Fly.io to Discontinue GPU-Accelerated Machines by August 1, 2026
Key Takeaways
- ▸Fly.io is sunsetting GPU services effective August 1, 2026—a deadline for all GPU-based workloads
- ▸Current offerings included NVIDIA A10, L40S, A100 40GB, and A100 80GB GPUs, marketed for inference, LLM deployment, video encoding, and graphics acceleration
- ▸GPUs were particularly popular for running smaller models like Llama 2 and Stable Diffusion without high infrastructure costs
Source:
Summary
Fly.io has announced the deprecation of its GPU-accelerated machines, with service ending on August 1, 2026. The platform currently offers four GPU models—NVIDIA A10, L40S, A100 40G PCIe, and A100 80GB SXM—deployed as single-GPU machines for inference, encoding/decoding, graphics rendering, and smaller model fine-tuning. The GPUs have been available in select regions including iad, sjc, syd, and ams. The discontinuation represents a significant shift in Fly.io's infrastructure strategy, affecting developers running AI inference, generative AI, and machine learning workloads on the platform. Users have until August 1 to migrate their workloads to alternative GPU hosting providers.
- Developers must migrate existing GPU workloads to alternative platforms before the deadline
