Fal's new funding round tells me the market has started to price AI inference the way I price hotels and buildings: by capacity, not by code. A company that had to turn customers away this summer just raised $1.3 billion mostly to buy more computing power, and investors valued it at about 20 times its revenue. That is less than half the multiple paid for a legal AI application last month. The gap is the real story. AI money is splitting into two different kinds of asset, and they should be underwritten differently.
Fal is a San Francisco company founded in 2021 by Burkay Gur and Gorkem Yurtseven. It runs a cloud platform where developers call image, video and audio generation models through a single API, and studios, ad agencies and game developers use it in production. On November 3 it announced $1.3 billion in new funding at a $16 billion valuation, as reported by SiliconANGLE, citing Bloomberg. Bessemer Venture Partners led, with Nvidia, T. Rowe Price, Sound Ventures and HOF Capital joining. The valuation was $8 billion in April, when Sequoia led a $350 million round, and $4.5 billion in December 2025. Investors have now put more than $1.9 billion into the company.
Annualized revenue passed $800 million earlier this year, according to a person familiar with the matter, and the company has 185 employees. The new money goes to compute and hiring. Gur said demand outran capacity over the summer and the company was "not able to serve our customers," so it rate-limited them. It is not alone: Fireworks AI raised $1.5 billion at $17.5 billion in July, and Modal raised $750 million at $15.75 billion in October. People who know the sector's finances say gross margins stay thin because compute is expensive to buy or lease.
Here is the number that matters to me. Divide Fal's $16 billion valuation by its $800 million of annualized revenue and you get 20 times. When Legora raised in October, it was valued at $8.8 billion on a little over $200 million of recurring revenue, which works out to roughly 44 times. So investors are paying about 2.2 times as much for each dollar of legal application revenue as for each dollar of inference revenue underneath it. The raise itself tells the same story: $1.3 billion divided by $800 million is about 1.6 years of current revenue, spent largely on capacity. That is the profile of a capital-heavy business, not a software one.
I recognize the pattern from hotels. At EMC Acquisitions, when a property is full and turning guests away, it does not have a demand problem. It has a keys problem, and the fix costs capital: you add rooms or buy the next building. Revenue is capped by capacity, and the return depends on what you paid for that capacity. Fal's summer reads exactly like a sold-out sign. The questions I would ask are the ones we ask on any hotel deal. What does each unit of capacity cost, how long is it committed, and what happens to rates when competitors open rooms across the street? With Fireworks, Modal and Fal all raising large rounds since July, a lot of new rooms are being built at the same time.
None of this makes Fal a bad investment. Gur's view that scale wins is probably right for this layer, which means the winners will be the companies with the cheapest and most reliable access to compute. Having Nvidia on the cap table may matter more than any product feature. But investors should price these companies as infrastructure and expect infrastructure returns. Through ELP Ventures, I would rather back founders building on top of these platforms, where every drop in the price of inference becomes their margin. When a city fills up with new hotel rooms, the travelers usually benefit before the owners do.