GPUs on Fly.io are available to everyone!

Source: fly.io Date: 2025-02-12 URL: https://fly.io/blog/gpu-ga/

Summary

Release announcement for general availability of GPU instances on Fly.io, open to all users. Available GPUs: A100 (40GB and 80GB) and L40S (48GB), provisioned by setting vm.size in app configuration. Target workloads: LLM inference, image generation, text transcription. A10 GPUs listed as coming soon. New accounts may require additional verification.

Implications

GPU market thread / edge deployment economics. This GA announcement predates the “We Were Wrong About GPUs” postmortem by ~2 months, and the L40S price cut by a similar window — making it a useful baseline for understanding how quickly Fly’s GPU strategy pivoted. At GA, the pitch was broad (“unlock the power of LLMs, image generation, transcription”); by the postmortem, the honest assessment was that the GPU product served a narrow use case. The A10 mention is interesting — it would become their most popular GPU, which contradicted their expectation that developers would want more powerful options. The GA moment is the high-water mark of Fly’s GPU ambitions before the strategic reset.