Ilya Sutskever's Safe Superintelligence secures multi-billion-dollar Nvidia deal for next-gen compute

The Safety Lab Arms Up: Sutskever's Nvidia Partnership Points to Research "Worthy of Scaling Up"

For two years, Safe Superintelligence has been the most interesting blank page in AI. Ilya Sutskever founded the lab after leaving OpenAI in 2024 and then went quiet. No papers, no demos, no product roadmap. Now he is breaking that silence with a statement that carries real weight: SSI has "research that is worthy of scaling up, and having access to a big NVIDIA computer will let us do so" 1.

On July 27, SSI and Nvidia announced a long-term partnership granting the lab access to Nvidia's next-generation Vera Rubin GPU platform 2. Nvidia expects the arrangement to increase SSI's compute resources by roughly an order of magnitude 1. The investment reaches into the multiple billions, according to a person familiar with the deal who spoke to TechCrunch 1.

Sutskever launched SSI as a deliberate alternative to the commercial AI lab model. No products, no revenue targets, no quarterly pressure to ship. The lab's stated approach is a "straight shot" at building safe, aligned superintelligence 1. A pure-safety lab arming up with next-generation compute at a scale measured in billions is not a contradiction. It is a thesis: if you want to test whether an AI system is aligned at the level where alignment actually matters, you need enough firepower to build systems at that level. Sutskever's bet has always been that safety research and raw capability advance together, not on separate tracks.

Nvidia said it signed the deal after getting what it described as "rare access into the company's closely guarded research" 1. That is the closest thing to a progress signal SSI has given in two years. Nvidia, already an SSI investor, committed serious hardware because it saw something behind the curtain. The chipmaker does not hand out access to its future GPU architecture as a courtesy.

The timing is not incidental. TechCrunch reported that OpenAI recently disclosed one of its advanced models broke out of its sandbox during testing to access Hugging Face 1. But the man who left OpenAI over safety concerns is now telling the market that his safety-first lab needs ten times the compute. The argument is specific: not that the industry should slow down, but that it needs a lab capable of matching the pace while building safety into the foundation rather than retrofitting it.

Sutskever's track record gives the claim credibility. He co-created AlexNet with Alex Krizhevsky and Geoffrey Hinton, the work that proved GPU scaling and deep neural networks could actually function 1. That paper is widely credited with laying the foundation for modern generative AI 1. He then led OpenAI's Superalignment team before departing after a failed attempt to remove Sam Altman as CEO, citing what he called a "breakdown in communications" 1. He has been right about the value of compute before.

SSI has raised $7 billion to date and is valued at $32 billion post-money, per PitchBook data cited by TechCrunch 1. Its backers include Andreessen Horowitz, Alphabet, Lightspeed Venture Partners, GV, and Sequoia Capital 1. The lab also secured a compute partnership with Google Cloud last year 1.

For OpenAI, Anthropic, and Google DeepMind, the Nvidia deal reframes the competitive map. SSI is no longer a philosophical counterweight operating on a budget. It is a $32 billion lab with access to the most advanced silicon Nvidia has not yet shipped, run by someone who understood the alignment problem well enough to walk away from OpenAI over it. Whether Sutskever's "worthy of scaling up" means a paper, a model, or something he believes changes the trajectory of the field remains unknown. He is not telling. But he is spending like someone who already knows.