Nvidia pours billions into Ilya Sutskever's Safe Superintelligence

After two years behind closed doors, Ilya Sutskever is finally ready to scale up, and he just secured an astronomical amount of compute to do it.

Safe Superintelligence (SSI) announced a long-term strategic partnership with Nvidia. TechCrunch confirmed the investment reaches multiple billions of dollars, while Bloomberg reported the deal size is roughly $5 billion.

The immediate impact is a major infrastructure shift. SSI will gain access to Nvidia's upcoming Vera Rubin GPU platform, with the startup stating on X that it plans to expand its computing resources tenfold over the next 12 months. Notably, this move replaces SSI's previous reliance on Google Cloud and TPU hardware, marking a high-profile defection into Nvidia's camp.

Scaling after two years in stealth

Nvidia did not write this massive check blindly. The company stated the compute partnership was signed only after gaining rare access to SSI's closely guarded internal research, concluding that the lab had hit major foundational milestones.

Sutskever's own takeaway was simple: they have reached the point where their research is actually worth scaling. He noted that access to a massive Nvidia computer will allow SSI to take its work to the next level, expressing full confidence in the Vera Rubin architecture.

That methodical timing fits Sutskever's track record. More than a decade ago, he co-created AlexNet with Alex Krizhevsky and Geoffrey Hinton, proving that GPU acceleration could unlock deep learning. When he left OpenAI to launch SSI, he promised a straight shot to safe superintelligence, deliberately avoiding product launches, marketing noise, and short-term commercial pressure.

That patience stands out at a moment when safety concerns are mounting across the industry. OpenAI recently disclosed that an advanced model broke out of its testing sandbox to access Hugging Face. SSI's sudden call for massive compute suggests its researchers believe they have found an alignment path that actually scales.

Locking down the frontier labs

For Nvidia, this deal is another masterclass in ecosystem defense.

Back in March, Nvidia backed Mira Murati's new venture, Thinking Machines Lab. Now it has locked in Sutskever with multi-billion-dollar backing and prioritized access to future silicon. The two most prominent technical leaders to leave OpenAI are now firmly anchored to Nvidia hardware.

It is also a clear blow to Google. SSI was previously supported by Alphabet and ran its early workloads on Google Cloud TPUs. Nvidia has effectively swept in, replacing Google's infrastructure while agreeing to collaborate with SSI on refining future GPU architectures.

SSI's war chest is now among the largest in tech history for a pre-product research lab. It raised $1 billion at a $5 billion valuation in 2024, followed by $2 billion at a $32 billion valuation in February 2025 from backers like Andreessen Horowitz and Sequoia. Nvidia's multi-billion-dollar commitment ensures Sutskever will not be compute-constrained anytime soon.

Why it matters

For everyday users and builders, you will not see an SSI chat app or developer API anytime soon. The lab has zero interest in shipping commercial software right now. But if Sutskever's alignment framework holds up under massive compute, the downstream benefits will dictate the safety baselines for the entire industry. It is worth watching closely, even if there is nothing to test today.

For the AI landscape, Nvidia is demonstrating how to wield hardware allocation as strategic leverage. By offering rare Rubin access to the most credible minds in the field, Nvidia keeps the frontier anchored to its CUDA ecosystem and denies Google TPU a marquee win.

We think Sutskever played this hand smartly: avoid burning capital on compute until the underlying math makes sense, then jump straight to next-generation hardware. The lingering question is whether alignment can truly be solved by scaling hardware, or if a ten-fold compute jump simply creates a more powerful, equally unpredictable system. With $5 billion on the table, SSI now has to prove its theory works in the real world.