OpenAI has been predicting rapid progress for scientific research with the advent of AI, but predictions now seem to be getting a lot more concrete.
OpenAI CEO Sam Altman recently shared a striking anecdote from a meeting with a physicist who had been using one of the company’s latest internal systems — one that left the scientist convinced that AI is about to compress decades of progress in theoretical physics into just a few years. For Altman, this isn’t just an exciting data point. It’s confirmation of a belief he’s long held: that accelerating science is among the most important things AI can do.
“My most exciting meeting of last week was with a physicist who’s using one of our latest internal systems,” Altman said. “He said, ‘My mind has been completely blown. I didn’t think this would ever happen. We are going to make decades worth of theoretical physics progress in the next couple of years.'”
Altman acknowledged the skepticism such a claim invites. “Some people will say, ‘Okay, who cares? You can write a better equation.’ But I am a believer that scientific progress is one of the most — maybe the most — important thing we can do to make the world better and better.”
He went on to lay out the practical stakes: “Obviously, we can do things like develop new cures for diseases, but there’s so much more — material science, new ways to produce clean and cheap energy, abundant energy — that if we are really on the precipice of being able to use AI to dramatically accelerate science, then I think we can solve some of the biggest problems in the world.”
Altman also pointed to the institutional muscle now behind that ambition. “We now have one of the most — maybe the most — well-funded foundation in the world, and the ability to use that amount of capital with the technology to go create a bunch of new scientific understanding for the world. I hope that’ll be one of our greatest contributions ever.”
The remarks land at a moment when OpenAI is moving from vision to execution on science. The company’s OpenAI for Science initiative is already partnering with researchers across disciplines, and in February 2026, GPT-5.2 derived a new result in theoretical physics — a small but telling proof of concept. The OpenAI Foundation, restructured following the company’s recapitalization in late 2025, has committed at least $1 billion in 2026 to life sciences, AI resilience, and community programs, with a longer-term $25 billion commitment to curing diseases.
Altman’s comments also fit a pattern of increasingly specific predictions from AI leaders about scientific acceleration. He previously said that AI could compress a hundred years of science into five or ten, and outlined a roadmap where 2026 would be the year AI begins making large scientific discoveries. He’s also set internal targets to build an automated AI research intern by September 2026 and a full AI researcher by March 2028. Anthropic CEO Dario Amodei has made similarly sweeping claims, suggesting AI could deliver a century’s worth of scientific progress in five to ten years.
What’s shifting now is the texture of the claim. Altman isn’t citing benchmarks or extrapolating from scaling curves — he’s relaying a reaction from a working physicist using a system that doesn’t yet exist for the public. If that reaction is any guide, the gap between prediction and reality may be closing faster than most people expect.