Co-founded two years ago by OpenAI's former chief scientist, the lab says the partnership will multiply its computing power tenfold within a year.
Nvidia has bought equity in Safe Superintelligence, Ilya Sutskever's lab, as part of a long-term strategic partnership the two companies announced on Monday. Neither side put a number on the investment, describing its size only as substantial.
One person briefed on the arrangement puts the equity investment at $5 billion, Reuters reported, and two briefed people said the agreement had been pulled together in the space of a few weeks.
The substance of the deal is compute. SSI gets access to Vera Rubin, the coming generation of Nvidia's GPU computing platform, and the lab said on X that Nvidia's money would let it increase its compute tenfold over the next 12 months.
Nvidia says it was allowed a rare look at SSI's closely guarded research before deciding to invest. The lab, Sutskever says, has reached work that merits a far larger machine, and the partnership gives it one.
That due-diligence look is the only read on SSI's work from outside the company. Sutskever founded it with Daniel Gross, previously head of AI at Apple, and Daniel Levy, previously of OpenAI. Where the rest of the industry showcases benchmarks, papers and launches, SSI has said almost nothing, asking backers largely to place their faith in Sutskever himself.
PitchBook puts its fundraising at roughly $7 billion on a $32 billion post-money valuation. Nvidia was already an investor before this agreement, as are Alphabet, Andreessen Horowitz and Sequoia.
The compute arrives against Sutskever's own repeated argument that feeding models more data and running them on bigger machines is running out of road, and that further progress rests on fundamentally new science. In a rare interview last year he said the era of scaling was nearing its end. Today's large language models, he has argued, post strong benchmark scores yet generalize poorly and prove brittle on real tasks, and SSI is hunting a fresh principle of machine learning rather than ever-larger models.
The investment also lands in a week of renewed scrutiny of Nvidia's practice of buying equity in companies that are also its customers: $6.6 billion into OpenAI in October, Elon Musk's xAI in November, Anthropic as well. Critics argue that a chipmaker financing its own customers' purchases makes real appetite hard to separate from demand engineered by the seller; Nvidia's founder and chief executive, Jensen Huang, has repeatedly rejected the circular-financing charge, calling the stakes good investments.
The SSI deal is structured differently. SSI has no product earning revenue, and no cloud commitment in the tens of billions has been disclosed in return, so what Nvidia is buying is a share of a wager, plus a tie to a researcher who helped lay the foundations of modern AI.
Whether the wager pays cannot be judged until SSI puts something on display for the first time. No date has been set for that.