←News detail

HuxiuUpdated Oct 25 18:26

Fluidstack Completes $1.5 Billion Financing at a Valuation of $18 Billion

AI summary

AI infrastructure company Fluidstack has completed $1.5 billion in financing at a post-money valuation of $18 billion, led by quantitative trading giant Jane Street, with its valuation rising from $7.5 billion to $18 billion within 60 days. Its core asset is a multi-billion-dollar compute contract signed with Anthropic, and it is participating in Anthropic's plan to deploy up to 1 million Google TPUs, becoming the first publicly disclosed TPU data center operator other than Google. The article points out that this circular financing, in which the funder simultaneously acts as customer and guarantor, carries systemic risk, and that Fluidstack's revenue is highly concentrated in a single customer such as Anthropic.

Why read this

AI score 71

The article breaks down the financing structure behind Fluidstack's valuation more than doubling 2.4x in two months, allowing readers to see clearly how compute contracts are priced as cash-flow instruments.

Topics

Tags

Body · AI translation

[[**AI infrastructure company Fluidstack's valuation soared from $7.5 billion to $18 billion within 60 days, completing a $1.5 billion financing round led by quantitative trading giant Jane Street. Its core asset is a massive multi-year contract backed by Anthropic, which the market views as a typical case of "sugar-daddy-style financing."**

**Key Points**

**1. The valuation surge stems from Anthropic's "sugar-daddy" contract**

Fluidstack does not produce chips or large models; instead, it has become the first publicly known TPU data center operator outside of Google by virtue of a multi-billion-dollar computing power contract signed with Anthropic (including participation in its deployment of 1 million TPUs). Its valuation has multiplied 2.4 times in 60 days. What the market is pricing is not the company's revenue, but this cash flow certificate guaranteed by Anthropic's balance sheet.

**2. Jane Street is betting capital on the entire computing power industry chain**

Jane Street not only led Fluidstack's round but also simultaneously invested in Anthropic, Etched, CoreWeave, Crusoe, and other model, chip, cloud-layer, and power companies. Its logic is not a venture capital portfolio, but rather converting trading profits into computing power futures, making it one of the largest buyers in the AI computing power market.

**3. The asset-light model relies on engineering efficiency and long-term power locking**

Fluidstack's Atlas OS platform can deploy thousands of GPU clusters within 48 hours and build a data center in 6 months (traditionally requiring 18-36 months). Its revenue comes from service fees for "getting the power connected," not from hardware appreciation. Revenue is projected at $660 million in 2026, 10 times that of 2024, but revenue is highly concentrated in single customers such as Anthropic.

**4. Sugar-daddy-style financing carries systemic risk**

Money circulates within the system: Jane Street is both an investor and a customer, and Fluidstack's revenue depends on Anthropic's long-term contract. If the ultimate buyer's demand for computing power falls short of expectations or an IPO is blocked, the pyramid could collapse from the bottom. 2026-09-27 14:50 "Sugar-daddy-style financing," valuation of 120 billion, China Venture quick overview

Recently, foreign media reported that AI infrastructure company Fluidstack completed a $1.5 billion financing round at a post-money valuation of $18 billion (approximately 120 billion RMB), led by quantitative trading giant Jane Street, which has frequently made media headlines this year.

It should be noted that just two months ago, Fluidstack had only just completed a $750 million financing round at a post-money valuation of $7.5 billion. In other words, within 60 days, the company's valuation has multiplied 2.4 times.

Even more incredible is that this company contains almost none of the popular Silicon Valley elements: no large models, no GPU manufacturing, no consumer products, and no legendary founder narrative.

If one must find something relevant, it is that it holds an auditable, multi-year mega-contract backed by Anthropic. To put it plainly, just as the market jokes that CoreWeave is "kept" by Nvidia, Fluidstack is in fact a company "kept" by Anthropic.

From reselling idle graphics cards to a valuation of 120 billion

Going back to 2017, Fluidstack's two founders, Gary WuF (Wu Jiali, British Chinese) and Jamie Cox, met while studying at Oxford.

Gary studied economics and management and taught himself programming, while Jamie studied classical literature. The two observed a very intuitive contradiction: AI researchers pursuing PhDs and small laboratories found it difficult to quickly obtain sufficient GPUs from AWS, Microsoft, and Google Cloud, as cloud resources from big companies involved long queues and were expensive. On the other side, a large number of gamers, internet cafes, and small laboratories had high-priced graphics cards that were used only two or three days a week, sitting idle most of the time.

So the two came up with an idea: to build an Airbnb for GPUs, aggregating idle graphics cards from individuals and small institutions, packaging them and renting them to researchers who needed computing power, while gamers could earn pocket money from their idle GPUs. In 2017, Fluidstack was officially founded in Oxford, with its headquarters in London.

Their early business was not limited to GPU rental; they also aggregated idle bandwidth from users worldwide to do CDN content distribution, with clients including the veteran video website Dailymotion. In 2018, Jamie dropped out of Oxford and received the "Thiel Fellowship," and the two began to devote themselves full-time to the startup.

However, because the business model of the P2P idle GPU market had inherent flaws, many research users abandoned it after a trial, so the company got off to a bad start, with slow growth and meager revenue.

The turning point came in 2020, when a third co-founder, César Maklary, joined. He holds a master's degree in aeronautical engineering from Imperial College and previously served as an aerodynamics engineer at the Red Bull Racing F1 team, specializing in large complex systems, cooling, power, and hardware engineering.

Soon the pandemic arrived. During this period, reselling scattered graphics cards could not make much money, so the three shifted to leasing GPU pools in bulk from enterprises and university laboratories and then subletting them to AI companies. At the same time, the team developed its own bare-metal scheduling system, Atlas OS, to unify management and operations. No one could have predicted that this capability would become the company's biggest moat.

After ChatGPT launched, orders poured in like a tide. In 2022, Fluidstack's revenue was $1.8 million; in 2023, $30 million; in 2024, the audited financial statements filed with the UK companies registry showed $66.2 million.

The company's core selling point is that clusters of thousands of GPUs that others need months to deliver can be deployed and brought online by Fluidstack within 48 hours. Its best-known case was a one-time deployment of 2,500 GPUs for Poolside. Its client list is also expanding rapidly, with frontier AI startups such as Mistral AI, Character.ai, Poolside, and Black Forest Labs signing on one after another.

But up to this point, it was still a "small but beautiful" presence in Europe's AI circle. In February 2025, French President Macron announced at the AI Action Summit that Fluidstack would take part in building a €10 billion, 1-gigawatt AI data center in northern France. Overnight, this little-known company became the hope of European AI sovereignty.

Soon, a dramatic turn occurred. In November 2025, Anthropic announced a $50 billion US computing investment plan, and Fluidstack was chosen as the builder and operator of custom data centers, and also became deeply involved in Anthropic's plan to deploy up to 1 million Google TPUs, making it the first publicly known TPU data center operator besides Google.

A few months later, as the US contracts were secured, Fluidstack withdrew from the French project, and its global headquarters moved from London to New York. Capital voted with its feet, and geography followed the contracts. Thereafter, riding on Anthropic's fame, Fluidstack became a frequent name in the news, though its founding team has remained low-key and rarely appeared in public.

The "zero-chip" model sought after by capital

Although Fluidstack was founded back in 2017, before its valuation took off, its fundraising record was so modest that it hardly looked like a star company.

Its only real funding round in its history took place in 2019. That year, Fluidstack completed a $3 million seed round, with investors including Seedcamp, Mercuri, and 7 Global Capital, among others. At that time the company was still doing P2P idle GPU and CDN businesses and had very little visibility.

Its second funding round came five years later. In 2024, during the critical period when its business shifted to enterprise GPU clusters, Fluidstack obtained a SAFE (a future equity agreement without equity pricing) and $37.5 million in debt financing; the investors in this round were not disclosed.

UK registration records show that as of the end of 2024, Fluidstack had completed only $30 million in cumulative equity financing, plus $38 million in debt. In February 2025, it secured a $200 million Series A led by private equity firm Cacti, and after that, Fluidstack's fundraising pace accelerated.

Money began chasing this company this year. At the start of the year, a $450 million Series A+ was led by the Situational Awareness fund under former OpenAI researcher Leopold Aschenbrenner; in July, the size of that round increased to $830 million, with the lead investor and valuation unchanged, though BlackRock, Google, Jane Street, and other investors were added.

Then came the latest round, a new $1.5 billion financing led by quantitative trading giant Jane Street, which sent the company's valuation soaring to $18 billion, with cumulative equity financing of about $2.6 billion, alongside tens of billions in matching debt financing for data center construction. This round also propelled the three founders into the ranks of "Europe's youngest self-made billionaires," except that they have already moved to New York.

At this moment, the picture of Fluidstack in the eyes of capital is not a cloud vendor, nor a customer agent for Nvidia, but a light-asset platform for "computing matchmaking and delivery":

Most of the data center shells, power, and chips are rented. It locks in nearly 1 gigawatt of power for the next decade from converted Bitcoin miners such as TeraWulf, Cipher Mining, and Hut 8, signs long-term agreements with real estate and capital parties, and then packages it into long-term computing contracts leased to Anthropic, Meta, Mistral, Poolside, and Black Forest Labs.

On the revenue side, according to an institutional investment memorandum, in 2026 Fluidstack will operate 1.3 gigawatts of computing capacity across more than 10 sites, with revenue of $660 million, more than double the 2025 expected value of $200 million and 10 times the 2024 actual value.

The key is that what it earns is not money from hardware appreciation, but service fees for "connecting 1 gigawatt of power at the right time." Its Atlas OS platform can automatically bring bare-metal GPU clusters online within days, and its modularly built computer rooms can be powered up in about 6 months, while the construction period for traditional data centers is 18 to 36 months.

Jane Street's other path

By the traditional price-to-sales measure, $18 billion against projected 2026 revenue of $660 million is 27 times—a figure that would be considered aggressive in any normal market for a business carrying construction risk, power-approval risk, and single-customer concentration.

But viewed through a different pricing framework, this is not pricing revenue; it is pricing a contract. Once a multi-year agreement on the scale of $50 billion is locked in, what remains is an engineering-execution problem—and engineering execution is precisely what institutions like Jane Street are best at assessing in the language of probability.

So to understand Fluidstack's valuation, one must first understand the people writing the check. We have previously detailed this company's legendary story in "Average Annual Salary of 18.25 Million Yuan, the Company Has No CEO," so we will not repeat it here.

It is the world's largest ETF market maker. The Financial Times called it "an extremely profitable anarchist commune." In 2025, Jane Street's net trading revenue was $39.6 billion and adjusted EBITDA was $31.2 billion, setting a Wall Street record.

The first quarter of 2026 was even more ferocious: single-quarter trading revenue of $16.1 billion and profit of $10.3 billion. Own capital (members' equity) stood at about $45 billion, nearly 20 times higher than in 2016. In Chinese public discourse, the company has already acquired a commonly accepted nickname: "America's High-Flyer Quant."

But unlike Liang Wenfeng's choice, Jane Street did not choose to train large models; instead, it uses capital to place bets at every layer of the AI industry chain. In 2026, for the first time, the company moved a large amount of money from its profit pool out of trading and into investment.

First, look at what happened to it this year:

On April 15, CoreWeave announced that Jane Street had signed an approximately $6 billion AI cloud platform contract, while also buying $1 billion of CoreWeave stock at $109 per share, securing priority access to Nvidia's Vera Rubin systems.

On August 18, inference chip company Etched announced a $700 million funding round at a $21 billion valuation, led by Jane Street. It is also Etched's first real paying customer. It had already invested more than $100 million in Etched before July, accompanying the valuation all the way from $5 billion to $21 billion.

In early September, Jane Street was reported to have signed a five-year, approximately $13 billion cloud contract with Crusoe. On the same day, again Jane Street led Fluidstack's $1.5 billion round.

Looking over Jane Street's investment map, a clear main line emerges: the model layer, the compute layer, and the infrastructure layer—full-stack holdings—and the targets it invests in also rise quickly. Specifically, the model layer includes Anthropic and Thinking Machines; the chip layer includes Etched and MatX; the cloud layer includes CoreWeave, Nscale, and Crusoe; the infrastructure layer includes Fluidstack; and the power layer includes X-energy.

Therefore, industry observers believe this is not at all the portfolio logic of venture capital, but rather a position sheet for cross-asset market making.

As one of the largest buyers in the AI compute market, Jane Street understands the real supply and demand at every layer better than any VC. By leading Fluidstack's round, it is essentially turning profit into compute futures.

Jane Street has also stepped in pitfalls. In February, the liquidation administrator of the TerraUSD collapse sued it for using a former intern to obtain inside information and sell ahead of others. In July, Reuters reported that its heavily invested AI hedge fund Situational Awareness suffered a single-month loss of about $15 billion after highly leveraged bets on AI hardware stocks collapsed, and its portfolio was sold off to Citadel at a steep discount.

This is fate; genius is always accompanied by controversy.

Finally, back to this financing itself. Some market views hold that this is a typical "kept" financing. The more academic name is anchor contract financing or circular financing, meaning the party providing the money is simultaneously the customer, guarantor, and even upstream supplier of the party receiving the money.

This was not invented by Fluidstack, nor even by this era.

What Jane Street is buying is not Fluidstack's execution capability, nor even the future of AI; it is buying a cash-flow certificate guaranteed by Anthropic's balance sheet. A valuation that rose 2.4 times in two months is not pricing a company; it is pricing a contract.

Of course, the cost is obvious. Money circulates within the system once and becomes revenue—so who ultimately pays for the system's real final demand? The ultimate buyers of compute are highly concentrated among a limited number of names. If the IPO process or revenue falls short of expectations, the pyramid could collapse directly from the bottom. Channel: Finance and Economics. This content comes from the internet; original link. The views represent only the author and do not represent Huxiu's position. For copyright issues, please contact hezuo@huxiu.com, and we will verify and handle it promptly. Everyone who is changing and wants to change the world is on the Huxiu APP [ ]

Source

Huxiu