Ilya Sutskever's Safe Superintelligence Startup | Generated by AI

Home 2026.06

Question: How is Ilya Sutskever’s company doing now?

Answer:

Sutskever is now CEO of Safe Superintelligence Inc. (SSI), the AI startup he launched in June 2024. Co-founder Daniel Gross left the company as of June 29, 2025, and Daniel Levy became president.

Funding & Valuation

SSI has raised $3 billion in capital, with the funding timeline showing explosive growth: $1 billion in September 2024 (from a16z, Sequoia, DST Global, SV Angel), then a $30 billion valuation in March 2025 (sixfold increase in 6 months). The company is valued at $32 billion, though it has ~50 employees and has not generated revenue.

Strategic Positioning

SSI’s positioning is “one goal, one product”: safe superintelligence, with no side products, SaaS, or consulting. It’s a straight-shot research lab with no intermediate commercial product. Sutskever has no immediate plans to release a product, explicitly wanting to avoid the “rat race” of current AI market competition.

Technical Resources

Google Cloud announced a partnership in April 2025 to provide TPUs (Tensor Processing Units) for SSI’s research, marking a significant technical collaboration. This is key—they’re compute-constrained like every other lab but have serious institutional backing.

Recent Drama

Meta CEO Mark Zuckerberg was in advanced talks to hire Gross and former GitHub CEO Nat Friedman, and reportedly attempted to acquire all of SSI. The fact that Zuckerberg couldn’t poach them or acquire the company at $32B suggests either (a) Sutskever convinced them to stay, or (b) the valuation/terms were unrealistic. Sutskever taking the CEO role himself signals commitment.

The Real Strategy

What’s interesting here: Sutskever acknowledged that if the timeline to superintelligence is longer than anticipated, SSI may release a product. He could also release one sooner if he felt it would be useful for the world to see powerful AI in action. This is pragmatism wrapped in principle—he’s building toward AGI but maintains flexibility.

This positions SSI differently from both OpenAI (product-first, commercial) and DeepSeek (open weights, aggressive iteration). Sutskever is betting that with $3B and GPT-scale compute, he can build a path to superintelligence that’s genuinely safer by design, not bolt-on alignment research. Whether that thesis holds depends on whether safety can actually be baked into architecture/training rather than bolted on post-hoc—which is still an open empirical question.

References:

Back Donate