From $7.5B to $18B in 4 Months: Inside the AI Data Center Startup Everyone Wants a Piece Of

A 2.4x Jump in Four Months

$7.5 billion. That was Fluidstack's valuation in December 2025.

$18 billion. That's the number on the table four months later, in April 2026. Same team, same business model — 2.4x the price.

It's the cleanest single data point for where AI's center of gravity has moved.

The Oxford Spinout Behind the Number

Fluidstack was founded in 2017 at Oxford University by seven co-founders: Jamie Cox, Peixian Wu, Gary Wu, César Maklary, Peter Francis, Reaal Khalil, and Daniel Fedder.

Gary Wu serves as CEO today. He read Economics and Management at Oxford and is a self-taught software engineer. Co-founder César Maklary holds the President role. In 2025, Rob Perdue joined as COO to add high-growth scaling experience to the leadership bench.

The company didn't start as a hyperscale operator. It launched as an "availability engine" — aggregating idle capacity from Tier 4 data centers and routing it to compute-hungry customers. That model matured into dedicated cluster operations as the AI boom intensified. Fluidstack now manages over 100,000 GPUs across its global footprint, with customers including Mistral, Character.AI, Poolside, and Black Forest Labs.

The Product — AI-Native, Not Just GPU Rental

Here's what separates Fluidstack from AWS or Azure.

Hyperscalers are general-purpose clouds. A department store of CPU VMs, storage tiers, networking primitives, managed databases, and hundreds of other services. Fluidstack does one thing: AI workloads. InfiniBand-first networking, top-tier NVIDIA H100 and H200 GPUs, and orchestration tuned for training and large-scale inference. That's it.

That narrowness delivers two things. Higher effective performance per GPU for AI-specific workloads. And faster provisioning of thousands-to-tens-of-thousands-of-GPU clusters, often in days rather than months. That's why Mistral runs on Fluidstack. It's also why Anthropic chose Fluidstack for custom data center buildouts.

In H2 2025, Fluidstack announced a €10B commitment to build a 1-gigawatt AI supercomputer in France. As U.S. demand surged, the company relocated headquarters from London to New York and redirected parts of the French project toward U.S. sites.

Source: commons.wikimedia.org · CC-BY 2.0

Deal Structure — Valuation Timeline

The round isn't closed yet. Here's what Bloomberg and TechCrunch have reported from the negotiation stage.

Date	Event	Valuation	Lead Investor(s)
2022	Series A	Undisclosed	—
Mid-2024	Growth round	Undisclosed (lower)	—
Dec 2025	Pre-IPO-style round	$7.5B	Undisclosed
Apr 2026 (in talks)	Growth round	$18B	Jane Street, Situational Awareness

The current $1B round is being co-led by Jane Street and Situational Awareness, with Morgan Stanley advising.

Jane Street is one of the largest quantitative trading firms in the world. When a trading house — not a traditional VC — deploys at this scale into AI infrastructure, the read is straightforward: this sector looks like a predictable-returns play, not a speculative one.

Situational Awareness is the fund operated by Leopold Aschenbrenner, the former OpenAI researcher whose 2024 essay of the same name argued that AGI could arrive faster than most people assumed. He's now deploying capital directly into the compute infrastructure that scenario requires.

Use of Funds — The Anthropic $50B Contract Explains It All

To understand where this money goes, start with Anthropic.

In November 2025, Anthropic selected Fluidstack as its custom data center partner. The contract covers design, construction, and operation of new data center campuses in New York and Texas. Reported contract value: $50B. It sits among the largest single AI infrastructure deals ever disclosed.

Executing that contract requires massive up-front capital. Power purchase agreements, site acquisition, multi-year GPU pre-orders, networking hardware, talent hiring — all in the billions. The current $1B round is growth capital to fund that pre-investment.

Reported primary uses include:

• Anthropic-dedicated site construction — power and cooling infrastructure for New York and Texas campuses
• Large-scale pre-orders of NVIDIA H200 and B200 systems
• U.S. operations hiring — engineering, site reliability, field operations
• Partial wind-down of U.K. and French assets in favor of U.S. concentration

Gary Wu stated in the company blog that "building the team that will drive our next phase of growth" is the priority. Capital translates into hardware, which translates directly into Anthropic revenue — a clean flow-through structure.

Source: commons.wikimedia.org · CC-BY-SA 3.0

Competitive Landscape — GPU Cloud Snapshot

To judge whether $18B is reasonable, compare sector peers side by side.

Company	Focus	Recent Valuation	Recent Quarterly Revenue	Key Contract
CoreWeave	AI cloud (public)	Approximately $35B+ (market cap)	Approximately $1B / quarter	Meta $21B (to 2032), Anthropic
Nebius	AI cloud	Publicly listed	Undisclosed (growing fast)	Nvidia $2B investment
Fluidstack	AI data centers	$18B (in talks)	Undisclosed	Anthropic $50B
Lambda Labs	GPU cloud	Approximately $10B	Undisclosed	Nvidia partnership
Crusoe	Green AI infra	Approximately $5B	Undisclosed	4.5GW natural gas secured

CoreWeave announced its expanded Meta agreement in April 2026 — a long-term $21B commitment to December 2032, stacked on an existing $14.2B arrangement. CoreWeave is doing roughly $1B in quarterly revenue and guiding to $5B annually. But its 4.8x debt-to-equity ratio and GPU-collateralized debt model have raised sustainability questions in analyst coverage.

Fluidstack hasn't disclosed revenue figures, but the Anthropic single-contract value is more than 2x CoreWeave's Meta contract. The case for $18B rests largely on that one relationship.

Sector Trend — Capital Flows Over Six Months

Capital flow into AI infrastructure over the last six months is outside normal ranges.

In November 2025, CoreWeave raised billions more in debt alongside the Meta deal, and Nebius took in $2B from Nvidia. In Q1 2026, global VC hit $300B with AI capturing 81%; infrastructure was the second-largest sub-category after foundation models.

Money flows into AI model companies. More than half leaves again almost immediately, heading to Nvidia, CoreWeave, and Fluidstack. Those are the actual beneficiaries.

When trading houses like Jane Street start anchoring infrastructure rounds, you're seeing institutional consensus that this flow-through structure holds for the foreseeable future.

What Changes from Here

Three concrete implications.

First, AI's bottleneck has fully shifted from models to infrastructure. That's why Anthropic signs $50B data center contracts and why infrastructure valuations move 2.4x in four months.

Second, GPU cloud pricing isn't coming down anytime soon. Expect the supply-constrained regime to hold at least through 2027. If you're budgeting cloud GPU hours for 2026 or 2027, don't assume 2025-level prices.

Third, large model companies and infrastructure providers are binding 1:1. Anthropic–Fluidstack, OpenAI–CoreWeave, Google–Nebius — these long-term, single-provider relationships are becoming the norm. Model company lock-in to a single infrastructure partner will intensify over the next 12 months.

The future of AI comes down to two questions, not one: who builds the smartest models, and who can run them at the largest scale fastest.

References

• TechCrunch: AI data center startup Fluidstack in talks for $1B round at $18B valuation
• Bloomberg: Jane Street in Talks to Back Fluidstack at $18 Billion Valuation
• Fluidstack: Selected by Anthropic to deliver custom data centers in the US
• Fluidstack: Strengthens leadership team with key executive hires
• Sacra: Fluidstack revenue, funding & growth rate
• Crunchbase: Gary Wu profile