From $8 billion to $16 billion in seven months: fal’s financial ascent

Two funding rounds totaling $1.65 billion bring fal’s valuation to $16 billion. The platform wants to speed up image and video generation.

Sixteen billion dollars in valuation, versus eight billion seven months earlier. The U.S. platform fal, specialized in running generative image, video, and audio models, has just closed a $1.3 billion fundraising round. This Series F, led by Bessemer Venture Partners, follows a $350 million financing completed in April under the leadership of Sequoia Capital. The two transactions represent $1.65 billion and place the company among the best-funded players in generative artificial intelligence infrastructure.

The amounts and valuations were reported by Bloomberg and relayed by TechCrunch. In its official announcement, the company confirms the two funding rounds, without detailing the financial terms. NVIDIA is participating in the Series F as a strategic investor, alongside Sound Ventures, HOF Capital, and T. Rowe Price, according to Bloomberg. This new capital injection is primarily intended to strengthen fal’s computing capacity and support its research work on generative models. Growth running up against compute capacity

Founded in 2021 by Burkay Gur and Gorkem Yurtseven, fal developed around infrastructure that allows developers and companies to run generative models without directly managing the servers required. Its platform brings together thousands of models accessible through a common programming interface, with usage-based billing.

This activity relies on inference, in other words the execution of an already trained model to produce an output. Unlike language models, which generate tokens progressively, visual creation systems often require multiple computing steps to produce or modify an image or a video sequence. Their memory, processing power, and execution time needs can vary considerably depending on the models and the requested formats.

The rapid expansion of these uses therefore requires infrastructure capable of absorbing large volumes of requests. Burkay Gur explained to Bloomberg that the company had to limit certain uses over the summer, due to insufficient capacity to meet demand. The financing is meant precisely to help acquire more computing resources and strengthen a team of around 185 employees.

According to the same information, fal’s annualized revenue would have exceeded $800 million over the year. This figure corresponds to an extrapolation of the observed revenue pace over a given period, not to actual annual revenue collected. It is also distinct from profitability, in a business where buying or renting computing power represents a major expense. From hosting to improving models

fal’s strategy is evolving in parallel with its infrastructure. Long focused on distributing and running models developed by other labs, the company is now working directly on post-training improvements, in collaboration with their creators.

This evolution took shape with H3 Max, an adapted version of MiniMax’s video model, then with FLUX 3 Max, developed with Black Forest Labs. The goal is to jointly improve both the quality of the results and the speed at which they are produced, rather than addressing those two dimensions separately.

For FLUX 3 Max, figures published by fal show a ten-second video generation completed in 12.45 seconds, compared with 126.44 seconds for FLUX 3 in the benchmark used. The model also achieves a score of 1,139 Elo in a preliminary image-to-video generation evaluation, compared with 1,089 for its original version. These results illustrate the gains claimed by the provider, without establishing a general superiority across all formats, hardware, and production conditions.

This work brings fal closer to the labs that develop the models it distributes. In its investment analysis, Bessemer says companies can now commission specially adapted versions of models, available exclusively on fal’s infrastructure. The terms of ownership, portability, and commercialization of these customized versions are not, however, detailed. Infrastructure already used in production pipelines

Professional customers are another area of development. Entertainment studios, advertising agencies, e-commerce platforms, and video game publishers are gradually integrating image, video, and audio generation into their production workflows.

The company had already announced in May a strategic partnership with Amazon Web Services to support this growth. It then claimed more than 2.5 million developers using its platform and cited Amazon MGM Studios, Canva, and Adobe among the companies running production workloads on its infrastructure. The AWS deal is aimed in particular at improving the availability of computing resources and the ability to handle demand fluctuations.

For client companies, the challenge is not limited to choosing a high-performing model. A single production may require several successive systems to generate, modify, enhance, or assemble different media. Access to these models from a shared infrastructure reduces integration work, but it also creates dependence on the provider’s availability, pricing, and technical conditions.

This positioning puts fal up against other inference specialists, including Fireworks AI, Baseten, and Modal, as well as the major cloud providers. Competition is focused as much on the models available as on execution costs, response times, and the ability to guarantee stable service at scale. Investments aimed at 2027

The financing will make it possible to pursue two costly activities simultaneously: building infrastructure capable of serving more requests and investing in research to improve the models offered. According to Bessemer, which had already backed the company since its Series B, this combination is now at the heart of its positioning.

The $16 billion valuation marks a rapid increase from the $4.5 billion recorded in the December 2025 Series D and the $8 billion in April 2026. It reflects investors’ expectations for generative AI growth in creative industries, without prejudging the company’s future margins.

The official announcement does not specify