Fal was worth $8 billion in April. It’s now worth $16 billion.
The San Francisco startup, which hosts and runs AI video, image, and audio models, raised $1.3 billion in a Series F led by Bessemer Venture Partners, TechCrunch AI reports, citing Bloomberg. Nvidia, Sound Ventures, and T. Rowe Price also joined the round.
The quick read
- The deal: $1.3 billion at a $16 billion valuation, double the April price.
- The total: Fal announced its Series E and F together, with $1.65 billion raised across both. Sequoia led the Series E, which works out to about $350 million by simple subtraction.
- The revenue: Fal passed $800 million in annualized revenue earlier this year, according to Bloomberg.
- The catch: Inference companies still run on thin gross margins because compute costs so much.
- My read: Demand looks strong. Profitability is the open question.
What Fal actually sells
Inference is the process of running an AI model that’s already been trained. Think of training as building the engine and inference as burning fuel every time someone hits the gas. Each time an app turns a prompt into a video clip, that’s inference, and the bill keeps coming as long as people use the product.
TechCrunch AI says demand is coming especially from customers that rely on open source models. That makes sense. An open model is free to download, but someone still has to host it and keep it running, and many teams would rather pay for that than manage their own GPUs.
Why this matters
What stands out is how fast investors are repricing this part of the AI stack. At $16 billion, Fal is valued at about 20 times that $800 million revenue figure. That’s my math, and the revenue number dates from earlier this year, so the real multiple may be lower today.
Fal isn’t alone either. Baseten, Fireworks AI, and Modal have also been growing quickly and raising large rounds at higher valuations. Frontier labs get most of the headlines, but a growing share of venture money is now chasing the companies that keep models running in production.
Fal has also been at this for a while. Burkay Gur, a former Coinbase machine learning leader and Oracle engineer, co-founded it in 2021 with Gorkem Yurtseven, who was previously a developer at Amazon. That’s more than a year before ChatGPT set off the current boom.
The margin problem
This is the part I’d watch most closely. According to several people familiar with inference companies’ financials, gross margins remain thin because buying or leasing compute is so expensive. The report doesn’t break out Fal’s own margins, so we can’t say where it lands in that group.
Still, an $800 million run rate tells you what customers pay. It doesn’t tell you what Fal keeps. And Nvidia’s spot on the cap table is worth noting, since it sells the kind of chips this whole category depends on. It gets to back the customers while also benefiting from the compute boom squeezing them.
What to expect next
- If you build on open models: Expect well-funded providers to fight for your business. Keep your setup portable, because thin margins mean today’s prices may not hold in either direction.
- If you run an AI startup: Check your inference costs against revenue now, before usage scales. Your provider’s economics can quickly become yours.
- If you track AI funding: Look past revenue growth and ask about gross margin. That number will separate the durable inference companies from the ones renting growth.
Fal now has fresh capital and a valuation that assumes this demand keeps climbing. The next test is whether it can turn that demand into healthy margins while compute stays this expensive. TechCrunch AI has more details in its original report.