Hello everyone,
Today I received an email confirming that Fly.io GPUs will be fully deprecated as of July 31, 2026. While this was foreshadowed in last year’s “We Were Wrong About GPUs” blog post, getting the official notice still stings.
I’ve been building my document processing pipeline around L40S machines — OCR, layout analysis, and similar workloads. Fly.io was chosen specifically because it was the easiest platform to build on: fly deploy, pick your region, done.
But having zero GPU options going forward feels like a dealbreaker for workloads like mine. I understand GPUs aren’t a cash cow feature — but they’re still a feature you expect from a hosting provider. It’s like a gym in a hotel — not every guest will use it, but the ones who need it will book a different hotel.
A couple of questions for the team and the community:
-
What’s the vision for customers currently relying on GPUs? Is there a recommended migration path or a partner integration planned? Should we be looking at splitting our architecture across providers?
-
Will there be absolutely no GPU option after August 1? Not even a single tier like L40S for inference workloads?
-
For those of you in a similar boat — what’s your plan? Considering Modal, RunPod, GCP Cloud Run GPUs? Would love to hear what others are thinking.
The discounted pricing through the sunset period is appreciated, but I’d much rather have a path forward on the platform I’ve already invested in.
