Analysis
Fluidstack has closed roughly $1.5 billion led by Jane Street Capital at an $18 billion valuation, according to Crunchbase News, which ranked it the second-largest round of the week behind Crusoe. Total funding now stands at about $2.6 billion.
The company was founded in 2017 as a University of Oxford spinout by Gary Wu, Cesar Maklary and Jamie Cox, and started life as a marketplace stitching together idle GPUs. It moved its headquarters from London to New York last December. What changed the trajectory was not the marketplace but the build business: Fluidstack now develops and operates purpose-built AI data centers for other people's chips, and is the first publicly known operator of Google TPU capacity outside Google itself, per Forbes.
In November 2025, Anthropic named Fluidstack as its partner for a $50 billion American compute program, starting with custom sites in Texas and New York. Its customer list also includes Meta, Mistral, Poolside and Black Forest Labs. Over the past year it has recruited more than a dozen engineers out of Tesla and SpaceX, which tells you the hiring bar shifted from cloud software to heavy construction and power systems.
“The company was founded in 2017 as a University of Oxford spinout by Gary Wu, Cesar Maklary and Jamie Cox, and started life as a marketplace stitching together idle GPUs.”
The comparison set is instructive. Crusoe closed $3 billion at $30 billion the same week and owns its own generation assets. CoreWeave, the public bellwether, owns roughly a million square feet of leased capacity and carries billions in GPU-backed debt. Fluidstack's model is lighter: it does not need to own the accelerators, because the customer or the chip vendor often does. That asset-light structure is exactly why Jane Street -- a proprietary trading firm that prices contracted cash flows for a living -- is a plausible lead here in a way a traditional crossover fund might not be.
The valuation math is aggressive on any conventional basis. Fluidstack was raising at $7.5 billion in December 2025, meaning the mark has more than doubled in about nine months. Reported revenue has scaled from roughly $1.8 million a few years ago toward a projected $660 million, which at $18 billion still implies a multiple in the high twenties on forward revenue for a business with construction risk, power procurement risk and heavy counterparty concentration.
That concentration is the bear case. If Anthropic's compute needs slow or its financing tightens -- the company is separately raising debt ahead of a possible listing, which Pulse covered last week -- a large share of Fluidstack's contracted pipeline is exposed to one buyer's balance sheet. Data center construction also runs on physical constraints venture capital cannot compress: interconnect queues in ERCOT and PJM routinely run 24 to 48 months, and turbine and transformer lead times remain measured in years. Announced gigawatts and energized gigawatts are very different numbers.
For founders, the useful signal is that infrastructure buyers now award multiyear contracts to companies with no operating history at that scale, provided they can prove site control and power. That is a bankability test, not a product test.
The number worth tracking is how much of the Anthropic program energizes on schedule in 2027. Slip a year and an $18 billion mark starts looking like a construction company's multiple.
Update (September 7, 2026): Pulse has follow-up coverage — Thailand Freezes All New Data Center Approvals.
Update (September 7, 2026): Pulse has follow-up coverage — Cerebras Is the Comp Every AI Listing Should Study.