Fluidstack Closes $1.5B Jane Street Round at $18B: The Neocloud Financing Google Built

A little-known UK-born startup just became one of the most closely watched companies in the AI infrastructure boom. Fluidstack, the “neocloud” operator that builds data centers purpose-built for AI workloads, has closed a new $1.5 billion round led by quantitative trading firm Jane Street at a valuation of more than $18 billion, Forbes reported on September 3 — more than double the $7.5 billion valuation the company carried as recently as this spring.

The round is the latest, and largest, milestone in one of the strangest corporate trajectories of the AI era: a GPU-leasing startup founded in 2017 by three twenty-somethings that has become the tip of the spear in Google’s campaign to break Nvidia’s grip on AI compute.

From broker to builder

Fluidstack was founded in 2017 by Gary Wu, Cesar Maklary, and James Cox. For most of its life it operated as a marketplace connecting researchers with spare GPU capacity — a broker, not a builder. That changed as frontier AI labs began demanding something the public clouds could not offer on acceptable terms: enormous, dedicated clusters with multi-year commitments and none of the margin stacking.

The company’s breakout moment was the $50 billion deal to build data centers for Anthropic, one of the largest private infrastructure projects in US history. Fluidstack now describes itself as leading the deployment of Anthropic’s $50B compute buildout, and its customer list has grown to include Meta and Mistral among others. The company manages more than 100,000 GPUs and raised an $830M Series A — a round size that would have counted as a mega-round for most startups, but that now reads as a footnote in its cap table.

The revenue numbers tell the story of the shift. Fluidstack grew from $1.8 million in revenue in 2022 to $66.2 million in 2024. Sacra estimated the company was at roughly $180M ARR by the end of 2024, with its original marketplace model contributing only about 38% of revenue. The rest comes from “Private Cloud” — Fluidstack-owned GPUs inside co-located facilities, dedicated to a single lab.

The Google machine

What makes Fluidstack’s rise genuinely unusual is who is underwriting it. Google, which is racing to sell its TPU AI chips as an alternative to Nvidia’s GPUs, has effectively wrapped Fluidstack in a financial embrace that de-risks the neocloud’s balance sheet in exchange for TPU distribution.

The mechanism, as reported by Reuters, the Wall Street Journal, and Data Center Dynamics, works roughly like this: Fluidstack commits to building data centers and stocking them with Google TPUs. Google, with its AA+ credit rating, provides lease backstops — first $3.2 billion, later expanded to as much as $1.8 billion in additional obligations — that swap Fluidstack’s financing risk for Google’s. In return Google has taken equity in the company, at one point roughly 14%, and later negotiated a further ~$100M investment for about 41 million shares, an 8% stake, tied to a $7.5B valuation.

For Google, the calculus is straightforward: Nvidia wins partly because it doesn’t just sell chips — it sells an ecosystem in which partners can finance, deploy, and operate at scale. Google is borrowing that playbook. Every Fluidstack data center filled with TPUs is capacity that is not filled with Nvidia GPUs, and Google is willing to spend its balance sheet to make that happen. Forbes frames the new $18B valuation as a direct read on how that bet is performing: the startup has become a working test bed for Google’s merchant-silicon ambitions.

The Jane Street imprimatur

The lead investor in the new round is Jane Street, the quant trading firm that has quietly become one of the most aggressive investors in AI infrastructure. This is not charity. Jane Street’s core business is pricing and trading complex instruments, and AI data center finance — with its decade-long power contracts, GPU depreciation schedules, hyperscaler backstops, and take-or-pay leases — is exactly the kind of structured-credit puzzle the firm trades for a living. Bloomberg first reported the two parties in talks back in April, when the parameters were around $1B at an $18B target; the round as closed came in at $1.5B, suggesting demand exceeded the original book.

The doubling from $7.5B to $18B in roughly four months invites the obvious question: what changed? The answer is that between those two marks, Fluidstack’s $50B Anthropic buildout moved from announcement to execution, and its contracted revenue base — reported at around $7 billion of base-term contracted revenue backstopped by Anthropic’s strategic commitments — gave investors a floor to underwrite against. When a neocloud has a hyperscaler-grade counterparty signing multi-year leases and a second hyperscaler backstopping its equipment financing, it stops trading like a venture bet and starts trading like infrastructure credit.

The neocloud math

Fluidstack is not alone at the top of this market. The six leading neoclouds — CoreWeave, Crusoe, Nebius, Nscale, Lambda, and Fluidstack — are collectively valued above $150 billion, and ABI Research forecasts GPU-as-a-service revenue from neocloud providers to surpass $250 billion by 2030, up from $42 billion in 2025. CoreWeave’s $14B Meta deal and $55B backlog showed the model works at scale; Fluidstack’s round shows capital markets will now fund the second tier at comparable multiples.

But the structure of Fluidstack’s rise carries its own risks. The company is, by its own economics, a leveraged landlord: it borrows against long-dated leases to buy chips that depreciate fast, and its equity story depends on the durability of a handful of contracts with a handful of customers. If Anthropic or Google’s commitments waver — or if TPU economics shift under Google’s own cloud strategy — the same leverage that doubled the valuation can compress it just as quickly. Analysis of neocloud economics suggests roughly $3 billion of free cash flow per gigawatt of capacity on average over an equipment cycle; whether Fluidstack’s deal structure lets it capture that full spread remains the open question.

What is not in question is the direction. AI labs need compute on terms public clouds won’t give, chipmakers need distribution partners they can finance, and deep-pocketed investors need yield in a world starved of it. Fluidstack has positioned itself at the intersection of all three — and with $1.5 billion more in the bank and an $18B target now validated by one of the sharpest trading desks in the world, it has the fuel to keep building.