OpenAI's Maths Push Costs Hundreds of Millions, Altman Says
Sam Altman says OpenAI spent about $10 million on one Navier-Stokes attempt and has more proofs in reserve. What that spending gap means for Indian research.
Key points
- OpenAI chief executive Sam Altman said the company will keep releasing machine-generated maths results despite complaints from mathematicians.
- Altman put OpenAI's maths spending in the hundreds of millions of dollars, including around $10 million on its claimed Navier-Stokes solution.
- The gap with Indian research budgets is wide, but verifying machine proofs could become skilled work for India's mathematicians.
Photo: Christopher Bowns / Wikimedia Commons, CC BY-SA 2.0 (cropped). Representational image.
The News
OpenAI chief executive Sam Altman has said the company has spent in the order of hundreds of millions of dollars on mathematical research, including around $10 million, "maybe a little more", on its claimed Navier-Stokes solution. Navier-Stokes is one of the Millennium Prize problems, the set of famous open questions in mathematics.
Speaking on the Cheeky Pint podcast, in an episode released on Thursday, Altman said he has empathy for mathematicians who are struggling to absorb the volume of results the company has put out. He was clear, however, that OpenAI will not slow down. Asked whether it holds more than the hundreds of proofs already released, he answered: "We do. Of course."
Altman also admitted the push "did get out of control" for a short spell, describing an internal Slack channel where staff threw the hardest problems they could find at the company's models. He argued that the public's claim on new scientific knowledge outweighs the disruption, while conceding that OpenAI must explain its findings better.
Why It Matters
The remarks land amid sharp criticism from researchers. Mathematicians quoted by The Verge say OpenAI's releases often arrive with thin write-ups, weak attribution and little verification, leaving academics to do the checking. At the same time, the company is being accused of holding back detail on a claimed result involving Yang-Mills theory, which underpins modern particle physics, on the advice of an expert group it selected itself.
The episode recalls 1976, when Kenneth Appel and Wolfgang Haken used a computer to prove the four colour theorem and many mathematicians refused to accept a proof no human could check by hand. The difference now is scale and money. A single problem attracting roughly $10 million in compute and staff time is a budget few university departments anywhere could match.
Indian Angle
India's research system works with far smaller cheques. The Anusandhan National Research Foundation, set up under a 2023 law, was designed around a Rs 50,000 crore outlay over five years for all of Indian science. At prevailing exchange rates, OpenAI's spend on one Navier-Stokes attempt would be well over Rs 80 crore, a sizeable sum against any single Indian institute's research budget. The worry that academics could be priced out of their own fields is therefore sharper for groups at places such as TIFR, the Indian Statistical Institute and IMSc Chennai.
There is also an opening. India produces a large pool of trained mathematicians, and checking and formalising machine-generated proofs is skilled, labour-intensive work. Indian researchers could build a role as verifiers and collaborators rather than spectators, provided they have access to the models and the compute.
That links directly to the IndiaAI Mission, approved in March 2024 with an outlay of Rs 10,371.92 crore, much of it aimed at shared GPU capacity. Domestic model builders such as Sarvam AI have so far focused on Indian-language products. Altman's comments suggest that reasoning and scientific work is where the global frontier labs are now competing hardest, and Indian policymakers may need to decide whether public compute should support that kind of research too.
FAQ
What did Sam Altman say about OpenAI's maths work?
On the Cheeky Pint podcast, Altman said he sympathises with mathematicians overwhelmed by OpenAI's output but that the company will keep releasing results. He said OpenAI holds more discoveries than the hundreds of proofs already published, and that its spending on maths runs to hundreds of millions of dollars.
How much did OpenAI spend on Navier-Stokes?
Altman estimated around $10 million, "maybe a little more", on the company's claimed Navier-Stokes solution. Navier-Stokes is one of the Millennium Prize problems. Any formal recognition would rest with the Clay Mathematics Institute, which administers those prizes.
Why are mathematicians unhappy?
Researchers say OpenAI's announcements often come with brief, hard-to-follow write-ups, limited attribution and little verification, leaving academics to do the checking. Some are also frustrated that full details of a claimed Yang-Mills result have not been made public.
What does this mean for Indian researchers?
The spending gap between frontier labs and Indian institutes is large, which raises fears of being outpaced. But verifying and formalising machine-generated proofs needs trained mathematicians, a field where India has depth, if researchers can secure model and compute access through programmes such as the IndiaAI Mission.
This story was reported by The Verge. Read the full original coverage at The Verge.