Fal raises $1.3B at a $16B valuation after demand outran its compute
The generative media inference startup says demand forced it to rate-limit customers this summer. Bessemer led the round, with Nvidia, Sound Ventures, HOF Capital and T. Rowe Price joining.
Fal, the San Francisco startup that runs image, video and audio generation models for developers, has raised a $1.3 billion Series F led by Bessemer Venture Partners. The round values the company at $16 billion, double the $8 billion it was worth in April, TechCrunch reported, citing Bloomberg.
Nvidia, Sound Ventures, HOF Capital and T. Rowe Price also took part, and Bessemer partner Talia Goldberg is joining the board of Features & Labels Inc., the company's formal name, according to SiliconANGLE. Fal used the same announcement to disclose a Series E led by Sequoia Capital that closed earlier in the year. Bloomberg put that round at $350 million at the $8 billion valuation, and fal said the two rounds total $1.65 billion.
A raise driven by a compute shortage
Co-founder and chief executive Burkay Gur said the size of the round comes down to compute. "We hit a point in summer where we were not able to serve our customers and had to rate limit them," he told Bloomberg, as quoted by Investing.com. The company plans to spend the money on more computing capacity and on growing its 185-person team.
Demand shows up in the revenue figures. Fal passed $800 million in annualized revenue earlier this year, a person familiar with the matter told Bloomberg. In December 2025, when fal raised a $140 million Series D led by Sequoia at a $4.5 billion valuation, Bloomberg reported that it had passed $200 million in revenue as of that October. The valuation has now risen from $4.5 billion to $16 billion in under a year, and SiliconANGLE puts total investment in the company above $1.9 billion.
Why media inference is its own business
Inference is the work of running a model that has already been trained to produce outputs. Bessemer's investment note argues that media is a different serving problem from text. Language models produce one token at a time, while image and video models refine a whole frame over many passes, which changes how memory, compute and latency have to be handled. Bessemer says fal has worked on that problem all the way down to the kernel.
The business has two sides. Developers reach thousands of models through a single API, and model labs use fal for distribution and serverless hosting. Bessemer lists Adobe, Canva, Shopify, World Labs and Amazon among the companies that have moved production volume onto the platform, and fal says entertainment studios, ad agencies, e-commerce brands and game developers run its models inside the pipelines that deliver their finished work.
Over the past year fal has also started improving the models it serves. Its research team post-trains open weights from other labs and tunes the serving stack alongside them. In August it released H3 Max, built on MiniMax's H3 weights, and in October it launched FLUX 3 Max with Black Forest Labs. In fal's own benchmark, FLUX 3 Max generated a 10-second clip in 12.45 seconds, against 277.80 seconds for Seedance 2.5, and fal says it scored 47 Elo points above the original FLUX 3 in Artificial Analysis's image-to-video ranking. Some of these partnerships carry licensing fees or revenue-sharing terms, Bloomberg reported. Gur said fal has no near-term plan to pretrain its own model but did not rule it out.
A crowded and capital-hungry field
Fal's new price puts it just behind Fireworks AI, which raised $1.505 billion at a $17.5 billion valuation in July and said it had passed $1 billion in annualized revenue. Baseten and Modal have also raised large rounds at rising valuations. Growth has not solved the cost side, though. Several people familiar with inference companies' finances told TechCrunch that gross margins across the sector remain thin, because buying or leasing GPUs is expensive.
That makes chips part of the story. Nvidia, which fal describes as a strategic investor, also backed Fireworks, and its growing reach across the AI stack is drawing scrutiny, as in the Justice Department's questions about Nvidia's Hugging Face deal. Specialist chipmakers are raising money to compete for the same inference workloads, including FuriosaAI, which raised $620 million to double its output. Labs are also pushing generated media toward interactive formats, such as Google DeepMind's Genie 4 world model.
Fal was founded in 2021 by Gur, who previously led machine learning work at Coinbase and was an engineer at Oracle, and chief technology officer Gorkem Yurtseven, a former Amazon developer. Fal says the new money funds its work across 2027.
