Platform-as-a-Service provider Railway has turned to AI to design its data center deployments.

In a recent blog post, Railway revealed it has turned to Anthropic's Claude to design the buildout of its "Gen 2" data center sites.

Claude planned the data center deployment for four geographies, eight or nine different data centers, with "a half-dozen suppliers, dozens of network providers, dozens of technicians, hundreds of line items, thousands of cables, and lots of velcro."

According to the company, this needed to line up within a two-to-three week installation window.

The company currently runs on a combination of Google Cloud Platform (GCP), Amazon Web Services (AWS), and its own bare-metal servers deployed in data centers.

For the company's first rollout of bare-metal servers, Railway deployed Gen 1 over eight phases across 18 months. "We'd build an initial site skeleton in each region, fill in 20 percent capacity, take a couple of months doing other stuff, then repeat at the next site. From the initial skeleton, we executed copy-pasta scale-ups in batches of 15-20 percent of site capacity and slowly filled them out."

Due to growing demand and industry constraints such as supply chain issues related to massive demand for "everything from DRAM to glass fiber," this is simply not fast enough.

Railway gave Claude a "framework" to operate within, which included version control for physical infrastructure that meant changes were tracked, applied, and inspected; design rules checks against devices, cables, racks, or sites to verify that they meet our requirements; a library of parts; and an attribute and constraint system.

Additionally, the company made its internal DCIM "real-time" and added a curated set of MCP tools with skills documenting processes.

A single global contractor was used for the actual buildout, with teams in every geography the company is live in.

In terms of hardware specs, Gen 2 comprises "the latest generation AMD Zen 5c EYPC CPUs with 96 cores (192 threads) with DDR5, five times more storage than Gen 1, and dual 100G ConnectX-6 NICs." It is also in the same chassis as the Gen 1 storage server, reducing four SKUs to two.

Railway noted that each Gen 2 site is built "beside its Gen 1 sibling rather than on top of it," enabling the company to "link the two over our own dark fiber: 400G links, across 4 diverse paths, with DWDM where we need more wavelengths. And in most regions we can also reach the nearest hyperscaler region in well under a millisecond, which matters for our Enterprise customers who keep one foot in AWS or GCP."

As of the start of June, Gen 2 was live in three of four regions: US West (California), US East (Virginia), and Amsterdam, the Netherlands. The remaining region is in Singapore.

Railway was originally built on Google Cloud Platform, but previously stated that this came with "a multitude of problems that have posed an existential risk to our business," including impacting pricing, and making it "rare" that the company understands why things break upstream. Railway previously wrote: "Despite multi-million dollar annual spend, we get about as much support from them as you would spending $100."

Notably, Railway recently ran into an issue with Google when the hyperscaler suspended one of its projects. A person familiar with the situation, however, told DCD that this was done because Google Cloud detected a spike in cryptocurrency mining, in violation of its terms of service.

The source added that multiple customer accounts were suspended, per its standard procedure, but Railway's account in particular has a multi-year history of failing to address abusive activity occurring on its platform.

Communications between Railway and Google Cloud were established fairly quickly, but no official support ticket was submitted.