Safe Superintelligence secures Nvidia investment and Vera Rubin access

Why it matters

Sutskever has persuaded Nvidia to fund an order-of-magnitude expansion of SSI's computing capacity. The deal also gives Nvidia rare access to the lab's closely guarded work and technical input from a potential frontier model developer.

Ilya Sutskever has secured an investment and a large increase in computing power for Safe Superintelligence, the research lab he built around one goal and one product: safe superintelligence.

Nvidia made what it called a "substantial" investment, Bloomberg reported on July 27. Neither the investment amount nor SSI's new valuation was disclosed. The agreement also gives SSI access to Nvidia's Vera Rubin systems, which the companies say will increase SSI's compute by an order of magnitude.

The deal is the clearest public indication that SSI is ready to test its research at far greater scale. Nvidia said it committed after receiving rare access to SSI's work, which the chipmaker described as closely guarded. The partners will also collaborate on current and future Nvidia computing platforms, giving the chipmaker technical feedback from one of the field's most closely watched researchers.

Sutskever reaches the scaling phase

Sutskever helped establish the scaling approach that turned neural networks into the foundation of the current AI industry. At the University of Toronto, he worked with Geoffrey Hinton and Alex Krizhevsky on AlexNet, then co-founded DNNresearch with them before Google acquired it. He later worked at Google Brain and co-authored the 2014 paper "Sequence to Sequence Learning with Neural Networks" with Oriol Vinyals and Quoc V. Le. He then co-founded OpenAI, where he later became chief scientist.

He left OpenAI in May 2024 and introduced SSI the following month with Daniel Levy and Daniel Gross. Sutskever told Reuters SSI made sense because he had "identified a mountain" different from OpenAI's. He also questioned the industry's loose use of scaling as a strategy, telling Reuters: "Everyone just says scaling hypothesis. Everyone neglects to ask, what are we scaling?"

SSI's structure reflects that concern. Its public mission statement promises one goal and one product, with capabilities and safety developed together. SSI says the model frees its researchers from product cycles and short-term commercial pressure.

Sutskever became SSI's chief executive in 2025 after Gross left, TechCrunch reported. The same report said Levy became president. A Stanford-hosted biography says Levy led OpenAI's Optimization team. It describes his doctoral research as spanning machine learning, optimization and privacy, and lists research placements at Facebook Applied Machine Learning, Google Brain and Google Research.

The Nvidia agreement points to a simpler conclusion: SSI now wants to test its private research direction at far greater scale. Nvidia's decision carries unusual weight because its announcement says the chipmaker examined research SSI has kept private.

Nvidia gets technical input from SSI

For Nvidia, the investment secures a role beyond supplying GPUs. SSI will provide input into the development of Nvidia's present and future compute platforms while training on Vera Rubin, the chipmaker's next-generation platform.

Nvidia describes Vera Rubin as its next-generation platform for agentic AI factories. Its partnership announcement identifies SSI as a partner and investee, gives the lab access to Vera Rubin systems and says the companies will collaborate on technical advances to Nvidia's computing platforms. It does not identify SSI as a customer.

The collaboration also gives SSI access to a new compute generation while those systems are ramping into the market. The agreement fits Nvidia's expanding role as a financier of AI development. RuntimeWire reported earlier on July 27 that Nvidia was weighing a guarantee tied to an OpenAI data center lease in Ohio.

Those arrangements put Nvidia's capital, hardware and technical organization beside the labs most likely to generate future demand. In SSI's case, the chipmaker is receiving access and engineering input from a research program that Nvidia says has been closely guarded.

A multibillion-dollar wager on a research lab

Reuters reported that SSI raised $1 billion in cash in September 2024 from investors including Andreessen Horowitz, Sequoia Capital, DST Global, SV Angel and NFDG. The report said SSI would use the money for computing power and a small research staff.

TechCrunch reported in April 2025 that SSI had raised another $2 billion at a $32 billion valuation in a round led by Greenoaks. SSI did not confirm that round in the report. Nvidia's partnership announcement identifies Andreessen Horowitz, DST Global, Greenoaks and Sequoia Capital as SSI investors.

The new capital comes with no disclosed price, ownership stake or valuation. SSI has not disclosed its spending, revenue or a timetable for showing its research. Investors are underwriting Sutskever's record and the possibility that he has found a productive direction beyond the standard recipe of larger datasets and larger training runs.

Nvidia has now seen enough to supply both money and access to its next-generation compute platform. SSI still has to turn that private conviction into evidence the rest of the field can evaluate.