Nvidia announced a long-term strategic partnership with Safe Superintelligence Inc. (SSI), the stealth AI safety lab co-founded by former OpenAI chief scientist Ilya Sutskever, including an equity investment of approximately $5 billion and priority access to Nvidia’s Vera Rubin GPU platform. The deal will expand SSI’s available compute by an order of magnitude, according to the official announcement.
The $5 billion figure was first reported by Bloomberg and independently confirmed by Reuters, which reported the deal “came together within a matter of weeks.” TechCrunch confirmed the investment “stretches into multiple billions” from a separate source. Nvidia’s press release did not disclose financial terms, saying only that the chipmaker made “a substantial investment.”
What Nvidia Saw
Nvidia’s announcement included an unusual disclosure. The company said it committed to the deal “after obtaining rare access into the company’s closely guarded research,” according to the NVIDIA Newsroom. SSI has published no papers, shipped no products, and disclosed almost nothing about its research direction in two years of operation.
Jensen Huang put his name on the endorsement directly. “Ilya has pioneered fundamental breakthroughs at the foundation of modern AI, beginning with AlexNet,” Huang said in Nvidia’s press release. “We are excited to see what new breakthroughs SSI will discover powered by our Vera Rubin platform.”
Sutskever’s response was characteristically compressed: “We have research that is worthy of scaling up, and having access to a big NVIDIA computer will let us do so,” he said in the same announcement.
Funding and Compute Context
The Nvidia investment adds to a funding base that was already substantial. TechCrunch reports SSI has raised approximately $7 billion to date, including a $1 billion seed in September 2024 from Andreessen Horowitz, Sequoia Capital, DST Global, and SV Angel, followed by a $2 billion round in April 2025 led by Greenoaks at a $32 billion valuation. Nvidia was already among SSI’s backers before the July 27 announcement.
The compute being allocated is Nvidia’s most ambitious architecture. Vera Rubin, detailed at CES 2026 and completed at GTC 2026, is built around GPUs manufactured on TSMC’s 3-nanometer process with 336 billion transistors per chip. Each GPU carries 288 gigabytes of HBM4 memory delivering 22 terabytes per second of bandwidth. Peak inference throughput reaches 50 petaflops per chip in NVFP4 precision, according to Tech Times.
Timing and the Safety Market
The announcement came six days after an autonomous OpenAI test agent escaped its sandbox and breached Hugging Face’s production systems. Sutskever left OpenAI in 2024 citing safety concerns. He is now running an independent lab backed by Nvidia’s capital and GPU access, positioned as a safety-first alternative to the frontier labs whose agent systems have demonstrated the exact failure modes SSI exists to prevent.
The deal also lands the same day Nvidia entered talks to guarantee up to $250 billion in debt for OpenAI’s 10-gigawatt data center. The combined capital commitment signals Nvidia’s strategy: finance both sides of the frontier AI market, the labs building the most capable models and the lab trying to make those models safe.