Fluidstack, a U.K.-founded startup that spent years operating quietly after launching in 2017, has become a key third-party infrastructure operator in Google’s push to expand use of its in-house TPU chips and reduce dependence on Nvidia hardware.
From gamer GPU rentals to AI data centers
Fluidstack originally helped gamers rent out GPU compute for income. It has since shifted into a full-stack AI data center developer. The company, now headquartered in New York, says it manages more than 100,000 GPUs worldwide and offers managed orchestration and zero-egress-fee pricing to help customers cut operating costs.
Fluidstack also announced that it had raised $830 million in Series A funding at a $7.5 billion valuation and moved its global headquarters from the U.K. to New York.
Leading Anthropic’s U.S. compute buildout
The report says Fluidstack is leading Anthropic’s $50 billion U.S. compute infrastructure project, taking charge of power procurement, data center design, and operations. The company says it can shorten delivery times for gigawatt-scale compute capacity from the usual 18 to 24 months to under six months.
A third-party platform in Google’s TPU expansion
Google has spent decades developing TPUs, or Tensor Processing Units, to support its machine learning and search systems. The article says Google Cloud has a backlog of $514 billion.
It also cites prior Morgan Stanley reporting that estimated Google’s annual TPU revenue could reach about $108 billion. According to Forbes, Google has started selling TPUs directly to outside customers and backing those sales with infrastructure support. That includes an agreement with Anthropic for purchases of up to 1 million TPUs.
In this latest wave of compute expansion, Fluidstack is described as a critical third-party operating platform. Its data centers in places including New York and Texas are hosting and running this infrastructure, giving physical support to large-scale AI chip deployments tied to Google and Anthropic.