Nvidia has agreed to lob $5bn at Safe Superintelligence, the hush-hush AI outfit run by OpenAI co-founder Ilya Sutskever.
According to the Financial Times Nvidia and SSI announced the deal as part of a “long-term partnership” which will see the start-up use Nvidia’s latest Vera Rubin hardware.
The $4.75tn chipmaker has agreed to invest about $5bn. Big cheques like this are usually tied to a start-up hitting certain milestones, and the pair declined to discuss the numbers, saying only that the investment was “substantial”.
The cash should help SSI build the computing muscle needed to exploit what it claims is a research breakthrough.
“We have research that is worthy of scaling up, and having access to a big Nvidia computer will let us do so,” Sutskever said.
SSI said the investment would let it increase its available computing power tenfold during the next 12 months.
The deal is the latest in a run of circular financing arrangements between AI chipmakers and model developers. Last week, Nvidia rival AMD said it would invest up to $5bn in Anthropic.
SSI, which still has only a few dozen employees, has been working on a new approach to AI research. It breaks with the method behind large language models developed by OpenAI, Google and Anthropic.
The start-up was valued at $32bn last year when it raised $2bn from investors including Greenoaks, Lightspeed Venture Partners and Andreessen Horowitz. Nvidia is already an investor in the company.
Despite its towering valuation and shiny list of clever people, hardly anyone knows what SSI is actually building. It has yet to release a product or publish any research.
Nvidia chief executive Jensen Huang said Sutskever had already “pioneered fundamental breakthroughs at the foundation of modern AI”.
“We are excited to see what new breakthroughs SSI will discover powered by our Vera Rubin platform,” Huang said, referring to Nvidia’s newest generation of AI systems.
Sutskever was among the first AI engineers to grasp the so-called “scaling law”, which showed LLMs became more capable with more data and computing power.
In November, he suggested the returns from that approach were fading. “Now that compute is big, compute is now very big, in some sense we are back to the age of research,” Sutskever said in a podcast interview.
