Upstart cloud provider Railway turns heads with speed
Railway began as a PaaS provider but now looks to disrupt hyperscalers with a bespoke back-end hardware design and built-in infrastructure best practices for developers.
A cloud provider says it has won thousands of paying customers from startups to Fortune 500 companies with its own take on efficient hardware design and built-in automation for infrastructure that doesn't include either GPUs or Kubernetes.
Railway, founded as a PaaS provider based on Google Cloud Platform in 2020, moved into its own co-location data centers in 2024. It also departed from its PaaS roots by offering users access to lower-level cloud resources such as VMs, containers and databases, without requiring developers to manage every provisioning, monitoring and security detail, according to founder and CEO Jake Cooper.
Jake Cooper
"We spent the last five years building our own hardware, our own networking layer, orchestration engine, from dashboard to data center," Cooper said. "We've got the standard build / deploy kind of APIs for containers and primitives for people to be able to go in and use [clones], snapshots and things like that at the VM level … [but we also] manage all of the telemetry for [customers with] logs, traces and custom metrics out of the box with a full querying engine."
Railway, which is privately owned, this week revealed it had secured a fresh $100 million Series B funding round. Cooper said the cloud provider has amassed 100,000 paid users at 25,000 businesses, ranging from small and medium businesses to Fortune 500 companies. A Railway press release named companies such as Bilt, Profound, HappyRobot, Intuit’s GoCo, TripAdvisor’s Cruise Critic and MGM Resorts as customers.
Railway competes with cloud hyperscalers and neocloud players, as well as cloud platforms from Render, Northflank and Replit, all of which have similar value propositions for developer ease of use.
However, the depth of Railway's vertical integration, with a custom rack-scale hardware design and a bare-metal orchestration layer that eschews Kubernetes, also means app developers can get services up and running in seconds rather than minutes, Cooper said.
Undercutting big cloud providers' pricing
In addition to better performance, Railway claims its cloud infrastructure costs are lower than those of even large hyperscalers. For example, it prices network egress for application services at $0.05 per GB. By comparison, AWS charges $0.09 per GB for volumes between 100 GB and 10 TB per month; its volume discounts reach $0.05 per GB only with greater than 100 TB transfer from EC2 to the internet per month, according to the AWS pricing page.
Railway's control over its hardware design means that it can pack in application services at high density while charging users by the second for only the underlying resources they actually use, Cooper said. He compared this to services such as AWS Lambda and Google Cloud Run. AWS Lambda charges by the number of requests contained in users' code functions and the duration it takes that code to execute. Google Cloud Run offers variable pricing according to cloud resources used per second, per requests or per job, which also includes network transfer costs.
Railway charges according to the underlying cloud resources used on a per-second basis by users' application services, and its Pro level pricing plan starts at $20 per month. It does not disclose enterprise pricing. However, because of the density of workloads on Railway's hardware, Cooper said, the total cost of its cloud platform is lower over time for users than comparable serverless computing products.
"We've essentially built a version of Cloud Run or Lambda, plus the hardware that goes along with it, that allows us to get to a level of what we call ultra-density, where we can run 10,000 processes on a machine," he said.
Cloud hyperscalers can counter this, at least for small companies, with generous free tiers, along with grants such as the Google for Startups program, which awards up to $350,000 in cloud credits to startups that pass its application process. Railway's startup program, launched in 2025, awarded up to $5,000 in credits to 10 high-growth startups.
How do you move as quickly as humanly possible, or agentically possible, and build a differentiated experience?
Jake CooperFounder and CEO, Railway
AI apps and an open GPU question
In some ways, Railway is benefiting from a rising tide of neoclouds driven by growing demand for AI infrastructure. Railway positions itself as a good deployment target for AI-driven workloads because of its performance and counts AI infrastructure startup Kernel among its customers.
"Deployment in seconds is almost table stakes for a lot of agents," Cooper said. "How do you move as quickly as humanly possible, or agentically possible, and build a differentiated experience?"
However, unlike most of its neocloud counterparts, Railway does not offer GPU hosting.
"We have focused on optimizing CPU use cases given how we've built everything from the ground up," he wrote in a follow-up email to Informa TechTarget this week. "We even have templates for running open source models on Railway today that work because of our performance. We have the full intention of becoming the best intelligent cloud for integrating with and running AI applications, but cannot disclose explicit GPU plans at this time."
Industry analysts said the lack of GPU support could hinder Railway's efforts to capture more AI workloads.
"The Series B funding is good, but that does not guarantee success in a crowded field," said Larry Carvalho, an independent consultant at RobustCloud.
Railway is forging a unique path among neoclouds, said Jim Frey, an analyst at Omdia, a division of Informa TechTarget.
"Most neoclouds, like CoreWeave or Vultr, are focusing on accelerating access to the massive infrastructure needed to harness AI at scale," he said. "Railway is taking a somewhat similar approach [by] offering application hosting infrastructure that is simpler and less expensive than traditional cloud providers but appears focused more heavily on smaller developers and enterprises that aren't necessarily trying to embrace AI, [and] just need a faster path from development to production."
Beth Pariseau, a senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism. Have a tip? Email her.
Jake Cooper
