Analysis
Between Monday and Friday, three companies whose entire business is renting, routing or powering other people's AI workloads closed $4.8 billion in private capital. Crusoe took $3 billion at a $30 billion valuation, Fluidstack took $1.5 billion at $18 billion led by Jane Street, and Gimlet Labs took $300 million at $3 billion led by Andreessen Horowitz. Crunchbase News counted Crusoe and Fluidstack as the two largest rounds of the week globally.
Here is the week's compute stack, deal by deal:
- Crusoe -- $3B Series F at a $30B valuation (Atreides Management and Valor Equity Partners): Denver-based, founded in 2018 by Chase Lochmiller and Cully Cavness to burn stranded natural gas for compute, now building gigawatt-scale AI campuses including the Abilene, Texas site tied to Stargate. Competitors: CoreWeave, Nebius, Lambda. Crunchbase
- Fluidstack -- $1.5B at an $18B valuation (Jane Street Capital): founded in 2017 as an Oxford spinout by Gary Wu, Cesar Maklary and Jamie Cox, relocated from London to New York, and the first known operator of Google TPU capacity outside Google itself. Anthropic picked it in November 2025 for a $50 billion US data center program in Texas and New York. Competitors: Crusoe, CoreWeave, Together AI. Forbes
- Gimlet Labs -- $300M Series B at a $3B valuation (Andreessen Horowitz, with Arm, M12, Sapphire, Menlo and Factory): San Francisco, building a multi-silicon inference cloud that splits a single inference job across GPUs, CPUs and accelerators from Nvidia, AMD, Intel, Arm, Cerebras and d-Matrix. Total raised now $392 million. Competitors: Baseten, Fireworks AI, Together AI. SiliconANGLE
“Anthropic picked it in November 2025 for a $50 billion US data center program in Texas and New York.”
The repricing is happening in months
What separates this week from the 2021 growth-round era is the speed of the marks. Fluidstack was raising at $7.5 billion in December 2025 and closed at $18 billion nine months later. Crusoe was valued around $10 billion in its December 2025 round and printed $30 billion this week. Nscale, the London neocloud, is in talks for $3.5 billion of pre-IPO financing at the same $30 billion cap, roughly double its March mark, according to Yahoo Finance.
Those step-ups are not being underwritten off revenue multiples in any conventional sense. They are being underwritten off contracted backlog: Nscale has briefed investors on roughly $103 billion of total contract value, most of it a single $45 billion Anthropic agreement signed August 26. Fluidstack's Anthropic relationship is worth a reported $50 billion in build commitments. SB Energy, SoftBank's data center and power arm now in IPO registration, disclosed about $439 billion of backlog.
Four customers, one risk profile
Strip away the branding and this is a supply chain with four creditworthy buyers -- Anthropic, OpenAI, Google and Meta -- plus a long tail of labs that may not survive their next raise. Every valuation in the list above is, in practice, a credit view on those counterparties honoring multiyear compute commitments.
That is the part the headline numbers hide. Backlog is not revenue. A $45 billion contract is a promise to buy capacity that in many cases has not been built, on sites where power interconnects are still queued. Announced capital and deployed capital are separated by 18 to 36 months of substation upgrades, turbine deliveries and permitting. If demand for frontier training slows even modestly in 2027, the take-or-pay terms inside those contracts become the most important documents in the sector -- and none of them are public.
What the application layer gets
The rest of the week's top-10 list shows what non-infrastructure companies are clearing: Upwind Security at $300 million, Medici Brands at $250 million, HiBob at $166 million, Lyte at $165 million, TabaPay at $155 million, Thyme Care at $125 million. Real rounds, all of them, and all an order of magnitude below the compute tier.
For GPs, the practical question is whether an $18 billion or $30 billion entry price on a capacity reseller can return a fund. The bull case is that these become utilities with 20-year contracted cash flows. The bear case is that they are leasing companies with venture multiples, and the exit math only works if the IPO window stays open. Nscale, CoreWeave and SB Energy will settle that argument in public markets long before the private marks reset.
The number to track through October is not the next mega-round. It is how much of Crusoe's and Fluidstack's contracted backlog converts to recognized revenue in the fourth quarter, because that ratio is the only thing standing between a utility multiple and a repricing.