Company Analysis: Fluidstack and the AI Infrastructure Race

The CODEW | Company Analysis

Fluidstack has closed an $830 million Series A at a $7.5 billion valuation, led by Situational Awareness — the AI-focused investment fund founded by former OpenAI researcher Leopold Aschenbrenner, with backing from Stripe's Collison brothers, former GitHub CEO Nat Friedman, and investor Daniel Gross. The round, reported to have closed in December, is one of the largest infrastructure financings of the year and cements Fluidstack's position among a small group of "neocloud" providers racing to control the physical compute layer beneath the AI boom.

Company Analysis cover: Fluidstack and the AI infrastructure race — The CODEW, July 25, 2026.

But the December round may already be old news. By April, TechCrunch reported Fluidstack was in talks for a further $1 billion round at an $18 billion valuation — more than double its price from just months earlier — with Google reportedly considering a $100 million contribution tied to its broader push to distribute TPU accelerators through third-party data center partners. Whether or not that round closes at the reported terms, the pace of repricing illustrates just how quickly capital is chasing this category, and how far Fluidstack has traveled from its origins as a scrappy European GPU marketplace.

From Oxford Side Project to $50 Billion Anthropic Partner

Fluidstack was founded in 2017, emerging out of Oxford University, and spent its early years as a marketplace pairing startups and AI hobbyists with idle GPU capacity — including, according to reporting, unused compute from video gamers' rigs. The company raised a modest $3 million seed round in March 2019 from Seedcamp, Mercuri and 7 Global Capital, followed by a $35 million Series A in May 2021 led by MaC Venture Capital. For years, Fluidstack was a minor player in a crowded field of GPU rental services.

That changed dramatically starting in 2024 and 2025. Fluidstack's revenue reportedly ballooned to $66 million as the company pivoted from a compute marketplace into a full-stack AI infrastructure builder — designing, financing and operating dedicated data centers rather than simply reselling spare capacity. The turning point came in February 2025 at the AI Action Summit in Paris, where French President Emmanuel Macron announced more than $113 billion in AI-related deals, including an $11 billion "AI factory" agreement with Fluidstack — at the time still a little-known British startup. CEO Gary Wu and president Cesar Maklary represented the company at the summit alongside heads of state and Big Tech executives, a striking moment for a company that, months earlier, had been operating well outside the spotlight occupied by CoreWeave, Crusoe and the hyperscalers.

The real inflection point came in November, when Anthropic signed a data center partnership with Fluidstack reported to be worth roughly $50 billion, covering custom-built facilities in Texas and New York designed specifically for Anthropic's training and inference needs. The deal was significant enough that Fluidstack relocated its global headquarters from the UK to New York to be closer to the center of gravity of its most important customer relationship. In a signal of how sharply the company's strategic priorities shifted toward the U.S. market, Fluidstack also pulled out of a separate €10 billion AI infrastructure project in France to concentrate resources on American opportunities — a reversal of the very deal that had put the company on the map less than a year earlier.

Fluidstack's customer base now extends beyond Anthropic to include Meta, Mistral (the French frontier lab where Fluidstack built an early reputation), Character.AI, Poolside and Black Forest Labs — a roster concentrated almost entirely among frontier AI labs and AI-native companies rather than general enterprise customers.

What Fluidstack Actually Builds

Fluidstack's pitch centers on speed: the company says it can bring gigawatts of new compute capacity online in as little as six months, compared with the 18 to 24 months typically required for hyperscale data center buildouts. It manages the full stack required to do this — securing power agreements, designing and constructing physical data centers, and running the software layer that turns racks of GPUs into usable compute for AI labs.

That software layer is built on two proprietary systems: Atlas OS, a bare-metal operating system that automates server imaging and provisions Kubernetes or Slurm environments without requiring manual hardware configuration, and a platform called Lighthouse. Together, these let Fluidstack offer GPU clusters ranging from as few as 8 GPUs to clusters of 30,000 units, accessible via web console or API, with billing options spanning on-demand hourly rates, reserved instances with 30-day-plus commitments, and fully custom private cloud deployments for the largest customers. The company reports managing more than 100,000 GPUs across its network, with roughly 268 employees as of mid-2026 — a lean headcount relative to the scale of infrastructure and capital it is deploying, though estimates of employee count vary by source.

The Capital Stack: Treating GPUs Like Infrastructure, Not Software

What distinguishes Fluidstack's financing approach from a typical software startup is the layering of equity with asset-backed debt. Beyond its equity rounds, Fluidstack has arranged debt financing capacity of up to $10 billion through Macquarie Group, structured as a senior debt facility collateralized against the GPUs themselves, with financed hardware slated for deployment in an Icelandic data center powered by renewable energy. In 2024, the company had already layered in $24.7 million through a SAFE agreement and $37.5 million in separate debt financing.

This structure — equity for growth and strategic positioning, project-style debt secured against physical, depreciating hardware — mirrors financing more commonly seen in toll roads, pipelines or power plants than in venture-backed technology companies. It reflects a broader shift in how capital markets are beginning to treat AI compute: not as software risk, where the primary uncertainty is product-market fit, but as infrastructure risk, where the primary uncertainties are financing cost, utilization rates, and asset depreciation. Total disclosed equity funding for Fluidstack is estimated at roughly $1.65 billion to $1.78 billion across all rounds, not counting the multibillion-dollar debt facility.

The Competitive Field

Fluidstack competes in the "neocloud" category — pure-play GPU infrastructure specialists competing both against each other and against the hyperscalers (AWS, Azure, Google Cloud, Oracle) for AI training and inference workloads:

CoreWeave is the most established player in the category, having completed an IPO and built a reported backlog north of $55 billion, anchored by a deep relationship with Microsoft. It remains the scale benchmark the rest of the neocloud field is measured against.
Crusoe differentiates on sustainability and unconventional power sourcing — including stranded natural gas and renewables — and has pursued its own aggressive multi-gigawatt buildout plans, positioning itself as a lower-carbon alternative within the same infrastructure race.
Lambda built its early reputation on GPU hardware sales to research teams and competes largely on price and simplicity relative to larger, more capital-intensive rivals.
Nebius, spun out of Russian internet company Yandex, has expanded rapidly with ambitions to match hyperscaler-grade service quality and geographic reach.
TeraWulf, a former bitcoin miner repositioning toward AI infrastructure, has entered into transactions with Fluidstack directly — PitchBook data shows a merger/acquisition-style deal between Fluidstack and TeraWulf's Abernathy High-Performance Computing Campus, illustrating how neocloud players are increasingly acquiring or partnering with repurposed power and data center assets rather than building entirely from scratch.

Against this field, Fluidstack's principal differentiators are speed of deployment and its direct, deeply integrated relationship with Anthropic — a relationship significant enough to have reshaped the company's geographic priorities and headquarters location. That gives Fluidstack a level of revenue visibility that smaller or more diversified neoclouds may lack, but it also concentrates risk: Fluidstack's near-term fortunes are unusually tied to the continued scaling ambitions of a small number of frontier labs.

Structural Risks Facing the Category

The neocloud sector as a whole faces risks that are structural rather than company-specific, and Fluidstack is exposed to all of them:

Hardware depreciation. GPU technology cycles quickly — new chip generations from Nvidia and others can meaningfully devalue existing fleets within a few years. Rental rates for previous-generation chips like the H100 have already fallen sharply from their peak as newer hardware entered the market, a sign of commoditization pressure. Debt collateralized against racks of GPUs is, in effect, backed by an asset that could lose 40–60% of its value well before the debt is repaid — a mismatch that has drawn scrutiny from credit analysts as GPU-backed lending has become a more common financing mechanism across the sector.

Customer concentration. With Anthropic representing a reported $50 billion commitment and a relationship significant enough to drive a headquarters relocation, Fluidstack's growth trajectory is closely tied to the continued scaling of a small number of frontier AI labs. A slowdown in frontier training spend, a shift in Anthropic's infrastructure strategy, or increased AI lab appetite for building infrastructure in-house would disproportionately affect Fluidstack relative to more diversified competitors.

Geographic and political exposure. Fluidstack's decision to exit a €10 billion French government project in favor of U.S. opportunities shows how quickly strategic priorities can shift in this sector — and how exposed infrastructure providers are to national AI policy, sovereign investment priorities, and shifting political winds around data center siting, energy use and export controls.

Financing sustainability. The Macquarie-backed, GPU-collateralized debt structure is still a relatively novel mechanism at this scale. Its performance through a full hardware depreciation cycle — rather than just its initial closing — will be an important signal for whether "AI infrastructure as an asset class" is durable or a temporary artifact of extraordinarily strong current demand.

Valuation velocity. A jump from a reported $7.5 billion valuation in December to talks of $18 billion by April, in a matter of months, is an aggressive repricing even by AI-era standards. Rapid markups can reflect genuinely accelerating fundamentals — new contracts, expanded capacity, stronger revenue — but they can also outpace the underlying economics of a capital-intensive, thin-margin infrastructure business, particularly one whose core asset class is a depreciating commodity.

What to Watch

Whether the reported $18 billion round closes, and on what terms — including whether Google's rumored $100 million TPU-distribution-linked investment materializes, which would signal deepening ties between Fluidstack and a second major hyperscaler beyond its GPU-centric customer base.
Utilization and margin data, which remain largely undisclosed across the neocloud sector. As more debt gets collateralized against GPU fleets, the sustainability of that financing depends heavily on utilization rates holding up as newer chip generations arrive.
Consolidation. With CoreWeave, Crusoe, Lambda, Nebius, TeraWulf and others all raising and building at scale simultaneously, most analysts expect the neocloud market to consolidate around a smaller number of winners over the next several years, favoring companies with the deepest hyperscaler-grade customer relationships and the most efficient capital structures.
Anthropic's own infrastructure strategy. Because so much of Fluidstack's near-term valuation rests on its relationship with a single frontier lab, any shift in how Anthropic sources compute — whether toward other neoclouds, hyperscalers, or in-house buildout — would be the single largest swing factor in Fluidstack's trajectory.

The Broader Neocloud Financing Environment

Fluidstack's layered capital stack is not happening in isolation — it reflects a financing environment that has shifted markedly over the past year as AI infrastructure spending has scaled into the hundreds of billions of dollars. Traditional venture equity alone can no longer fund the buildout pace frontier labs are demanding, which has pushed neocloud operators toward increasingly creative capital structures: GPU-backed asset financing, project-finance-style debt facilities, and direct hyperscaler participation in what would once have been pure independent infrastructure plays. Analysts have noted that lenders including Macquarie, Blue Owl and others are increasingly comfortable extending large facilities against collateral — racks of GPUs — that would have been considered "nebulous and tricky to value" only a few years ago. The willingness of institutional lenders to underwrite that risk at Fluidstack's scale is itself a signal of how mainstream AI infrastructure financing has become, even as the underlying collateral remains genuinely difficult to price given the pace of chip generational change.

This financing environment also helps explain why neocloud valuations have become so volatile and so fast-moving. Because so much of the capital stack sits in debt rather than equity, a relatively modest amount of new equity can support an outsized valuation increase — which may partly explain how Fluidstack's reported valuation could more than double in a matter of months without a commensurate jump in disclosed revenue or contracted backlog. It also means that neocloud equity investors are, in effect, underwriting a residual claim on a business whose senior obligations are secured by rapidly depreciating hardware — a risk profile more common in leveraged infrastructure buyouts than in traditional venture investing.

The Bigger Picture

Fluidstack's journey — from an Oxford-born GPU marketplace pairing idle gaming rigs with cash-strapped researchers, to a $7.5 billion infrastructure company negotiating a possible $18 billion valuation less than a year later — is one of the more dramatic repricing stories of the current AI cycle. It captures both the scale of investor conviction that compute will remain the binding constraint on AI progress, and the speed with which that conviction can inflate valuations for companies whose underlying assets — GPUs, power contracts, physical buildings — depreciate and shift in value far faster than the venture rounds built on top of them. Fluidstack's raise doesn't settle the AI infrastructure race. However, it does confirm that investors are still willing to write infrastructure-scale checks for compute providers, even as questions about hardware depreciation, customer concentration, and long-term margins remain unresolved for now.

Sources

This analysis is based on official company announcements, strategic partnership announcements, and industry reporting, including:

Supporting Industry Coverage

Building instead of buying? Start with a company that actually exists.

Every "build vs. buy" decision starts with a more basic one: is your company legally formed yet? Firstbase handles US incorporation end-to-end — LLC or C-Corp setup, EIN registration, a business bank account, and ongoing compliance — all from one dashboard, no lawyer required.

Start Your Business with Firstbase →

Company Analysis is a flagship editorial series from The CODEW that provides in-depth analysis of technology companies, examining their business models, competitive positioning, financial performance, products, partnerships, strategic decisions, and long-term market outlook.

The CODEW