Nvidia Takes a Stake in Sutskever’s Safe Superintelligence

Nvidia (NVDA ) has invested in Safe Superintelligence Inc. and agreed to supply the lab with its next-generation Vera Rubin systems, under a long-term partnership the two companies announced on July 27, 2026. Nvidia says the arrangement will expand SSI’s compute by an order of magnitude.

The release sets out the shape of the deal. Nvidia took a position in SSI alongside what it describes as rare access to the lab’s closely guarded research, and the two companies will work together on Nvidia’s current and future compute platforms, with SSI’s insights feeding the chipmaker’s design work. SSI was founded in 2024 and is led by Ilya Sutskever and Daniel Levy; Andreessen Horowitz, DST Global, Greenoaks and Sequoia Capital are among its investors.

The lab’s own site describes one goal and one product, a safe superintelligence, and a business model deliberately insulated from short-term commercial pressure: research first, with no interim product funding the work along the way.

“We have research that is worthy of scaling up, and having access to a big NVIDIA computer will let us do so,” said Sutskever, SSI’s cofounder and CEO.

What an order of magnitude buys

An order of magnitude is a ratio rather than an absolute, so what it delivers depends on the footprint SSI runs today, which the lab holds closely. Nvidia’s comparable arrangements give a sense of the scale these deals work at. When it announced a partnership with Thinking Machines Lab on March 10, 2026, the company named at least one gigawatt of Vera Rubin systems, with deployment targeted for early 2027.

What SSI would be getting

Vera Rubin is Nvidia’s current flagship, announced at its March 2026 GTC conference with seven chips in full production. The NVL72 rack pairs 72 Rubin GPUs with 36 Vera CPUs over NVLink 6. Nvidia says the configuration trains large mixture-of-experts models with a quarter of the GPUs that Blackwell required, and delivers up to ten times the inference throughput per watt at a tenth of the cost per token.

Those are Nvidia’s own figures, and it has published comparable ones for the Vera CPU in chip-design workloads. Vera Rubin systems began reaching partners in the second half of 2026, so SSI’s expansion tracks that ramp.

Nvidia’s book of customer stakes

The investment half of the deal is visible in Nvidia’s accounts even when individual cheques are not. In its first-quarter results for fiscal 2027, covering the three months to April 26, 2026, the company reported $43.4 billion in non-marketable securities, its holdings in private companies, up from $22.3 billion three months earlier. That book nearly doubled in a quarter, while data-center revenue reached $75.2 billion.

The structure is not unique to Nvidia. AMD’s Anthropic arrangement follows similar logic. What it means in practice is that some of the demand Nvidia reports now comes from customers it has also funded. For a pre-product research lab, that is how frontier-scale compute gets paid for, at a point when the rest of the buildout is running on a fast-growing AI debt market.

What to watch

• The size of the stake. Nvidia itemizes material equity positions in its quarterly filings. The next one will show whether the SSI investment clears that bar.
• Where the compute lands. Capacity at this scale needs a physical site and a power supply, details that typically surface through interconnection queues and construction records.
• The Vera Rubin ramp. SSI’s expansion is contingent on systems that only began shipping through partners this year.

For Nvidia, the deal buys early sight of two years of frontier research and a design partner whose findings feed back into its own chip roadmap. For SSI, it converts a research agenda into the compute needed to test it at scale.