Nvidia Invests Billions in Sutskever's Safe Superintelligence

Two years after Ilya Sutskever left OpenAI to build what he calls a straight shot at safe superintelligence, his lab has finally announced how it plans to pay for the compute. TechCrunch reported that Safe Superintelligence has entered a long-term partnership with Nvidia, and that Nvidia's investment 'stretches into multiple billions.' The deal gives SSI access to Nvidia's Vera Rubin GPU platform, which is expected to lift the lab's compute resources by an 'order of magnitude.'

The context that makes this notable is how much money is already inside SSI. The company has raised $7 billion to date at a $32 billion post-money valuation, from a cap table that includes Andreessen Horowitz, Alphabet, Lightspeed, GV and Sequoia. It has released no commercial product. Its entire value on paper rests on the bet that Sutskever, who co-authored AlexNet and led OpenAI's Superalignment team before departing in 2024, can go directly from private research to something that meaningfully matters, without the intermediate step of selling anything.

Sutskever's own framing of the deal leans on the hardware rather than the capital. 'We have research that is worthy of scaling up, and having access to a big NVIDIA computer will let us do so,' he said. Read plainly, that says the science was ready before the machine was. For Nvidia, taking an equity position in a frontier customer whose next hardware bill will be measured in the billions is a familiar shape of deal.

The honest caveat is that almost nothing about the terms is on the record. The dollar figure is attributed to an unnamed source, the delivery timeline for Vera Rubin is not spelled out, and SSI still has not shown outside researchers what it is actually building. Take the size of the round and the 'order of magnitude' claim as reported, not as settled.

What is worth watching is whether the model of no product, frontier research only, funded at multi-billion scale, holds up. If Sutskever's lab can convert this compute into a demonstrable capability lead, the pure-research path becomes viable again, and Nvidia keeps its position as the utility the frontier labs cannot route around.