Railway

A kid who modded games built the cloud AWS never bothered to.

Jake Cooper was 13 when he first opened a code editor - not to build a startup, not to impress anyone, but to mod first-person shooter games with C++. That particular instinct - just make the thing work the way it should - turned out to be a pretty good foundation for a cloud company.

By the time he was at the University of Victoria studying engineering, Cooper had already worked remotely for Wolfram Alpha, tangled with Docker and CI pipelines, interned at IBM and Hootsuite, and gone on to Bloomberg and Uber. When Uber sold its Jump bike division to Lime and Cooper found himself suddenly unattached, he started paying attention to a problem he'd tripped over everywhere he'd worked: deploying software was still absurdly painful.

You could be a brilliant engineer - truly exceptional at building things - and still spend half your time wrestling with Kubernetes, Terraform, IAM policies, VPCs, load balancer configs, and billing dashboards that seemed designed to confuse. The cloud had become infrastructure theater. Cooper's insight was almost embarrassingly simple: what if it just... wasn't?

Railway launched in June 2020 as a Platform-as-a-Service that handles the full deployment lifecycle with minimal fuss. Connect your GitHub repo, push code, Railway figures out the rest. Databases spin up instantly. Environments are unlimited. Rollbacks are instant. You can SSH into a running container. Bills scale down to near zero when nothing is running. There's no infrastructure tax on your attention.

The company never ran ads. Never cold-called prospects. Never did a launch campaign. Its first five years of growth came entirely from developers telling other developers - a $200,000-per-month influx of new users that no marketing team was responsible for and no CMO could take credit for.

Deploy anything. Manage nothing.

Railway is a PaaS - but that word undersells it. Think of it as the layer between "I have code" and "my code is running in production, scaled, monitored, and accessible to the world." All the parts developers used to manage manually - build configuration, container orchestration, database provisioning, networking, observability - Railway either automates or makes trivially easy.

Pricing is usage-based, billed per second of actual CPU and memory consumption. The Hobby tier is $5/month (with $5 of usage credit included). Pro is $20/month. Enterprise is custom. The math is deliberately hard to game - if you're not using resources, you're not paying for them.

They built their own data centers. In six months.

In January 2024, the Railway team made a decision that most startups would never seriously consider: exit Google Cloud entirely and build their own co-location infrastructure from scratch. The project became known internally as Railway Metal.

The stated reasons were about economics and control - owning hardware means better margins and the ability to guarantee performance. The unstated reason may be more interesting: Cooper had clearly grown impatient with the idea that running great infrastructure required renting it from someone else's cloud at a permanent premium.

"Closer to building a house than deploying a Terraform stack."

The Railway Metal blog post - a candid, technical blow-by-blow of constructing a real data center - became something of a minor classic in engineering circles. The team documented everything: the wiring inconsistencies, the infrastructure surprises, the sheer foreignness of physical hardware to a team that had only ever deployed software. By September 2024, nine months after the project began, the first Railway customers were running on Railway Metal. The entire cloud platform now runs on infrastructure Railway owns.

They built a tool. Then made it obsolete. On purpose.

Railway's open-source build tool Nixpacks - a system that automatically detects what language you're using and builds a Docker container for it - became unexpectedly popular. Competitors started using it. Other platforms shipped Nixpacks support as a feature. It accumulated over 3,500 GitHub stars and developed an ecosystem outside Railway's own platform.

Then Railway deprecated it and built Railpack instead. The move wasn't disrespect toward Nixpacks - it was a recognition that the constraints built into Nixpacks at the beginning had accumulated. Railpack delivers builds that are 38-77% smaller with meaningfully better caching and granular versioning. The open-source community retained Nixpacks; Railway moved forward.

This has become something of a pattern for Railway: build something in the open, let the community adopt it, and then build the next version when the first version's limits become apparent. It's a slow-burn approach to infrastructure that tends to produce better software than the move-fast alternative.

30 people. 14 countries. No office. No salespeople. For five years.

Railway has roughly 30-40 employees distributed across the United States, Japan, UAE, India, Germany, Mexico, Belgium, Spain, Canada, France, the UK, Singapore, Brazil, and Thailand. There is no headquarters, no central office, no mandatory timezone. Employees are actively discouraged from working weekends - Cooper has been explicit about this, which is unusual for a company that otherwise moves as quickly as Railway does.

New hires go through a four-week onboarding process. The company has a 32,000-member Discord that serves as both support channel and community hub. The brand voice - "Ship software peacefully" - is not a tagline invented by a marketing agency. It's a description of what Cooper actually wants the product to feel like.

Until January 2026 and the Series B, Railway had no dedicated sales team, no marketing function, and no formal go-to-market operation. The company described 2026 as "the year we play on the world stage" - meaning the first year they'd actually try to grow the way other companies grow.

$124M raised. The last $100M just arrived.

The angel investor list at the Series A is telling: Tom Preston-Werner (GitHub co-founder) and Guillermo Rauch (Vercel CEO) both wrote checks. These are people who understand infrastructure developer tools in a way that most investors don't, and their early bet on Railway says something about the credibility the company had built within that specific community.

From a game modder's frustration to a $100M cloud company.

From indie hackers to Fortune 500s - all without a salesperson.

Railway's user base is genuinely broad. The Hobby tier ($5/month) is designed for individual developers - people shipping side projects, building prototypes, learning by doing. The Pro tier ($20/month) targets startups and fast-moving teams. Enterprise handles the TripAdvisors, the Cognizants, the Automattics - companies where the question is no longer "can this scale?" but "can we get legal and compliance involved?"

The template marketplace is a quiet but meaningful part of how Railway grows. Anyone can publish a one-click deployment template - for a Postgres-backed Next.js app, a Ghost blog, a Plausible analytics instance, a Supabase alternative - and when someone deploys it, the creator earns 25% of the revenue. Over $1M has been paid out. This is how Railway simultaneously keeps its ecosystem of open-source tools aligned and turns the community into a distributed sales force it never had to hire.

The developer cloud for the AI era - and they mean it.

When Railway calls itself "the world's first intelligent cloud provider," the claim has a specific meaning that goes beyond marketing language. The Railway MCP Server - launched August 2025 - lets AI coding agents like Claude, Cursor, or Windsurf deploy applications, set environment variables, and retrieve logs directly from within a code editor. The developer never leaves their tool. Infrastructure just happens.

The design choice that stands out most is what the MCP Server doesn't do: it has no destructive operations. No delete-x tools. No "nuke this environment." Cooper's team made a deliberate call that AI agents operating on production infrastructure should not have the ability to accidentally remove things. The caution is notable in an industry where "move fast" is still a reflex.

This positions Railway interestingly for the era of agentic AI - a world where a significant fraction of deployments will be initiated not by a human typing a command but by an AI agent executing a plan. Railway is building to be the infrastructure those agents reach for.

Go deeper.