A lot of the money in AI has gone to the companies that train models. Fal is a reminder that there is a fortune in simply running them. The San Francisco startup, formally Features & Labels Inc, has raised $1.3 billion in a Series F led by Bessemer Venture Partners, with Nvidia, Sound Ventures, and T. Rowe Price joining in. The round values Fal at $16 billion, according to Bloomberg's reporting as relayed by TechCrunch, double the $8 billion it was worth in April.
Fal does one thing: it hosts and runs AI image, video, and audio models so that other companies do not have to stand up their own fleet of accelerators. That is the unglamorous plumbing of generative media, and business is booming. The company passed an annualized $800 million in revenue earlier this year, and it says the new capital will go toward more inference capacity and a larger research team.
The round that keeps raising
This is not Fal's first big cheque this year. It disclosed a prior Series E led by Sequoia, and together the two rounds brought in roughly $1.65 billion. The pace tells you something about where demand is heading. Much of it comes from customers running open-weight models, which shift the cost from training to serving and land that bill squarely on companies like Fal.
The valuation also puts Fal in close company with its rivals rather than out front. Fireworks AI raised $1.5 billion at a $17.5 billion valuation over the summer, which makes inference one of the few corners of this market with more than one well-funded contender and real competition on price.
A bet on serving, not training
What makes Fal interesting is the shape of the wager. Training a frontier model is a race a handful of labs can afford to run. Serving the output is a utility business, closer to a toll road, and it grows every time someone generates another clip or image regardless of which lab built the model underneath. Investors are paying a 20-times-revenue multiple on the idea that the toll road only gets busier.
Commentarii · 0