Safe Superintelligence (SSI): Inside Ilya Sutskever's Lab

Safe Superintelligence Inc. (SSI) is the company that put the goal in its name. Founded in June 2024 by Ilya Sutskever, Daniel Gross and Daniel Levy, it says it has “one goal and one product: a safe superintelligence,” and it has organized everything else around that sentence: no products, no public research, three posts on its updates page in more than two years, and, by most reports, about $8 billion raised.

That combination makes SSI one of the most discussed and least documented labs in the field. This page separates what SSI itself has said, what has been reported by named outlets, and what is rumor. It describes a stated goal. It does not imply that SSI, or anyone, has built Super Intelligence. SSI has not claimed to.

SSI at a glance

Fact	Detail	Basis
Founded	June 19, 2024	Announced by SSI
Founders	Ilya Sutskever, Daniel Gross, Daniel Levy	Announced by SSI
Leadership	Ilya Sutskever (CEO), Daniel Levy (President)	Confirmed by SSI, July 2025
Offices	Palo Alto, California and Tel Aviv, Israel	Stated on ssi.inc
Team	More than 50 people	Reported by Globes, August 2026
Funding	$1 billion confirmed; about $7 billion more reported	See “Money and compute” below
Compute	Google Cloud TPUs (2025); NVIDIA Vera Rubin (2026)	Announced by Google Cloud and NVIDIA
Products	None released	SSI’s own channels, checked October 21, 2026
Published research	None on SSI’s own channels	Same check

What SSI says it is building

SSI’s founding statement, which still opens its website, calls the company “the world’s first straight-shot SSI lab.” In SSI’s usage, the acronym stands for the goal as well as the company: building a safe superintelligence is, in its words, “the most important technical problem of our time.”

The statement makes three commitments that explain how the company behaves:

• Safety and capability together. SSI treats the two as technical problems to be solved at the same time, and says it intends to advance capabilities as fast as possible while keeping safety ahead of them.
• No product cycle. The company says its single focus means no distraction from management overhead or product releases, and that its business model is designed to keep the work insulated from short-term commercial pressure.
• The name is the roadmap. SSI describes itself as having one mission and one product, with the name, the mission and the product roadmap all being the same thing.

This is why SSI appears first in our map of who is building Super Intelligence and why its founding is a company-name event on the Rise of Super Intelligence timeline. Other labs adopted the word for a division (Meta Superintelligence Labs), a team (Microsoft AI) or a mission statement. SSI is the clearest case of a company whose legal identity is the research goal.

Where it came from

SSI is in large part a continuation of a project Sutskever began at OpenAI, where he was a co-founder and chief scientist. In July 2023 he co-led OpenAI’s Superalignment effort, which set out to solve the core technical problems of aligning superintelligent systems within four years, backed by a fifth of the compute OpenAI had secured at the time. As a member of OpenAI’s board, he voted in November 2023 to remove Sam Altman as chief executive and later said he regretted his part in it. He left OpenAI in May 2024, and the Superalignment team was dissolved the same month.

Five weeks later he announced SSI. His co-founders came from different directions. Daniel Gross, born in Jerusalem, had led AI work at Apple and been a partner at Y Combinator. Daniel Levy, who holds a computer science doctorate from Stanford, had been a researcher at OpenAI. SSI announced its Tel Aviv office on day one, and Sutskever, like Gross, grew up in Israel. CTech describes SSI as keeping a significant team there.

Timeline

Date	What happened	Basis
Jul 5, 2023	OpenAI launches Superalignment, co-led by Sutskever	OpenAI
May 2024	Sutskever leaves OpenAI; Superalignment team dissolved	Wikipedia, SI Builders timeline
Jun 19, 2024	SSI announced by Sutskever, Gross and Levy	SSI, Techmeme
Sep 4, 2024	SSI raises $1 billion; a valuation of about $5 billion is reported but not confirmed	SSI, The Register
Apr 9, 2025	Google Cloud says SSI is using its TPUs for research	TechCrunch
Apr 2025	$2 billion round at a $32 billion valuation reported, led by Greenoaks, with Alphabet and NVIDIA investing	Computing (citing Reuters and the Financial Times)
Jun 2025	Meta reported to have tried to acquire SSI and to have been turned down	CNBC
Jul 3, 2025	SSI confirms Gross left on June 29; Sutskever becomes CEO, Levy President	SSI
Nov 25, 2025	Sutskever’s long-form interview on SSI’s approach	Dwarkesh Podcast
Jul 26–27, 2026	Long-term partnership with NVIDIA, including an investment and access to Vera Rubin; $5 billion reported	SSI, NVIDIA, TechCrunch
Aug 2026	An investor says SSI will release its first model that month; SSI does not confirm	StartupHub.ai, Globes
Sep 29, 2026	Executive Order 14434 makes “SI” the federal label for AI; SSI is not among the accord’s signatories	White House, Infosecurity Magazine
Oct 21, 2026	No model, product or paper released; no update on ssi.inc since July 26	SI Builders check of SSI’s channels

Money and compute

Only one funding round has been announced by SSI itself. On September 4, 2024 it posted that it had raised $1 billion from NFDG (the fund run by Gross and Nat Friedman), a16z, Sequoia, DST Global and SV Angel. Coverage at the time put the valuation at about $5 billion and described a team that had barely reached double digits.

Everything after that comes from reporting. In April 2025, Reuters and the Financial Times reported a $2 billion round at a $32 billion valuation, led by Greenoaks with $500 million, with Alphabet, NVIDIA, a16z, DST Global and Lightspeed participating. In July 2026, NVIDIA’s investment was put at $5 billion by Bloomberg and at “multiple billions” by a TechCrunch source. Neither company disclosed the amount. Added together, those reports come to roughly $8 billion. We use that figure only as a sum of reports.

Compute has followed the money. In April 2025 Google Cloud said SSI would use its TPUs, and a source told TechCrunch that Google was SSI’s main compute provider. In July 2026 SSI announced a long-term partnership with NVIDIA that gives it access to the Vera Rubin platform; SSI said the deal would let it scale its compute tenfold. Sutskever’s explanation was one line: “We have research that is worthy of scaling up.” NVIDIA also said the two companies would work together on its current and future compute platforms.

The 2025 leadership change

SSI’s one public leadership change came in mid-2025. CNBC reported that Meta had tried to buy SSI earlier in the year, that Sutskever turned down both the acquisition and an offer to hire him, and that Mark Zuckerberg then negotiated with Gross, who was SSI’s chief executive at the time. Gross joined Meta’s new superintelligence organization, now Meta Superintelligence Labs; our MSL profile covers the episode from Meta’s side.

On July 3, 2025, Sutskever confirmed the change on SSI’s updates page: Gross had left on June 29, Sutskever was formally CEO, Levy was President, and the technical team reported to Sutskever. He acknowledged rumors of companies looking to acquire SSI and wrote that the company had the compute and the team and knew what to do.

What Sutskever has said about the approach

SSI publishes almost nothing about its research, so the most detailed account of its thinking is a single interview. In November 2025, Sutskever told Dwarkesh Patel that the field had moved through an age of research (roughly 2012 to 2020) and an age of scaling (2020 to 2025), and was now returning to research, “just with big computers.” The points that bear on SSI:

• A different bet. He said SSI has “a different technical approach” that it considers worth pursuing, without describing it, and argued that SSI’s research compute is closer to the large labs’ than it looks, because much of their compute goes to serving products.
• A learner, not a finished system. He described the target less as a system that already knows everything than as one that can learn any job quickly, comparing it to an eager, very capable 15-year-old.
• “Straight shot,” with caveats. He said the no-product plan could change if timelines turn out long, and that even a straight shot would involve releasing the system gradually.
• Timelines. Asked when a system that learns as well as a human might arrive, he gave a range of five to 20 years.
• Alignment. He suggested aiming for AI that is robustly aligned to care about sentient life, and argued that may be easier than caring about humans alone.
• Revenue. He gave no business model, saying SSI was focused on research and the answer would emerge from it.

These are statements of intent and opinion in an interview, not company commitments, and SSI has published nothing that tests them.

The August 2026 model rumor

In early August 2026, Gavin Baker, chief investment officer of Atreides Management, said on the Invest Like the Best podcast that SSI had said it would release its model in August, according to StartupHub.ai. Over the following weeks, posts on X from investors and researchers suggested that a model had been shown privately and that it might keep learning after training. Globes summarized the claims on August 26, noting that Sutskever had not commented.

As of October 21, 2026, none of this has been confirmed. SSI has published no model, API, paper or announcement, and its updates page has no entry after the NVIDIA partnership on July 26. Claims about the model’s architecture, including continual or test-time learning, are inferences by outside observers. SI Builders will log a release on the Rise of Super Intelligence timeline only when SSI announces it on its own channels.

SSI and the federal “SI” label

Since Executive Order 14434 on September 29, 2026, “SI” has also been the U.S. executive branch’s word for ordinary AI. The overlap is a coincidence of vocabulary. SSI’s acronym predates the order by more than two years and refers to the research goal, a system that would exceed human capability, not to AI in general. The federal “SI” currently means exactly what the statutory definition of AI means. Our guide on Super Intelligence vs AI explains the difference.

SSI was not among the companies whose leaders signed the White House Accord on Super Intelligence on the day of the order, according to coverage of the signing, and we have found no public statement from SSI about the rename. Its website continues to write “superintelligence” as one word, as it always has.

Open questions

• What is the technical approach? SSI has said it has one and that it is worth scaling. It has not said what it is.
• When, and how, will anything ship? The founding plan was a single straight shot. Sutskever has since allowed for a gradual release. Neither has a date.
• How does a no-product company pay for compute at this scale? The answer so far has been investors and compute partners. Sutskever has said the business model will follow from the research.
• What does “safe” mean in practice? SSI has not published a safety framework, evaluations or alignment research on its own channels.

How this page is maintained

Every claim links to a dated source. We mark which facts SSI has confirmed and which come from reporting, and we do not repeat figures or rumors without saying where they came from. Quotations are verbatim. We write “Super Intelligence” as two words in our own voice and keep the official spelling in the company’s name and in quotes.

This page was first published on 2026-10-21 and is updated when SSI announces something on its own channels or when a named outlet reports a material change. Corrections and additions, with sources, to corrections@sibuilders.io.