Layer 2.5 · head to head
Fly.io vs Modal
Inference next to your app, or inference as its own platform.
| Fly.io | Modal | |
|---|---|---|
| Can you leave | Movable with effort. Expect to redo images, storage wiring and networking. | Sticky. Leaving means rewriting against a different interface, not changing a hostname. |
| Counterparty | The GPU product is an extension of an established application platform. | Low financially; high in code. Modal's programming model is genuinely delightful and genuinely non-portable — that is the trade. |
| What it is | Purpose-built GPU cloud. Owns or leases its own datacentre capacity and sells it directly. | You never see a GPU. You send a request and pay per token or per second of execution. |
| How you reach it | You push a container image and it runs. No cluster to operate. | You supply code; scaling and idle time are the provider's problem. |
| Accelerators | A100, L40S | H100, A100, L40S, T4, and others |
| Regions | Global edge | US, EU |
| Pricing model | Per second, scale to zero. | Per second of GPU time, scale to zero, no idle charge. |
| Getting started | Self-serve. | Self-serve, generous free tier historically. |
| Capacity | Small GPU fleet relative to the CPU platform. | Managed pool. |
| Ownership | Private. | Private. |
Fly.io
For: Teams already running applications on Fly who want inference next to them.
The catch: A short list of parts. This is an adjacency play, not a GPU cloud.
Economics: Value comes from co-location with your app, not from the hourly rate.
Modal
For: Python teams who want infrastructure to disappear and are happy writing to a framework to get it.
The catch: Your deployment becomes Modal-shaped. Migrating off is a rewrite, not a redeploy. Price that in on day one, not year two.
Economics: Per-second billing with no idle cost is the cheapest possible shape for bursty inference. It is the wrong shape for a job that runs continuously.
Neither table row is a price
Deliberately. Published on-demand rates at this layer move weekly, and essentially nobody signing a real contract pays them — every serious buyer pays less than every list figure either of these companies publishes. Quoting one here would date this page within a month.
The GPU rental price index carries dated, sourced figures instead, and the durable finding there is the spread: the identical H100 rents from roughly $1.38 to $12.29 an hour depending only on who you rent it from.
The layers underneath both
Whichever you pick is renting you chips in a building that needs power. In 2026 that is the constraint that binds: Microsoft has disclosed an Azure backlog it cannot fill for want of megawatts rather than accelerators, and the US interconnection queue exceeds 2,600 GW with roughly 80% of projects withdrawing before they energise.
Layer 2 — Silicon · Layer 1 — Energy · The interconnection queue