Tranched AI Funding Rounds Push Valuations Past Revenue Reality

One round, two prices

For most of venture capital's history, everyone who bought into a funding round paid the same price per share. That convention is now breaking down among the most sought-after AI companies. Venture firms are more often splitting rounds into tranches. An early slice sells cheaply, then a later slice of the same round sells at a much higher valuation, and that higher number is usually the one that gets announced. Bloomberg reported on October 22 that the practice is spreading. Its report focused on Emulate, a startup founded by Google DeepMind alumni that has drawn competing bids at rising prices even though it is about three months old and has no product or public website.1114

The structure is not new. What has changed is how widely it is used and how large the gaps between tranche prices have become. Taken together, the reporting points to one conclusion: in the hottest part of the AI market, the headline valuation now tells you mainly what the last investor in paid.

The Ineffable benchmark

The clearest documented case is still Ineffable Intelligence, the London lab led by David Silver. Its fundraise was promoted as Europe's largest seed round ever, at $1.1 billion and a $5.1 billion valuation.13 Forbes, citing company filings, found that the round had two tranches. In the first, Sequoia and other investors put in $11 million at a pre-money valuation of roughly $55 million. About a month later, a $1.1 billion tranche followed at a $4 billion pre-money valuation, paid by Lightspeed, Index Ventures, DST Global and Sequoia.13 Forbes said that was more than 70 times the price for the same company within weeks. A person familiar with the deal said most of Sequoia's money went in at the higher price.23

Ineffable also shows how the headline number travels. Unicorn trackers list it as the most valuable startup to cross $1 billion for the first time in 2026, using the $5.1 billion figure.5 They do not show the much lower price that early backers paid.

Emulate and the new cohort of labs

Emulate is the newest case. It was founded in August by Jack Parker-Holder, Matthew McGill and Philip Ball, former DeepMind researchers who worked on the Genie world model. It has been raising a seed round of about $700 million, with later tranches putting its post-money valuation at roughly $3.7 billion to $4 billion.22 The exact figure differs depending on the report. A September report relayed from the Financial Times described talks at about $3.7 billion post-money, with Index and Lightspeed expected to co-lead. Summaries of Bloomberg's October story say only that the valuation has reached "the billions" through a two-step structure.23 Until the round closes and terms are filed, the safest assumption is that any single number for Emulate is the price of its final tranche, not a blended value.

Emulate is part of a wider group of research-led labs raising huge sums before they have any commercial product. Flapping Airplanes reportedly raised a $1 billion Series A at a $6 billion valuation, four times its January seed valuation, with nothing yet on sale. Mirendil, which has not yet opened API access, was reported in September to be in talks to raise up to $1 billion at $5 billion.15 Forbes estimated that these so-called neolabs account for about 16% of the roughly $283 billion invested over the past year in startups other than OpenAI and Anthropic.13

Not just a pre-product phenomenon

It would be easy to treat tranching as a quirk of companies selling pedigree and nothing else, but the coverage shows it reaching businesses with real revenue too. Baseten, an inference-infrastructure company, raised $1.5 billion in two tiers, one at an $11 billion valuation and one at $13 billion.13 It had been valued at $5 billion six months earlier, while its annualized revenue run rate reportedly grew from about $200 million to $600 million.21 Fal was reported in June to be raising $300 million to $350 million in two instalments, with only the higher-priced one carrying the roughly $8 billion headline. Three months earlier it had been valued at $4.5 billion.

The pattern started at smaller sizes. PitchBook data cites Lovable's Series A and Serval's Series B as earlier examples. Serval announced a $1 billion valuation, but Sequoia's lowest entry reportedly valued it at under $400 million.17 Aaru's Series A let lead investor Redpoint put most of its money in at $450 million, alongside a $1 billion price for other investors.

Who benefits

Opinions split most sharply over who gains from these deals. Critics describe it as a way to manufacture headlines. Mercor CEO Brendan Foody said in early June that he had seen about half a dozen rounds in six months where Sequoia invested in two tranches, and called it the "Sequoia Scam."13 Foundation Capital partner Jaya Gupta told Forbes that when fundraising runs on vibes, a billion-dollar headline is worth more than an accurate one.23

Defenders call it a fair reward for taking risk early. SignalFire venture partner Josh Constine argued that a tranched round rewards investors who committed early and add real value. He said top founders now have enough competing capital to set prices themselves.17 There is a real argument here. Early backers get more ownership, and founders get the big raise and the publicity that help them hire and buy computing power.

In our view, both readings are partly right, and they leave out who pays. Amplify partner Sarah Catanzaro has noted that a company's real worth is usually the blended valuation, a weighted average across tranches, and that this figure is usually not reported.25 Employees who join after an announcement may receive option prices close to the higher valuation rather than the blended one.23 The arithmetic also raises the bar for founders: when lead investors get a discount, the rest of the round has to be priced high enough to keep the blended figure where founders want it. The investors exposed to the most risk are the ones in the late tranches, because their returns depend on the company growing into the headline number.15 Limited partners, in turn, hold funds whose holdings are marked at late-tranche prices that nobody revisits until a round fails to close.21

The market backdrop

Tranching is spreading in a funding market with no recent precedent. Crunchbase counted a record $510 billion in global venture funding in the first half of 2026, more than in all of 2025, and OpenAI and Anthropic took 43% of it.4 Anthropic's private valuation rose from $183 billion in September 2025 to $965 billion in May 2026.18 OpenAI is now reportedly close to raising up to $30 billion at a $1.25 trillion pre-money valuation as a bridge to a planned 2027 IPO. That figure is below the $1.4 trillion Bloomberg reported a month earlier.

Counts of new unicorns vary by tracker and date, but they all point the same way. One analysis found that 49 of the 191 companies to reach unicorn status in 2026 were AI firms.3 Other compilations counted close to 90 new unicorns by early July, most of them AI-related.10 Crunchbase found that 19 of the first half's new unicorns raised quick follow-on rounds, often within six months, at double their earlier valuation or more. Etched, for example, went from $5 billion to $10 billion.8 Etched was later reported to be raising at $10.3 billion in a Sequoia-led round while talking with Jane Street at $20 billion.21

The exit market is also pushing these prices up. SpaceX agreed to buy Cursor maker Anysphere in an all-stock deal valuing it at $60 billion, described as the largest acquisition of a venture-backed startup on record. Cursor had been valued at $29.3 billion in November 2025.4 Deals like that give investors a reason to think even a headline price may turn out cheap.

There is one dissenting note on the wider market. One outlet argues that money flooding into lab teams with famous résumés sits alongside a funding squeeze for AI application companies.22 Baseten's and Fal's tiered rounds show the structure reaching infrastructure and application companies as well, which suggests the split runs between in-demand companies and everyone else, not between research labs and products.13

What to watch

The practical lesson is to read AI funding announcements differently. A seed round at $4 billion may contain a much lower price that never appears in the press release. The figures that matter are the blended valuation and the share of each tranche, and they mostly surface in filings, not announcements.25 As long as capital is plentiful and exits like the Cursor deal keep happening, tranching will probably spread further. The test comes when a late-tranche investor has to raise its next round at or below the headline price. Until then, the AI market's best-known valuations should be read as the highest price anyone paid, not a consensus on what the company is worth.