Safe Superintelligence Inc. (SSI)

IWhat it does

Safe Superintelligence Inc., known as SSI, is a private research lab. Its stated goal is to build a safe superintelligence, meaning an AI system that is more capable than people at most intellectual work and that stays under control. The company says this is its sole goal and its only product¹. It has not released a model, an app or an API that a customer can buy or try. Nothing I found says otherwise.

This is a different kind of company from most on this site. There is no sales line to analyse and no customer list. What exists is a team, a research direction, a lot of computing power on order, and investors who have paid a high price for the chance that the research works. So this page cannot do what a normal company page does. It can describe the bet, the money behind it, and what would have to be true for the bet to pay.

Ilya Sutskever, a co-founder of OpenAI and its former chief scientist, announced SSI on 19 June 2024 with Daniel Gross and Daniel Levy¹⁴. Sutskever co-created AlexNet, the 2012 image model that helped start the modern deep learning wave¹⁵, and he left OpenAI in May 2024, after he took part in the failed attempt in November 2023 to remove Sam Altman as chief executive¹⁴. SSI has offices in Palo Alto and Tel Aviv¹.

The company calls its method a straight shot: one goal, one product, no detours into commercial releases. TechCrunch describes it as avoiding product launches and short-term revenue pressure⁷. SSI says its team, investors and business model are all lined up behind that goal, and that the structure is meant to shield safety work from short-term commercial pressure¹. That is a claim about intent. I cannot test it from outside.

Very little about the research is public. When Nvidia announced its partnership in July 2026, it said it had been given rare access to SSI’s closely guarded research and that SSI had spent two years quietly developing a research direction aimed at powerful, robustly aligned AI³. Sutskever said the company has research worthy of scaling up³. Those are statements from the two parties to the deal, not independent evidence that the research works.

IIHow it makes money

It does not, as far as anything public shows. TechCrunch reported in February 2025 that SSI had not yet generated any revenue⁹. A January 2026 TechCrunch piece describes SSI as having no product cycles and no product beyond the foundation model still in development, and reports no revenue figure for it¹². I found no later source that reports revenue, a customer, a pilot or a pricing plan. I treat revenue as zero.

So the business model today is simple: investors fund research in return for equity. SSI has raised about $3 billion in two rounds, according to TechCrunch⁷¹². In September 2024 it raised $1 billion from NFDG, Andreessen Horowitz, Sequoia Capital, DST Global and SV Angel, as the company’s own updates page records². A second round of about $2 billion followed in early 2025⁷, reported by the Financial Times as led by Greenoaks⁸.

In July 2026 Nvidia announced a long-term partnership and an investment³. Nvidia’s own release gave no amount and called it substantial³. TechCrunch’s source said the sum was in the multiple billions, and TechCrunch relayed Bloomberg’s report of $5 billion⁷. Reuters, citing a person briefed on the deal, also called it a $5 billion equity investment²⁵. Treat $5 billion as press-reported, not confirmed by the company.

The deal is also a compute arrangement. SSI gets access to Nvidia’s next-generation Vera Rubin platform, which Nvidia says raises SSI’s computing capacity by an order of magnitude, roughly ten times³. A chipmaker investing in a lab that then buys its chips is a circular flow of money. It is common in this sector now, and it makes the headline investment figure less clean than a cash investment from a pure financial backer.

What a future revenue model could look like is open. The two obvious routes are selling access to a very capable model, as OpenAI and Anthropic do, or licensing the technology. Sutskever has reportedly described reasons SSI might change course, such as long AI timelines or the value of powerful AI reaching the world¹². I treat any revenue model as unproven and put it in my scenarios only as an assumption.

What is reported about SSI’s funding and money

IIIThe market and the competition

SSI competes in the market for frontier AI models, the most capable general systems. That market has a few very well funded players. OpenAI announced on 31 March 2026 that it had closed a round of $122 billion in committed capital at a post-money valuation of $852 billion¹⁸. Anthropic announced a $65 billion Series H at a $965 billion post-money valuation on 28 May 2026¹⁷, and Anthropic filed a Form S-1 for a proposed IPO on 26 October 2026⁵. Google and Meta run their own labs with large balance sheets.

Next to those, SSI is small and early. It has had roughly $3 billion of outside money before Nvidia¹² and was believed to have around 20 employees in March 2025²². The big labs earn large revenue today. Anthropic’s amended S-1 reports revenue of $4.6 billion for 2025 and $33.2 billion for the first nine months of 2026²³. OpenAI has published no audited accounts. It said at its March 2026 funding close that it was generating about $2 billion of revenue a month¹⁸²⁹, and Fortune reported leaked, unaudited financials showing 2025 revenue of $13.07 billion²⁴.

The closest peers are the newer, research-heavy labs. Thinking Machines Lab, founded by former OpenAI chief technology officer Mira Murati, raised about $2 billion at a $12 billion valuation in July 2025¹⁶. It announced its own gigawatt-scale Nvidia partnership in March 2026⁶ and has released open-weight models and a fine-tuning API⁶. That is the contrast: Thinking Machines ships products, SSI does not. Mistral AI in Europe raised €3 billion in September 2026 at a post-money valuation of more than €21 billion, the company said¹⁹.

There is also a talent market. Meta tried to buy SSI in early 2025 and Sutskever refused, according to anonymous sources¹³. Meta then hired SSI’s chief executive Daniel Gross¹³, and Gross left effective 29 June 2025². In a lab where the main asset is a small group of researchers, that is a real exposure.

The market is also where the compute is. Training the largest systems takes very large amounts of chips and power. SSI’s main cloud provider was reported to be Google Cloud using its TPU chips as of April 2025¹¹, and it has now added Nvidia as a partner³. Compute is a cost SSI must keep paying. The big labs fund it from revenue. SSI funds it from investor money.

IVThe moat and how it could be disrupted

I rate the moat thin. That is not a comment on the quality of the people. It is that a moat is a durable advantage that protects earnings, and SSI has no earnings yet. What it has is a reputation, a small team of researchers, privileged access to a large amount of compute, and an unpublished research idea.

Each of those can erode. Reputation does not stop a rival from reaching the same result. Researchers can leave, and some already have, since one of the three founders went to Meta in 2025¹⁰. Compute is bought on the open market by anyone with money, and larger labs have far more of it. An unpublished idea is protected only while it stays secret and while nobody else finds the same thing. In AI, similar ideas tend to appear in several places at about the same time.

If SSI did reach a clearly superior system first, a moat could form quickly: the model itself, the data it would generate, and the lead time. That is the case the investors are paying for. But a lead in model quality has proved short in this market. Rivals have released comparable systems within months, and open-weight models such as Thinking Machines’ Inkling, at 975 billion parameters, are released for anyone to use²⁶.

The company’s own structure is a second reason I stay cautious. It has said it will avoid commercial products¹⁷. That keeps the research focused, but it also means no customers, no feedback from real use and no switching costs. A company that sells nothing cannot lock anyone in. If SSI later decides to sell, it would be starting that work behind labs with years of distribution and enterprise contracts.

VThe numbers we can see

There are very few. SSI is not an SEC registrant. A search of the SEC’s EDGAR company database for the name returned no matching registrant⁴. That is not proof that no private-placement notice exists, because such notices can be filed under a different legal name, but I found none. There are no audited accounts and no company-reported revenue, loss or cash balance.

The valuation history is press-reported. The September 2024 round is described as a $5 billion valuation⁷. In February 2025 Reuters reported talks at a valuation of at least $20 billion, four times the earlier figure⁹. In April 2025 the Financial Times reported a $32 billion valuation, on an additional $2 billion⁸. In March 2025 Calcalist wrote that SSI was reportedly raising more than $2 billion at a valuation above $30 billion²². I show both and use $32 billion as the press-reported figure, since three TechCrunch pieces and CNBC use it⁷⁸¹⁰¹³.

There is no confirmed valuation after the Nvidia investment. Neither Nvidia’s release³ nor the TechCrunch report⁷ states one.

The best I can say about spending is an order-of-magnitude statement. Money raised before Nvidia was about $3 billion¹². A team of about 20 people (the March 2025 estimate) on typical frontier-lab salaries would use only a small part of that each year, so compute is the likely main cost. That is my inference, not a reported number. I have no figure for cash left.

Valuation marks for SSI and comparable private labs (SSI’s marks are press-reported; the others come from company announcements or press reports)

VIWho runs it and who owns it

Ilya Sutskever is chief executive. He took the role after Daniel Gross left effective 29 June 2025, and Daniel Levy became President²¹⁰. Gross had been CEO and co-founder; he had previously led AI efforts at Apple and was reported to be joining Meta¹³. Sutskever has a strong research record, with AlexNet¹⁵ and a founding role at OpenAI¹⁰. He is better known as a scientist than as a business operator, and TechCrunch’s Jan 2026 piece makes that point about him¹². I cannot judge how well he will run a larger company, or how the company copes with a leader who has said little in public about its commercial plans.

SSI described the leadership change in a message to the team and investors. It said it was not pursuing acquisition interest and that the technical team reports to Sutskever. It closed with the line ’We have the compute, we have the team, and we know what to do’². Whether that is right is exactly what investors cannot yet check.

Ownership is not public in detail. Named backers are Andreessen Horowitz, DST Global, Greenoaks, Sequoia Capital³, NFDG and SV Angel². TechCrunch also lists Alphabet, Lightspeed Venture Partners and GV among the backers⁷; I did not find a primary source for those. Nvidia is a new holder³. Founders hold stakes but the sizes are unknown. There is no public cap table, no information on preferred share rights, and no information on how much of the company employees hold.

A note on governance. SSI is a for-profit company, and I found no public statement about a special board or a safety charter that I could verify. Its commitment to safety over speed is a stated mission¹, not a binding rule I can point to. An investor would want to read the shareholder agreement to know whether it can be enforced.

VIIValuation: is the last price supported?

Equity value in US$ billions. Not a share price: a private company has none. Latest mark: primary round, press-reported (Financial Times via TechCrunch); post-money status not stated.

What the last valuation assumes

At about $32 billion, the April 2025 mark assumes that SSI has a meaningful chance of building a frontier system and then earning revenue on the order of tens of billions a year. With no revenue today, the price rests on belief in the team and the unpublished research. For the mark to hold on cash-flow grounds, I would need to believe in a path like my base case, in which SSI reaches the frontier and sells at scale by the early 2030s, and I would also need to believe that this outcome is more likely than not.

Bear, base and bull

• Bear. The research does not reach the frontier. SSI sells a small amount of technology or is bought for its team: $1 billion of revenue in 2032 at 5 times sales, about $5 billion. Discounted six years at 25% (a venture-stage rate, well above the 10% or so used for mature companies), with 10% dilution, that is about $1.2 billion, plus $5 billion of assumed leftover cash. About $6 billion. Revenue and cash are my assumptions.
• Base. SSI reaches a competitive model and sells it: $20 billion of revenue in 2032, a figure I chose, not reported. A multiple of 8 times sales sits between Alphabet at 7.0 times²¹ and Nvidia at 16.0 times²⁰. Value in 2032 is $160 billion. Discounted at 25% for six years gives about $42 billion, less 25% dilution from further funding, plus $5 billion of cash. About $37 billion. In my model this case is conditional on success and is not probability-weighted, so the true expected value is lower. Put another way, the base figure answers the question of what the company is worth if it works, not what it is worth today.
• Bull. SSI leads the frontier: $80 billion of revenue in 2032 at 10 times sales, $800 billion, in line with the very large marks on OpenAI and Anthropic¹⁷¹⁸. Discounted at 25% gives about $210 billion, 30% dilution, plus cash. About $150 billion.

Comparables used

Where I come down

The April 2025 mark of $32 billion is a little below my base value of $37 billion, but my base is a success case and not an expected value. If I weight the three cases by the odds I gave earlier, the expected value is well below $37 billion. For example, with 60% on the bear case, 30% on the base and 10% on the bull (my own weights, not measured odds), the weighted value is about $30 billion. The result moves a lot when the weights move. A lab that is paid for the right to be one of the few to reach frontier AI should trade at a premium, and the market may be right: Thinking Machines was priced at $12 billion for a team with products, Mistral at more than €21 billion, and the leaders near $1 trillion^16191718. On those comparisons $32 billion for a team with Sutskever’s record is not out of line. But the mark is more than 18 months old and was set before the Nvidia deal, so it is stale. A new round would probably be priced higher, and I have no figure for it.

What I cannot judge

I cannot judge whether the research works, how much cash SSI has left, who owns what, or what valuation the Nvidia deal implied. I did not read audited accounts because none are public. The $5 billion Nvidia figure and the $32 billion valuation come from press reports I could not trace to primary documents. The revenue and multiple inputs in my model are my own assumptions, so the range of $6 billion to $150 billion should be read as an indication of uncertainty and not as a forecast. This is research, not investment advice.

VIIIThe IPO

No IPO plan reported. I found no report of an IPO plan, a confidential filing or a banker mandate for SSI. EDGAR shows no matching registrant⁴. SSI said in July 2025 that it was not pursuing acquisition interest and was focused on its work². Larger peers are moving: Anthropic filed a Form S-1 on 26 October 2026⁵, and OpenAI said in June 2026 that it had confidentially submitted an S-1²⁷. CNBC reported, citing two sources, that OpenAI’s chief financial officer told employees in August the company will be public in 2027, possibly sooner²⁸. Those listings could set public-market prices for AI labs, which would bear on SSI’s next mark.

What to watch

• A new priced round or tender, and the valuation attached to it
• Any Form D or other SEC notice under SSI’s legal name
• How Anthropic’s S-1 discloses revenue, losses and compute costs, and how its IPO is received
• Whether SSI publishes research or releases a model, or states a revenue plan
• Departures of senior researchers, or new acquisition approaches
• Confirmation of the size and terms of the Nvidia investment

IXWhat could go wrong

What most likely destroys this business is simple: the research does not produce a system that is clearly better than what the large labs ship, and the money runs out before it does. Frontier labs burn billions per year. In leaked, unaudited figures reported by Fortune, OpenAI’s 2025 operating loss was $20.9 billion²⁴. SSI is a fraction of that scale, but it has no revenue to offset its costs.

Several things have to go right at the same time. First, the research direction has to work, which I will put at one in three or less for reaching a frontier-leading system, though that figure is my own judgement and not a measured probability. Second, SSI has to keep its key researchers. Third, it has to raise further money at a rising valuation, probably tens of billions more. Fourth, it has to turn the result into something that earns money, despite its stated aversion to products. Fifth, it has to do this before larger labs reach similar capability. If each of these were about 60 to 70 percent likely and independent, the combined chance would be roughly 10 to 15 percent. They are not truly independent, but the sum shows how much has to go right.

What an owner can take out is also limited. SSI pays no dividends and has no revenue. A holder gets value only through a sale of the company, a later funding round with secondary sales, a tender, or an IPO. Meta’s interest shows an acquisition is possible¹³, and SSI said in 2025 it was not pursuing one². Employee and early-investor liquidity is unknown.

There are other risks. Nvidia is both investor and supplier, so the economics of the compute deal may favour Nvidia³. The safety claim is a promise and could be tested if competition gets tough. And the whole sector is priced on hopes, so a drop in confidence in AI spending would lower the price of any lab that has no revenue.

The next best use of the same dollar is, for a public-market investor, a company that sells the picks and shovels. Nvidia trades at 16.0 times sales on $303 billion of trailing revenue²⁰, and Alphabet at 7.0 times on $471 billion²¹. Those are businesses with revenue. They carry their own risk, but they do not need a research breakthrough. This is a comparison, not advice.