Nvidia Backs Safe Superintelligence’s Tenfold Compute Plan

Nvidia is investing in Safe Superintelligence (SSI) and has entered a long-term partnership with the lab. SSI is the AI research lab led by co-founder and chief executive Ilya Sutskever, and it gains planned access to Nvidia’s advanced Vera Rubin AI computing platform. Sutskever expects that infrastructure to expand the processing resources available for SSI’s work.

SSI’s plan calls for a tenfold capacity increase during the 12 months following the partnership announcement. Nvidia’s equity investment is about $5 billion, according to a person briefed on the deal, but Nvidia and SSI did not confirm that amount or disclose financial terms. Installed, usable hardware will provide the partnership’s first measurable result.

Nvidia is both an infrastructure supplier and financial partner, extending a financing model it has used with AI-cloud providers. SSI is pursuing safe superintelligence, a hypothetical AI capability that exceeds humans across numerous tasks, with no interim commercial release planned. Nvidia is financing a long-duration research goal and supplying infrastructure for more experiments, not hardware for a product approaching customers.

Vera Rubin Changes SSI’s Compute Path

Vera Rubin gives SSI a route to more computing capacity without turning its research plan into a finished result. SSI’s results will depend on putting delivered hardware to sustained use.

SSI previously used Google Cloud for research. Google’s tensor processing units (TPUs) are chips designed for machine-learning calculations, and SSI so far relied mainly on Google’s TPU systems. Selecting Vera Rubin for the next expansion adds a major Nvidia route, but the disclosed terms do not establish that SSI will abandon all Google infrastructure.

Nvidia and SSI plan to collaborate on computing platforms using insights from SSI’s work. SSI gets advanced infrastructure, while Nvidia may receive technical insights from a frontier research customer. Neither company defined the scope of that access or how directly those insights may inform Nvidia’s platforms.

Sutskever tied the hardware demand to SSI’s readiness to scale its work.

“We have research that is worthy of scaling up, and having access to a big Nvidia computer will let us do so.”

Ilya Sutskever, Safe Superintelligence co-founder and chief executive (via NVIDIA Newsroom)

More computing power can support larger or more numerous experiments, but it cannot validate SSI’s methods or establish that their work will produce safe superintelligence. SSI’s researchers must turn installed systems into sustained capacity and repeatable experiments before the expansion changes what the lab can test.

A Safety Lab With No Interim Product

Sutskever and two partners formed SSI in 2024 around a single safe-superintelligence goal. SSI’s focused research mission took priority over recurring product cycles. 

So far SSI has disclosed little about its research publicly. It raised two financing rounds: $1 billion in 2024 and $2 billion in 2025, with the second valuing SSI at a reported $32 billion. The first round arrived before SSI offered any public AI model or product-revenue benchmark.

SSI defines safe superintelligence as its sole product and roadmap, with safety and capabilities developed together, and plans to withhold interim commercial models until it reaches that goal. Its policy leaves outsiders with few intermediate measures of research progress, while compute growth can be measured sooner than scientific success.

SSI must now turn planned Vera Rubin systems into usable capacity. Its computing-capacity plan calls for a tenfold increase during the 12 months following the announcement. Whether SSI deploys and operates enough capacity to achieve that increase will be the partnership’s clearest near-term test.