Railway vs Fly.io — and three alternatives worth pricing

The products sell different units

Railway meters CPU and memory usage, then adds storage and public egress. Fly.io sells Machines and volumes by provisioned time, with started and stopped states priced differently. Render mixes instance plans with workspace and network rules. Cloud Run meters active container time and requests. Cloudflare Workers meters requests and CPU time; Containers add active vCPU, memory, disk, and egress.

That means a single monthly total can mislead. A quiet application favors scale-to-zero. A latency-sensitive application with a warm minimum pays for idle capacity. A media-heavy application may spend more on transfer than compute.^{[1][3][5][7][8]}

One workload, three stages

The chart uses the traffic envelope above, a cacheable response mix, 10 GB of persistent storage after the MVP stage, and no database. The MVP line deliberately uses available free or minimum plans; production readiness is not implied.

Cloudflare is inexpensive only when the workload fits Workers. Moving an arbitrary long-running container to Workers is an architecture change, not a pricing toggle. Cloudflare Containers are now generally available, but their active-resource model belongs in a separate calculation.^[8][9]

The forgotten-resource test

Deleting the web application does not necessarily delete its bill. Fly.io explicitly warns that volumes bill while Machines are stopped and managed Postgres, Upstash, and Tigris services can survive app deletion. A dedicated IPv4 address and snapshots are separate meters.^[4]

Railway’s common trap is topology: a database connection through its public URL creates egress, while private networking avoids it. Cloud Run source deployment can create Cloud Build and Artifact Registry charges outside the Cloud Run line item.^[2][7]

Render’s free database expires, free services sleep after inactivity, and the presence of a payment method changes whether excess bandwidth or build usage becomes downtime or a charge.^[6]

Which one should a solo founder choose?

Choose Railway when the repository-to-service mental model and a readable bill matter more than squeezing every infrastructure unit. Choose Fly.io when regions, Machines, private networking, and low-level control are product requirements and you accept more operational ownership.

Choose Render when you want a conventional PaaS and can live with plan-gated observability. Choose Cloud Run for bursty containerized APIs when a Google Cloud billing account and adjacent services are acceptable. Choose Cloudflare Workers when the application is naturally request-driven and edge-compatible. A zero egress line alone is not enough.

• Before revenue: optimize for a safe failure mode and an exportable repository.
• At first revenue: pay for warm latency, backups, and enough telemetry to explain a support ticket.
• Before a launch: estimate egress separately and rehearse the provider’s spend cap or maximum-scale behavior.
• After the launch: compare measured resource-hours with the envelope and revise the model instead of changing provider from one invoice.