The most ambitious AI labs are not just buying compute time. They are buying infrastructure by the megawatt, with contracts that stretch for decades. This shift from renting cycles to commissioning physical capacity is the story behind Fluidstack, a London-born GPU cloud platform that has quietly become a critical supplier for labs like Anthropic and Mistral. The company, which began in 2017 as a marketplace for spare GPU capacity, has recently anchored its strategy on a 25-year, 168-megawatt joint venture with bitcoin miner TeraWulf in Texas, a deal with a reported $9.5 billion revenue potential [Sacra, Oct 2025].
From Marketplace to Megawatts
Fluidstack’s initial wedge was an asset-light marketplace, connecting AI teams to third-party GPU capacity. That model still accounts for an estimated 38% of its revenue [Sacra, 2025]. The strategic pivot, however, is into owned infrastructure through its private cloud offering, which now drives the majority of its business. The company claims to manage over 100,000 GPUs [DCD, 2026], and its ability to deploy thousands of H200s with InfiniBand networking has become a key selling point. A telling case study was its work with AI startup Poolside, where Fluidstack says it provisioned more than 2,500 GPUs within 48 hours [Fluidstack blog, 2025].
The Infrastructure Moats
The scale of ambition is visible in the company’s recent moves and partnerships. Its joint venture with TeraWulf, announced in late 2025, is backed by Google, which has reportedly guaranteed $1.3 billion of the project’s obligations [Sacra, Oct 2025]. Project financing was successfully priced, and the Abernathy, Texas site is on track for commissioning in the second half of 2026 [TeraWulf, Dec 2025]. Furthermore, Fluidstack was selected by Anthropic as a partner to deliver custom data centers in Texas and New York as part of Anthropic’s $50 billion U.S. infrastructure investment [Fluidstack blog, 2025]. To manage this growth, the company has relocated its global headquarters to New York City [PRNewswire, Dec 2025] and brought in seasoned operators like Rob Perdue, former COO of The Trade Desk, as its new COO [Yahoo Finance, Feb 2025].
The Risks in a Capital-Intensive Race
Fluidstack’s strategy carries significant, inherent risks that come with the territory of heavy infrastructure. The capital required is staggering, and the company’s funding history reflects this relentless need. Despite raising over $265 million in disclosed funding, including a $200 million Series A led by Cacti in early 2025 [Sacra, 2025], the company is reportedly already in talks for a $1 billion round at an $18 billion valuation [TechCrunch, Apr 2026]. The competitive landscape is defined by three primary pressures: capital intensity, commodity risk, and customer concentration.
What Success Looks Like
For Fluidstack, the next twelve months will be about execution on its monumental construction projects and proving that its hybrid model can achieve sustainable margins. Success means successfully commissioning the Texas joint venture, securing the reported $1 billion funding round, and expanding its roster of sovereign AI and financial services clients. In the high-stakes patient population of frontier AI labs, the standard of care is evolving from mere cloud credits to dedicated, performant physical real estate. Fluidstack is betting it can be the provider that delivers both the land and the keys.