πŸ“° Key Takeaways

Fal is a San Francisco-based startup that hosts and runs AI video, image, and audio models. It just raised $1.3 billion in its latest round, led by Bessemer Venture Partners, with Nvidia, Sound Ventures, and T. Rowe Price participating. According to Bloomberg, the round values Fal at $16 billion, twice the $8 billion valuation it had in April. The company also announced both a Series E and a Series F, which together raised $1.65 billion, with the earlier round led by Sequoia.

The backdrop for this Series F is a sharp rise in demand for inference services. Inference means running a trained AI model so it actually works and produces results, and demand is especially strong from customers who rely on open-source models. Bloomberg reports that Fal’s annualized revenue topped $800 million earlier this year.

Other inference providers like Baseten, Fireworks AI, and Modal are also growing fast and raising big rounds at higher valuations. But several people familiar with the finances of inference companies say gross margins in this business are still low, because buying or renting compute is so expensive.

Fal was founded in 2021 by Burkay Gur, a former head of machine learning at Coinbase who also worked as an engineer at Oracle, and Gorkem Yurtseven, who previously worked as a developer at Amazon.

πŸ’¬ JudyAI Lab Take

Fal raised $1.3 billion at a $16 billion valuation, double the $8 billion from April. With annualized revenue above $800 million, it looks like hosting and running inference for AI models is moving from a supporting role to a main event that can soak up big checks on its own.

This story points to a clear division of labor. Models themselves keep getting more varied, demand from customers using open-source models is especially strong, and the inference layer that actually runs those models and delivers stable results has become its own business. Baseten, Fireworks AI, and Modal are also growing fast at higher valuations, so this isn’t just one company’s story. The original article also notes that several people familiar with the finances say gross margins at these companies are still low, because buying or renting compute costs so much. For AI builders, that means demand is hot, but the cost structure of inference isn’t easy: revenue scale and profit margin are two different things. When we pick which layer of service to depend on, we should look at both its growth and its cost pressure.

A good next step is to audit your own project: write down how much you spend on inference services each month, and how hard it would be to switch to another provider.

πŸ“… Source Info

πŸ”— Further Reading