Ilya Sutskever's Safe Superintelligence raises $5 billion from Nvidia despite not yet releasing a product
The secretive AI startup has yet to publish research or launch a product, but Nvidia is making what the company calls a substantial investment that will increase its computing capacity tenfold over the next year.
For most AI startups, convincing investors requires showing products, publishing research or demonstrating commercial traction.
Safe Superintelligence (SSI) has done none of those things.
Yet the company founded by former OpenAI chief scientist Ilya Sutskever has secured another enormous vote of confidence. Nvidia is making what SSI described as a "substantial investment" as part of a long-term strategic partnership that will increase the company's computing capacity tenfold over the next 12 months.
According to Reuters, Nvidia's investment amounts to $5 billion, making it one of the largest strategic bets the chipmaker has made on a frontier AI company.
"We are announcing a long-term strategic partnership with NVIDIA," SSI wrote on X. "NVIDIA is making a substantial investment in SSI that will let us 10x our compute in the next 12 months. We reached the point where our research is worth scaling and with this partnership we will be able to."
Neither company disclosed the size of the investment or SSI's valuation. The startup previously raised around $7 billion and was reportedly valued at $32 billion, making it one of the world's most valuable privately held AI companies despite remaining largely invisible to the public.
The partnership gives SSI access to Nvidia's next-generation Vera Rubin platform and marks another sign that Nvidia is extending its influence beyond supplying chips to becoming a strategic investor in companies it believes could shape the future of artificial intelligence.
In announcing the partnership, Nvidia said it had been granted rare access to SSI's closely guarded research before deciding to invest.
Founded in 2024 by Sutskever and Daniel Levy, SSI now employs only a few dozen people, including a research team in Tel Aviv. The company has yet to release its first product or publish research describing the technology it is building.
That secrecy stands in sharp contrast to much of the AI industry, where companies frequently showcase benchmarks, research papers and product launches to attract customers and investors.
Instead, SSI has largely asked investors to trust Sutskever himself.
One of the most influential researchers in modern AI, Sutskever co-founded OpenAI and played a central role in breakthroughs ranging from AlexNet to GPT models and OpenAI's reasoning systems. After leaving OpenAI, he founded SSI with a single objective: building what he calls "safe superintelligence."
In a rare interview last year, Sutskever argued that the AI industry's current approach has reached its limits. He said the era of simply scaling models with more data and larger computers is ending and that future progress will require fundamentally new scientific ideas.
"The age of scaling is ending," he said, arguing that today's large language models remain surprisingly fragile despite their impressive benchmark performance.
Sutskever has repeatedly argued that current models suffer from poor generalization, performing well on tests but failing in unpredictable ways in real-world tasks. Rather than pursuing ever-larger models, SSI is attempting to discover what he has described as a new machine-learning principle that could produce more robust intelligence.
Until now, the company has offered little evidence publicly that this approach is working.
Its announcement with Nvidia may represent the clearest signal yet that SSI believes it has reached an important milestone internally.
"We reached the point where our research is worth scaling," the company wrote.
Whether that confidence ultimately translates into a product, or validates Sutskever's belief that the next AI breakthrough will come from research rather than brute-force scaling, remains one of the biggest unanswered questions in artificial intelligence.
