Railway raises $100M to build data centers and capitalize on 'agent speed' software development

Summary

Railway has closed a $100 million funding round to build out data centers and double down on its platform thesis that software deployment should require near-zero friction. Jake Cooper, co-founder and CEO, frames the raise around a single structural shift: the definition of a developer is dissolving, and Railway wants to be the infrastructure layer underneath whatever replaces it.

Cooper argues the productivity multiplier from AI-assisted coding is not 10x but closer to 10,000x, a figure he grounds in his own workflow. He describes writing 5,000 lines of code in a single day, between meetings, while spending most of that time reviewing logic and architecture rather than syntax. His internal label for this pace is 'agent speed,' and he has reorganized Railway's teams around it, including putting recruiting and operations staff on Claude Code subscriptions.

The Build-vs-Buy Calculus Is Shifting

Cooper's position is that the economic case for buying off-the-shelf software is weakening fast. Purpose-built internal tooling, historically expensive and slow to produce, is now trivial to assemble. He singles out ERP software as a category ripe for disruption, arguing that organizational structures which previously required scale to justify can now be replicated earlier and customized more deeply. Flexible platforms like Notion still have a role, but the threshold for when it makes sense to build proprietary tooling has dropped sharply.

He also walked back a prior public take. Cooper had posted that engineers would become product managers faster than PMs could become engineers. He now considers that wrong. His revised view is that three capabilities survive full agent automation: taste, agency, and structure. PMs hold much of that context already, and the engineering credential becomes less relevant as the bottleneck shifts from code generation to architectural judgment.

Data Centers as a Moat

The capital raise is not purely software-focused. Railway is building physical data centers and has developed internal tooling to virtualize rack configuration, allowing drag-and-drop server layout. The longer-term vision is a one-click checkout for data center procurement, integrating directly with the Super Micro API to auto-generate purchase orders. Cooper cited a rough illustrative figure of $25 million as the kind of order that workflow could eventually process programmatically.

On database portability, Cooper is skeptical that AI agents meaningfully reduce switching costs. Storage and data persistence remain hard physical constraints regardless of how capable the tooling layer becomes. Railway started as a one-click Postgres and Redis provisioning service and has since expanded up the stack toward web apps and CDN capabilities, with the data layer treated as the foundation rather than a commodity.

The product roadmap converges on a Terraform-style model for full-stack infrastructure, where intended state and actual state are continuously reconciled, environments can be forked instantly, and changes merge into production without manual intervention. Cooper describes that as the operational definition of building at agent speed.