AGI Startups in 2026: What Billion-Dollar Seed Rounds Teach Founders

SSI, Thinking Machines Lab, AMI Labs, Ineffable Intelligence and others raised billions before shipping a product. Here are the rounds exactly as reported, which ones actually closed, and the practical lessons a regular founder can take from them.

AGI startups are the young labs that say plainly they are building artificial general intelligence or superintelligence, and between 2024 and 2026 investors gave several of them billion-dollar rounds before they had a product to sell. Safe Superintelligence, Thinking Machines Lab, AMI Labs, Ineffable Intelligence, Recursive Superintelligence and Reflection AI were all funded on the record of their founders and a claim on computing power.

The short version: Thinking Machines Lab “officially closed a $2 billion seed round led by Andreessen Horowitz,” TechCrunch reported in July 2025, at a $12 billion valuation. Its much bigger 2026 round is still talk: The Next Web, summarising The Information’s updated report in September 2026, wrote “None of it is signed.”

A quick definition first. AGI, in common usage, means AI that can match people across most kinds of cognitive work; there is no single agreed definition, and some of these companies avoid the word entirely. SSI talks about “a safe superintelligence,” meaning a system beyond human level. For a founder, the label matters less than what the money behind it reveals.

Which AGI startups raised the biggest rounds?

The biggest reported rounds went to labs led by former OpenAI and Google DeepMind researchers, often within months of founding. Every figure below is as reported by the outlet named, on the date shown; only some were confirmed by the company.

Lab (lead founder) Round, as reported Reported by, date
Safe Superintelligence (Ilya Sutskever) $1B raised SSI’s own updates page, September 2024
Safe Superintelligence $2 billion at a $32 billion valuation Financial Times via TechCrunch, early 2025 (reported April 2025)
Thinking Machines Lab (Mira Murati) $2 billion seed at $12 billion, closed TechCrunch, July 2025
Reflection AI (Misha Laskin, Ioannis Antonoglou) $2 billion at $8 billion The New York Times via SiliconANGLE, October 2025
AMI Labs (Yann LeCun, executive chair) US$1.03 billion seed at US$3.5 billion Singapore EDB, March 2026
Recursive Superintelligence (Richard Socher, Tim Rocktäschel) At least $500 million at $4 billion pre-money Financial Times via The Decoder, April 2026
Ineffable Intelligence (David Silver) $1.1 billion seed at $5.1 billion Quartz, April 2026
Safe Superintelligence Nvidia investment, amount undisclosed TechCrunch, July 2026 (Bloomberg reported $5 billion)
Thinking Machines Lab Talks only: $5bn to $6bn at a pre-money of at least $40bn The Information via The Next Web, September 2026

Reflection AI shows how fast these prices moved. SiliconANGLE, citing the Times, said the company was valued at $545 million in March 2025, when it launched with $130 million, before the October round at $8 billion.

Why would investors pay billions for a lab with no product?

Because the rounds bought people and a head start on compute, not revenue. The companies’ own reported status makes that plain. When it closed its seed, TechCrunch said Thinking Machines Lab was under a year old and “has yet to reveal what it’s working on.” Quartz described Ineffable Intelligence, founded in November 2025, as having no product, no revenue and no published roadmap. The Decoder said Recursive Superintelligence was four months old and that its self-improving AI idea was still at the research stage, according to the FT. Tech.eu put AMI Labs at under six months old.

What those labs did have were founders with famous research records: a former OpenAI chief scientist, a former OpenAI chief technology officer, the former head of reinforcement learning at Google DeepMind, and LeCun, who left Meta in 2025. Investors were pricing the odds that a handful of people would build something frontier-grade first.

That logic does not transfer to a normal seed round, and our look at VC fundraising concentration in 2026 explains how unevenly this year’s money is spread. Using an AGI lab’s seed as a comparable for your own raise is like pricing a starter home off a stadium sale.

Why does Nvidia keep showing up on the cap table?

Because a frontier lab’s biggest cost is chips, and the chipmaker gains when those labs grow. Nvidia appears in the reported investor lists for Thinking Machines’ seed, Reflection AI (which it led), Recursive Superintelligence, Ineffable Intelligence and SSI.

The SSI deal spells out the trade. TechCrunch reported that it gives SSI access to Nvidia’s Vera Rubin platform, expected to lift the startup’s compute “by an order of magnitude.” Sutskever’s own explanation was blunt: “We have research that is worthy of scaling up, and having access to a big NVIDIA computer will let us do so.” Thinking Machines signed a multiyear partnership in March 2026 to deploy at least one gigawatt of Vera Rubin systems, and in the round now under discussion Nvidia would supply roughly half the capital, per The Next Web.

The founder lesson is about circular money. When a supplier invests, part of the cash may come back to it as purchases, and the investor gains leverage over your roadmap. Before you take a strategic check, work through the questions in what happens if your backer competes with your product.

What happens when the founders behind the valuation leave?

The valuation is exposed, because a team-priced round has little else to fall back on. Both of the best-known AGI labs have been tested on this.

At SSI, Sutskever confirmed that co-founder and CEO Daniel Gross left as of June 29, TechCrunch reported in July 2025; CNBC said Meta had poached him during its AI hiring spree. Zuckerberg had reportedly tried to buy the whole company. Sutskever’s public answer: “We are flattered by their attention but are focused on seeing our work through.”

Thinking Machines lost more. TechCrunch reported that co-founders including Lilian Weng and Luke Metz returned to OpenAI; The Next Web wrote that Andrew Tulloch joined Meta in late 2025 and that CTO Barret Zoph walked out in January 2026, with OpenAI announcing his return “within the hour.” The same report noted that Murati holds board voting control on major decisions, which keeps strategy stable when co-founders exit.

Ordinary startups face the same risk at smaller scale. Vesting schedules, a written co-founder agreement and clear decision rights are cheap to set up early and painful to negotiate after someone has a better offer.

How can you tell a closed round from a rumoured one?

Read the verbs. A round is closed when the company or its spokesperson confirms it; everything else is a report. Headlines about AGI startups routinely blur the two, and the Thinking Machines story shows why that matters: TechCrunch first relayed a $1 billion raise, The Next Web later relayed a $5bn to $6bn version, and neither was a done deal.

Wording in the story What it tells you
“officially closed” plus “a company spokesperson told” Closed and confirmed by the company
“in discussions to raise” / “if it is completed” Talks; terms can still change or collapse
“according to the Financial Times” / “did not comment” Reported by one outlet, unconfirmed
“undisclosed investment” plus another outlet’s number The amount is someone else’s estimate
“pre-money valuation” The price before the new cash is added
“a source with knowledge of” Anonymous figure; treat as a range

The Thinking Machines saga has one more lesson: late in 2025 it reportedly sought $50 billion or more, TechCrunch said, and The Next Web wrote that those talks did not close. For a wider method, see how to read venture funding data.

Do AGI startups still need revenue?

Their investors may be patient, but at least one lab is already charging customers. SSI is described by TechCrunch as taking a “straight shot” approach without being distracted by product releases or short-term revenue. Thinking Machines went the other way: in July 2026 it released Inkling, an open-weight model, and TechCrunch said it earns money through usage-based compute fees on its Tinker platform for adapting models to customers’ own data.

A source told TechCrunch that Thinking Machines’ annual revenue run rate is over $100 million. That is an unnamed source, not a filing, but it hints at the pressure that comes with a valuation in the tens of billions. If you ever describe your own product in AGI or superintelligence terms to raise money, read our companion piece on AI washing and the SEC before the deck goes out: a mission is not a capability claim, and regulators treat the two very differently.

The SSI home page, meanwhile, still describes “one goal and one product: a safe superintelligence,” with no product listed for sale.

Frequently Asked Questions

Which AGI startups have raised the most money?

Among the labs in this article, the largest reported single rounds were Thinking Machines Lab’s $2 billion seed (closed July 2025, TechCrunch), Safe Superintelligence’s $2 billion at a $32 billion valuation (reported by the Financial Times in April 2025) and Reflection AI’s $2 billion at $8 billion (The New York Times, via SiliconANGLE, October 2025). Nvidia’s 2026 investment in SSI is undisclosed.

Is Thinking Machines Lab valued at $40 billion?

Not as a closed deal. In September 2026, TechCrunch relayed a report by The Information that the company was in discussions to raise at a valuation of at least $40 billion. The Next Web’s September 2026 summary of the updated report said the $40bn figure was pre-money and that none of it was signed. Its only closed round in this article is the July 2025 seed at $12 billion.

How much did Nvidia invest in Safe Superintelligence?

SSI did not disclose the amount. TechCrunch reported in July 2026 that a source described the investment as stretching into multiple billions, and that Bloomberg reported a deal size of $5 billion. Treat both as reported figures, not confirmed ones.

What is the difference between pre-money and post-money valuation?

Pre-money valuation is what investors say the company is worth before the new cash arrives; post-money adds the new round on top. Two reports about the same deal can quote very different headline numbers simply because one uses pre-money and the other post-money, so check which one a story means.

Can a startup raise a seed round with no product?

The AGI labs above did, but they are outliers whose founders already ran research or products at OpenAI, Google DeepMind, Meta or Salesforce. Their reported rounds priced the team and access to computing power. Most seed investors still ask an ordinary startup for evidence of a problem, early users or traction.

Sources

This article is general information for founders, not legal, investment or securities advice. Funding figures and company statements are as reported by the sources cited on the dates shown and can change.