberg, K. (2025). Good old-fashioned engineering can close the 100,000-year "data gap" in robotics. Science Robotics, 10(105). https://doi.org/10.1126/scirobotics.aea7390

Goldman, S. (2026, January 6). A year ago, Nvidia's Jensen Huang said the 'ChatGPT moment' for robotics was around the corner. Now he says it's 'nearly here.' But is it? Fortune. https://fortune.com/2026/01/06/nvidia-jensen-huang-chatgpt-moment-for-robotics/

Johansson, R. S., & Vallbo, A. B. (1979). Tactile sensibility in the human hand: Relative and absolute densities of four types of mechanoreceptive units in glabrous skin. The Journal of Physiology, 286(1), 283–300. https://doi.org/10.1113/jphysiol.1979.sp012619

Kay, G. (2025, August 25). Inside the strategy shift at Optimus, Tesla's humanoid robot program. Business Insider. https://www.businessinsider.com/tesla-musk-optimus-humanoid-robot-training-motion-capture-cameras-2025-8

Korosec, K. (2026, March 27). Waymo's skyrocketing ridership in one chart. TechCrunch. https://techcrunch.com/2026/03/27/waymo-skyrocketing-ridership-in-one-chart/

Levin, B. (2025, September 24). Expanding our data engine for physical AI. Scale AI. https://scale.com/blog/physical-ai

Milmo, D. (2023, February 2). ChatGPT reaches 100 million users two months after launch. The Guardian. https://www.theguardian.com/technology/2023/feb/02/chatgpt-100-million-users-open-ai-fastest-growing-app

NVIDIA. (2025). NVIDIA launches Cosmos world foundation model platform to accelerate physical AI development [Press release]. https://nvidianews.nvidia.com/news/nvidia-launches-cosmos-world-foundation-model-platform-to-accelerate-physical-ai-development

NVIDIA. (2026). NVIDIA announces Alpamayo family of open-source AI models and tools to accelerate safe, reasoning-based autonomous vehicle development [Press release]. https://nvidianews.nvidia.com/news/alpamayo-autonomous-vehicle-development

Physical Intelligence. (2026, April 16). π0.7: A steerable model with emergent capabilities. https://www.pi.website/blog/pi07

Schwarz, D. (2026, August 20). Waymo profitability forecast: Rides, margins, and losses through 2027. FutureSearch. https://futuresearch.ai/waymo-financial-forecast/

Tong, M. (2026). Unitree sheds over 200b yuan in first trading week as analysts see rational repricing, upbeat in long-term prospects. Global Times. https://www.globaltimes.cn/page/202608/1369106.shtml

Waymo. (2026). Accelerating our global growth: Waymo raises $16 billion investment round. https://waymo.com/blog/2026/02/waymo-raises-usd16-billion-investment-round/