Skip to content

Railway: 3M Users, Own Data Centers, and a $200K Monthly AI Agent Bill

Jake Cooper's Railway is processing 100K signups per week and spending over $200K on coding agents monthly as it builds what it calls the agent-native cloud.

Railway hit 3 million users and is adding 100K signups per week. Founder Jake Cooper — previously at Bloomberg and Uber — laid out the numbers and the thesis on the Latent Space podcast: the next cloud platform won’t be built for humans clicking dashboards. It’ll be built for coding agents deploying code.

The concrete details from the conversation:

  • 100K signups/week — Railway’s current growth rate, driven largely by developers who want push-code-get-a-URL simplicity
  • $200K+ monthly spend on coding agents — Railway itself is using AI agents heavily for its own development, eating its own dogfood at scale
  • Own bare-metal data centers — the company moved the vast majority of workloads off public cloud onto its own metal, a bet on margins and control
  • “Death of PRs” — Cooper’s framing for where agent-driven development leads: agents ship directly to preview environments, not pull requests
  • Near-zero activation energy — the core design principle since day one: push code, get a URL, iterate

Cooper’s argument is specific. Agents don’t want to configure VPCs, set up load balancers, or navigate AWS’s 200+ services. They want an API that takes code and returns a running deployment. Railway’s architecture — instant deploys, ephemeral environments, programmatic control — was accidentally built for this before the agent wave hit.

The $200K/month agent spend number is the most telling data point. Railway isn’t just positioning as agent-friendly infrastructure — it’s stress-testing the workflow internally. Cooper rebuilt Railway’s network overlay over a single weekend, which says something about the velocity agents enable when the deployment layer cooperates.

NOTE

Cooper’s “death of PRs” claim isn’t about eliminating code review. It’s about agents deploying to preview environments directly, where the review happens against a running app rather than a diff.

Railway is betting that the cloud platform market resets when the primary deployer isn’t a human. At 3M users and growing 100K/week, the bet has traction — whether the “agent-native” framing holds up depends on whether agents actually prefer Railway’s API over raw Terraform. The full conversation is on Latent Space.