For two years, Safe Superintelligence Inc. (SSI) said practically nothing. No model, no product, no demo, no revenue – just a website with a manifesto and a $32 billion valuation. On Monday, the AI lab founded by OpenAI co-founder Ilya Sutskever broke that silence: SSI and Nvidia announced a long-term partnership, including an investment by the chipmaker.

The official announcement stays vague on the numbers. Nvidia has made a “substantial” investment, it says, with no financial details disclosed. Bloomberg, citing people familiar with the matter, reports a figure of around five billion dollars. TechCrunch writes that the investment runs into the billions. Neither company has confirmed any amount.

Ten times the compute

The core of the deal isn’t the money – it’s access to chips. SSI gains access to Nvidia’s new Vera Rubin platform and intends to increase its compute “by an order of magnitude,” meaning tenfold. The two companies also plan to collaborate on the technical advancement of Nvidia’s current and future compute platforms, with SSI contributing its perspective on the future of AI.

Vera Rubin is Nvidia’s successor generation to Blackwell, unveiled at GTC in spring 2026. The platform has been in full production since the first quarter of 2026, with partner shipments beginning in the second half of the year. At its heart is the Rubin R100 GPU with 336 billion transistors, built on TSMC’s 3-nanometer process, with 288 GB of HBM4 memory. The Vera Rubin NVL72 rack system packs 72 Rubin GPUs and 36 Vera CPUs. Nvidia promises up to ten times lower cost per token compared to Blackwell.

What stands out is how Nvidia explains the decision: the company says it entered the partnership after obtaining “rare access” to SSI’s closely guarded research. Anyone who wants to know what SSI has built over two years has to rely on the verdict of Nvidia’s due diligence.

“Research that is worthy of scaling up”

Sutskever puts it more soberly: “We have research that is worthy of scaling up, and having access to a big NVIDIA computer will let us do so.” He said the company is “incredibly proud” to be partnering with Jensen and the Nvidia team, and confident that its “big bet” on Vera Rubin will take SSI to the next level.

That sentence is the actual news. In recent months, Sutskever has repeatedly suggested that pure scaling would continue to deliver improvements, but that “something important” would still be missing. His saying now that his research is worth scaling reads as a claim that the missing piece has been found. Nobody outside SSI can verify it.

A lab that deliberately sells nothing

SSI was founded in June 2024 by Sutskever together with Daniel Gross (previously head of AI at Apple) and Daniel Levy (previously OpenAI), shortly after Sutskever left OpenAI – months after the failed attempt to remove Sam Altman as CEO. Gross left SSI for Meta in June 2025; Sutskever took over as CEO. Meta is said to have also approached SSI about an acquisition at the time, which SSI declined.

The business model is the absence of a business model: “one goal and one product,” namely a safe superintelligence. No product cycles, no management overhead, no short-term commercial pressures – that’s how the company describes itself. Its offices are in Palo Alto and Tel Aviv, and the team is deliberately kept small, at a few dozen people.

The funding has been generous nonetheless: one billion dollars in September 2024 at a $5 billion valuation, followed by two billion dollars at a $32 billion valuation in a round led by Greenoaks. Investors include Andreessen Horowitz, DST Global, Sequoia Capital, Lightspeed, SV Angel, Alphabet – and Nvidia, which was already on board before this deal. According to PitchBook data, SSI has raised roughly seven billion dollars to date. Google Cloud additionally supplies the lab with TPUs.

The circularity question resurfaces

The investment lands in a week in which Nvidia’s stakeholding strategy is once again under heavy scrutiny. Over the past two years, the company has invested in its own customers on a grand scale: $6.6 billion in OpenAI in October, six billion in Elon Musk’s xAI in November, up to ten billion in Anthropic. On top of that come stakes in neocloud providers such as CoreWeave, Nebius, Nscale and Lambda. Most recently, it was reported that Nvidia is weighing a $250 billion financing backstop for an OpenAI data center in Ohio.

The criticism is always the same: when a chipmaker helps finance its own customers’ purchases, the resulting revenue growth isn’t purely organic – genuine demand becomes hard to distinguish from financially engineered demand. Huang has repeatedly rejected the “circular financing” charge, arguing that the stakes are simply good investments. And Nvidia isn’t alone in this: Google has backstopped roughly $35 billion in lease payments for Anthropic.

In SSI’s case, though, the structure is somewhat different from OpenAI or Anthropic. There is no product generating revenue, and no double-digit-billion cloud commitment made public in return. What Nvidia is buying here is a share in a bet – and a relationship with the researcher who laid a good part of the groundwork for today’s AI.

Whether that bet pays off won’t become clear until SSI shows something for the first time. There is no timeline for that. Meanwhile, Mira Murati’s startup Thinking Machines Lab, which also counts Nvidia among its major investors, has already put out an open-weight model of its own.