Vercel vs Railway: where should your ai app run?

the full guide

quick answer: for the front end of most ai apps it's Vercel: #2 of 5 on our backends board at 8.5/10, because it's the nicest way going to ship Next.js, and its functions only bill CPU while your code is actually working, not while it's waiting on a model. Railway is #3 (8.0/10) and it's the better pick for anything that has to keep running: an API server, a queue worker, an agent loop, a database next to them.

one thing first: Railway paused new signups after a security incident on oct 21, 2026, and as of its status update in the early hours of oct 27 (ET) they're still paused, with no date to reopen. existing projects kept running. if you're new, you can't sign up today; if you're already on it and signed up before oct 21, Railway recommends rotating the keys you store there.

the short version

ranks and scores come straight from our backend board, and nobody pays to move up it.

why Vercel ranks higher

front ends are its whole job. push to git, get a preview URL, merge, it's live, with instant rollback when you need it. CDN, HTTPS, routing, firewall and DDoS mitigation come on every plan.

waiting on a model is cheap. Vercel's Fluid compute bills active CPU time, and its docs say time spent waiting on i/o, like calling an ai model or a database, doesn't count. that's most of a chat route's life, so streaming LLM responses from functions is a good fit.

the ai plumbing is there. AI Gateway (bring your own key, fallbacks, spend monitoring), Vercel Sandbox for running code your agent wrote, and Vercel Workflows that can pause and resume for minutes to months. pairs nicely with the Vercel AI SDK.

the catch: Hobby is for personal, non-commercial use, so anything with customers needs Pro. functions stop at their max duration (300s default; 800s on Pro), the filesystem is read-only apart from a 500 MB /tmp, and request and response bodies cap at 4.5 MB. Pro is $20 per developer seat, so a team pays per head.

where Railway wins

it runs servers, not just functions. a service is whatever your repo or Docker image builds: a FastAPI app, a Node worker, a cron, a Discord bot, an agent that loops all night. no timeouts to design around, and services in one project share a private network.

the database sits next door. add Postgres from the project canvas and connect with the DATABASE_URL it hands you; there's a pgvector template in the marketplace, volumes for persistent data, backups, and you can convert it to a high availability cluster when you need it.

you pay for what runs. CPU, memory and disk are metered per second, so stopped services cost nothing. Hobby is $5 a month including $5 of usage; Pro is $20 a month per workspace with $20 included and unlimited seats, which is kinder to a team than per-seat pricing.

sandboxes for agents. Railway Sandboxes are isolated Linux VMs for agents, builds and untrusted code, billed on the CPU actually used and memory held, so a sandbox waiting on a model costs almost nothing.

the honest bit: the signup pause above, and the incident behind it: Railway says some customers' encrypted secrets were exposed and asks everyone who signed up before oct 21 to rotate keys stored there. it's also less of a front end platform; you can serve a site from it, but you don't get Vercel's previews, CDN features and framework smarts out of the box.

which should you pick?

• a Next.js app with streaming ai chat routes: Vercel
• marketing site, docs or front end with previews per PR: Vercel
• a Python or Node API that's always on: Railway (if you're already in, or once signups reopen)
• queue workers, crons, long agent runs: Railway
• both, which is common: front end on Vercel, backend and workers on Railway
• you want Postgres, auth, storage and vectors as a service instead: Supabase is #1 (9.0/10). see Neon vs Supabase
• serverless Postgres with a branch per PR: Neon (#4)

the rest of the field

faq

is Railway better than Vercel?

not on our board: Vercel is #2 (8.5/10) and Railway #3 (8.0/10). Railway is the better pick for long-running backends, workers and agents; Vercel for front ends and serverless ai routes.

can i sign up for Railway right now?

not as of oct 27, 2026. Railway paused new signups after its oct 21 security incident and its status page says they remain paused, with no date yet. existing users' services stayed online.

Vercel vs Railway pricing: which is cheaper?

both start at $0 with limits. Vercel Pro is $20 a month per developer seat with $20 of usage included. Railway Hobby is $5 a month with $5 of usage, and Pro is $20 a month per workspace with $20 included and unlimited seats. which bill ends up smaller depends on what you run: per-second CPU and memory on Railway, per-seat plus usage on Vercel.

can Vercel run a long-running backend?

up to a point. functions run 300s by default and up to 800s on Pro (1800s in beta), and Vercel Workflows can pause and resume for much longer. for an always-on server or worker, Railway is the simpler fit.

can i use Vercel for commercial projects on the free plan?

no. Vercel says Hobby is for personal, non-commercial use. commercial projects need Pro.

sources

checked 2026-10-27

• Vercel pricing (Hobby $0 personal and non-commercial, Pro $20 with $20 included credit, $20 developer seat, Fluid compute rates, Sandbox, workflows, AI Gateway, instant rollback, firewall and DDoS): https://vercel.com/pricing
• Vercel Functions limits (300s default, 800s Pro max, 1800s beta, workflows for longer, 4.5 MB body limit, active CPU excludes waiting on i/o): https://vercel.com/docs/functions/limitations
• Vercel Functions runtimes (Node.js, Bun, Python, Go, Rust, Ruby, Wasm, edge, container; read-only filesystem with 500 MB /tmp): https://vercel.com/docs/functions/runtimes
• Railway pricing (free trial, Free, Hobby $5, Pro $20 per workspace with unlimited seats, per-second usage rates, sandboxes): https://railway.com/pricing
• Railway basics (services from a GitHub repo or Docker image, private network per project, volumes, backups): https://docs.railway.com/overview/the-basics
• Railway Postgres guide (one-click Postgres, database_url, high availability, pgvector template): https://docs.railway.com/guides/postgresql
• Railway status (new signups remain paused, no date yet; update oct 27, 03:22 UTC): https://status.railway.com/
• Railway incident report (oct 21 security incident, running services stayed online, rotate secrets stored before oct 21): https://blog.railway.com/p/preliminary-report-october-security-incident
• ranks, scores, verdicts and plans: numptr data/tools.json (backend, database and deploy board)