OpenAI spent 'order $10 million' on Navier–Stokes proof push

OpenAI chief executive Sam Altman has told Stripe's Cheeky Pint podcast that the company spent "order $10 million" on its September claim of a proof of the Navier–Stokes problem, a sum that anchors the commercial scale of the AI developer's move into scientific research.

In a Nov. 3 interview with podcaster Raj Shamani, Altman said OpenAI's recent mathematics results brought the prospect of a scientific-discovery machine closer than he had anticipated. "The math results were quite amazing to me and it really made me think that we are closer to this milestone than I expected us to be," he said.

The remarks follow a rapid sequence of releases. OpenAI announced a claimed Navier–Stokes proof in September, released 722 AI-generated mathematics manuscripts in early October, and reported further claimed advances in mathematics, computer science and theoretical physics later that month.

What remains unverified?

The mathematical claims are still under scrutiny. OpenAI published a paper and public proof files in Lean, software for checking formal mathematical arguments. Its Navier–Stokes claim covers fluid motion subject to a smooth external force. The Clay Mathematics Institute, which administers the Millennium Prize Problems, still lists the problem as active. Its rules require publication, at least two years of scrutiny, and general acceptance by the mathematics community.

The announcement also triggered a dispute over publication arrangements and research provenance. Mathematician Tristan Buckmaster, who had been working with Levent Alpöge on related fluid-equation problems, publicly questioned whether private drafts submitted through OpenAI's Codex coding tool had influenced the research. Buckmaster said he did not know whether their data had been used. OpenAI denied accessing their work before its public release.

The October manuscripts have needed corrections. OpenAI's Oct. 7 update recorded three withdrawals and revisions to 14 manuscripts. It reported Lean formalizations covering 300 of the remaining 719 top-line results, roughly 42%. The repository records further formalization additions and fixes later in October.

For R&D managers, that correction rate matters. Formalized mathematics can be checked by software; the same workflow does not extend to experimental claims.

How did the compute bet begin?

Altman traces the computing bet to GPT-4. OpenAI finished training the model around August 2022, months before its release. Seeing its capabilities gave the company confidence to commit to far more computing capacity. "We then decided we were going to go make a very big bet on compute," Altman said. He described GPT-4 as opening research directions that led to later reasoning models.

The Navier–Stokes release itself made Altman nervous. Recalling the period before September, he told Shamani: "This is like the real deal." He called it the last occasion he remembered his heart pounding with nerves, describing AI tackling problems beyond what people had managed.

Who is sceptical?

Yann LeCun, Meta's former chief AI scientist and co-founder of Advanced Machine Intelligence, has questioned whether language models' reasoning in mathematics and code extends to understanding the physical world. Altman is more optimistic. He described the goal as "a machine that can help us understand the world better," finding insights in physics, medical treatments and energy sources.

The commercial logic is already visible. In January, OpenAI CFO Sarah Friar outlined how licensing, intellectual-property agreements and outcome-based pricing could let the company share in the value of discoveries. Rival Anthropic, maker of Claude, has begun running its own drug-discovery programs.

What about drug discovery?

Drug discovery adds laboratory work, safety testing, clinical trials and regulatory review — steps formal math workflows cannot automate. Anthropic acquired Mithril Medicine, a one-person biotech developing treatments for rare muscular dystrophies. Anthropic said Mithril moved a possible treatment from idea to promising mouse results in under 18 months, and that a drug could enter human trials as soon as late 2028 if candidates hold up. In April, Anthropic bought Coefficient Bio for roughly $400 million.

Pharmaceutical money is also flowing. Eli Lilly and NVIDIA announced in January a joint investment of up to $1 billion over five years in an AI drug-discovery research lab.

In an Oct. 21 interview with AI commentator Matthew Berman, Altman said he hoped AI would produce a candidate treatment for a major disease within the next couple of years, with testing still to follow — a projection, not a measured result, and one that Lean-style verification cannot adjudicate.