AI Valuations 2026: Real Cash, Engineered Headlines

Two funding announcements landed this week that look unrelated and tell the same story. Legal AI startup Legora raised $550 million at an $8.8 billion valuation, and the year-old AI research lab Flapping Airplanes raised $1 billion at a $6 billion valuation. Both were reported as clean numbers: one round, one price. Meanwhile, Startup Fortune reported on October 22 that some of the biggest AI rounds of the year were nothing like that, with investors buying into the same legal round at wildly different prices.

In 2026, AI startups raised real money at prices most investors never paid. Ineffable Intelligence's $1.1 billion seed round was announced at a $5.1 billion valuation, but its first $11 million tranche priced the company at $55 million, so late investors paid more than 70 times as much per share as the funds that got in first. The capital is real; the headline numbers are edited.

The mega-round ledger

Start with what is on the record. The nine largest AI financings of 2026 for which numbers are disclosed moved about $12.6 billion of committed capital. The table below takes each round at its announced terms and computes one number founders rarely volunteer: the share of the company sold, which is the raise divided by the headline valuation.

Three things stand out. First, the established businesses sell tiny slivers: Databricks gave up 2.6% of the company for $5 billion, and Legora and Harvey each sold around 3 to 6% while carrying over $200 million in annual recurring revenue. Second, the research labs with no product sell far more of themselves: Flapping Airplanes parted with nearly 17% and Ineffable Intelligence with almost 22%. Third, the two rounds at the bottom of the table are the ones where the headline valuation is hardest to take at face value, because the price was not the same for everyone.

Same round, two prices

The mechanics are straightforward once you see one example, and Forbes documented it in a June investigation that Startup Fortune summarized on October 22. Ineffable Intelligence, the lab founded by longtime Google DeepMind scientist David Silver, raised an initial $11 million tranche from firms including Sequoia Capital at a pre-money valuation of roughly $55 million. Barely a month later, the same round added a follow-on tranche of $1.1 billion at a pre-money valuation of $4 billion, with Lightspeed, Index Ventures and DST Global among the buyers. Forbes noted that the company's implied value would more accurately be represented by a weighted blend of the two tranches, a figure that usually goes unreported.

That weighted blend is easy to compute, and the result is instructive. Weighting each tranche by the capital invested, ($11M x $55M + $1.1B x $4B) divided by $1.111B, gives a blended pre-money value of about $3.96 billion. The round was announced at $5.1 billion post-money and billed as Europe's largest seed round ever. The announcement was not false, but it priced the company at the last tranche's terms, not the average dollar's terms.

Baseten shows the same mechanic at a larger scale. The AI inference infrastructure company closed a $1.5 billion round in two tranches at $11 billion and $13 billion valuations, a jump from the $5 billion valuation it had announced five months earlier. Altimeter Capital, Conviction and Spark Capital led, with Sands Capital and Wellington Management as co-leads, and Baseten said revenue had grown 20 times over the preceding year. The split of the $1.5 billion across the two tranches was not disclosed, so the blended price sits somewhere between $11B and $13B.

The pattern kept appearing through the year. Forbes sources confirmed tranche splits at the synthetic-customer startup Aaru and the IT help-desk startup Serval. On October 22, Bloomberg reported that investors were fighting to buy into Emulate, a three-month-old DeepMind-alumni startup with no website or product, at escalating prices within the same round, with Index Ventures and Lightspeed initially in at around a $200 million valuation before later investors paid more. This is a reported practice at a handful of high-demand companies, not a universal fact about every AI round.

How tranche financing works, and why it inflates headlines

In a traditional venture round, every investor in a Series A or seed round pays the same price per share, which fixes one valuation for the company. A tranched round keeps the legal round open for weeks or months and closes it in stages. Early investors commit first at a baseline price, in exchange for taking the earliest risk and often writing the largest checks. Later investors enter the same round at a higher price, usually because demand for the deal has intensified or the company has hit an interim milestone.

Here is a simplified example to show the arithmetic. Suppose a startup raises $10 million at a $50 million pre-money valuation, then $100 million at a $500 million pre-money valuation, all in one round. The headline announced is $600 million post-money, the second tranche's terms. But the money-weighted blended pre-money value is ($10M x $50M + $100M x $500M) / $110M, or about $459 million. The headline is real in the sense that the last investors really paid it, and it is inflated in the sense that the average dollar in the round bought in far below it.

Why do venture firms want it this way? Startup Fortune's reporting points to two motives. AI labs need unusually large sums for computing infrastructure, and lead investors sought better terms in return for committing capital early. Later investors accepted higher prices because access to the most sought-after AI deals is scarce. As Bryan Yeo, group chief investment officer at Singapore's GIC, put it: there is "a little bit of a hype bubble going on in the early-stage venture space" where "any company startup with an AI label will be valued right up there at huge multiples of whatever the small revenue is."

There is an honest version of this structure, and founders should know where the line is. Tying later tranches to real milestones, like shipped product or signed revenue, is a legitimate way to share risk. The version that inflates headlines is when the price steps up on nothing but elapsed time and deal heat, and only the final price gets reported.

The revenue question

Tranches explain part of the 2026 valuation story; the rest is revenue, or the absence of it. Databricks closed its $5 billion round at $190 billion while disclosing a $7 billion-plus revenue run-rate growing over 80% year over year, per the company's own August 13 announcement. That is a steep 27x multiple, but it is a multiple of something real: the company also said it was adjusted cash-flow positive over the trailing twelve months. Legora's $8.8 billion price tag sits on $200 million-plus of ARR, a 44x multiple, after the company said it went from $1 million to $100 million in ARR in under 18 months and added the next $100 million in under six. Harvey, its closest competitor, raised $550 million at $15.5 billion in September with ARR above $200 million as of March, per TechCrunch.

Contrast that with the research labs. Flapping Airplanes quadrupled its valuation from a January round to $6 billion without a commercial product, per Bloomberg. Ineffable Intelligence had no near-term plans to launch products when its round was announced, per Forbes' reporting. These are bets on teams, mostly ex-DeepMind talent, not on businesses. ByteHerald looked at a similar pricing puzzle this month when investors priced Manus at $4 billion after Meta lost the deal: the number reflected what the winning bidder needed to pay, not what the company was worth.

The broader market context is a split screen. PitchBook-NVCA data compiled this week shows median pre-money valuations for 2026 rising at every stage but seed: Series A up 25% to $60 million, Series B up 34% to $187 million, Series C up 62% to $500 million, and Series D and later up 107% to $1.72 billion, while seed medians slipped 13% to $16 million. AI companies step up 2.2x between rounds against 1.5x for everyone else. In inference chips alone, the median pre-money valuation hit $1.77 billion in 2026 against $194 million in 2024, per PitchBook's October 20 report. Meanwhile, sellers on the secondary market Forge who last raised in 2021 accept a median 59% discount. That is the tension at the heart of these numbers: primary-market prices for AI keep climbing while Gartner warns $3.5 trillion of AI spending may never pay off. Somebody's expectations are wrong.

Three questions for the next headline

Valuation headlines will keep coming, so here is a durable filter, applicable to any mega-round, not just this week's. First, what price did every investor pay? If the round closed in tranches, the headline is the most expensive close. Second, how much of the company was sold? A $1 billion raise at $6 billion means founders and employees gave up nearly 17%; a $5 billion raise at $190 billion cost 2.6%. Dilution tells you how confident the insiders were. Third, what is the money-weighted blend? Weight each tranche by capital invested and you get the number the average dollar actually paid.

Cerebras is the cautionary tale for what happens after the headline. The chipmaker raised a $1 billion Series H at around a $23 billion valuation, went public in May, closed its first day of trading up 68%, and has since fallen more than 61% from that peak, per PitchBook. The private market got its number. The public market is still deciding whether it agrees.

So are 2026 AI valuations real, or engineered? Both. The checks cleared: about $12.6 billion across the nine largest disclosed rounds, led by genuine revenue at Databricks and Legora and genuine scarcity at the research labs. But the two most dramatic numbers of the year, Ineffable's $5.1 billion and Baseten's $13 billion, were set by the last investors in line, not the average one. A valuation is a price, and a price is only as honest as the question of who paid it. When the answer is "not everyone," the headline is an advertisement, not a measurement.