A high school dropout earns an annual salary of 155 million RMB, and his unicorn company is about to launch an IPO.

Recently, a company established just over two years ago submitted its listing application.

It is called Nscale, an AI infrastructure company headquartered in London, which plans to list on the New York Stock Exchange under the ticker symbol "NSCL".

According to the Financial Times, the maximum target valuation for this IPO is about 35 billion U.S. dollars, equivalent to about 237 billion yuan. In March this year, Nscale just completed a 2 billion U.S. dollar Series C financing with a valuation of 14.6 billion U.S. dollars at that time.

In the first half of 2026, Nscale recorded revenue of 140.6 million U.S. dollars, a year-on-year increase of 1252%; but its net loss reached 1.02 billion U.S. dollars in the same period. By the end of August, the total value of the company's long-term contracts had reached 103.4 billion U.S. dollars. However, this figure does not equal the realized revenue: the total value of contracts currently in execution is about 2.6 billion U.S. dollars, and most of the rest are future projects.

The bulk of the contracts come from Microsoft and Anthropic. The maximum contract amounts disclosed by the two companies are about 43.8 billion U.S. dollars and 44.6 billion U.S. dollars respectively, totaling about 88.4 billion U.S. dollars, accounting for about 85% of Nscale's total contract value of 103.4 billion U.S. dollars.

The most eye-catching figure is also Nscale's founder Josh Payne.

32-year-old Payne is Australian. He dropped out of high school, once worked in a coal mine, and later worked as a construction worker, miner, and Bitcoin miner. In 2024, Nscale was spun out from Arkon Energy, a company Payne founded back in 2020. In 2025, Payne received a total compensation of 23.207 million U.S. dollars from Nscale, equivalent to about 155 million yuan, the vast majority of which came from stock options.

On the eve of Nscale's listing, NVIDIA also promised to provide 1 billion U.S. dollars in convertible note financing to Nscale; at the same time, Nscale is also an important customer of NVIDIA GPUs.

How did a high school dropout build an AI infrastructure company targeting a 35 billion U.S. dollar valuation in just a few years?

He has always focused on "electricity"

Payne's entrepreneurial story did not start with AI.

Payne was born in 1993 in Newcastle, New South Wales, Australia. After dropping out of high school, he did not go to university, but went to work in a coal mine in the Hunter Valley. The shifts in the mine were long, and in his spare time, he read books on entrepreneurship.

Later, he left the coal mine to work on construction sites, working on large development projects in Sydney. In 2017, he co-founded a mining labor service company to supply workers for the mining and construction industries. It was during these experiences that he gradually came into contact with another key word: energy.

Around 2019, Payne entered the renewable energy sector and founded a Bitcoin mining company that was later renamed Arkon Energy. Bitcoin mining is a very "asset-heavy" business, with a large number of machines running at the same time, so electricity is naturally one of the most important costs.

Arkon quickly set its sights on Norway. In 2022, the company acquired a Bitcoin data center in northern Norway, targeting the local abundant and low-cost hydropower resources. This experience also became the starting point for Nscale's later business model.

After the rise of large AI models, the demand for GPUs has grown rapidly. But GPUs are only part of the computing power. To make tens of thousands or hundreds of thousands of GPUs actually operate, land, electricity, data centers, cooling systems, networks and software are also required. Payne's judgment is that instead of only doing one link in the chain, it is better to build the entire chain together.

Nscale summarizes its model as "power to token" — covering the whole path from electricity all the way to AI services. Around 2023, Payne began to prepare land and power resources for the new AI infrastructure business; in May 2024, Nscale was spun out from Arkon Energy and officially entered the market.

Shortly after the business started, Payne sent a direct unsolicited message on LinkedIn to Øyvind Eriksen, CEO of Norwegian industrial group Aker, to promote his idea of building an AI data center in Norway. Unexpectedly, this message eventually led to a cooperation. Aker later became an important shareholder of Nscale, and Eriksen also joined Nscale's board of directors.

In terms of financing, Nscale's speed is also very fast.

In December 2024, the company completed a 155 million U.S. dollar Series A financing; in September 2025, it completed a 1.1 billion U.S. dollar Series B financing led by Aker, with NVIDIA, Dell, Nokia and others participating in the investment.

In March this year, Nscale completed a 2 billion U.S. dollar Series C financing with a valuation of 14.6 billion U.S. dollars. The round was led by Aker and 8090 Industries, with NVIDIA, Dell, Nokia and others continuing to participate. Former Meta COO Sheryl Sandberg, former Yahoo President Susan Decker, former British Deputy Prime Minister Nick Clegg and others also joined the board of directors at this stage.

On September 15, three days before Nscale submitted its IPO application, the company signed at least a 3.1 billion U.S. dollar financing arrangement, of which NVIDIA promised to subscribe 1 billion U.S. dollars.

In just two years, Nscale has reached the doorstep of IPO.

Nscale's business logic

What exactly does Nscale sell?

Simply put, it is an infrastructure company for the AI era.

Large AI companies need a large number of GPUs, but building data centers by themselves not only requires purchasing GPUs, but also solving a series of problems such as land, electricity, computer rooms, networks, deployment and operation and maintenance. What Nscale does is to integrate these links to provide customers with a full set of infrastructure from electricity, data centers to GPU computing power, and then to cloud services.

Nscale currently has two main businesses: Nscale Infrastructure is responsible for power, data center and GPU infrastructure; Nscale Cloud provides cloud computing, inference, model deployment and other services. Among them, infrastructure is still the core of the entire business.

As of August 31, under Nscale's long-term "take-or-pay" contracts, about 2.6 billion U.S. dollars of contracts are actually in performance; if the contracts that have taken effect but have not yet started performance are included, the total contract value reaches 103.4 billion U.S. dollars.

The so-called "take-or-pay" means that customers lock in part of the computing power capacity in advance and pay according to the contract agreement, instead of paying entirely according to the actual usage. The weighted average term of such contracts is about 5.7 years.

In terms of customers, Microsoft and Anthropic are the two largest orders so far.

Since the end of 2025, Nscale has successively signed a number of agreements with Microsoft, with a maximum total contract amount of about 43.8 billion U.S. dollars, and the service period is extended to the end of 2033 at the latest.

On August 25 this year, Nscale signed a service agreement with Anthropic with a maximum total amount of about 44.6 billion U.S. dollars. According to the agreement, Nscale will deploy NVIDIA Vera Rubin NVL72 GPUs at the Monarch Compute Campus in West Virginia to provide dedicated GPU infrastructure for Anthropic.

Calculated based on the maximum contract amount, the two contracts total about 88.4 billion U.S. dollars, accounting for about 85% of Nscale's current total contract value of 103.4 billion U.S. dollars. Most of the projects corresponding to the contracts still require subsequent construction, deployment and delivery.

This can also be seen from the company's computing capacity.

As of August 31, Nscale's total operational computing capacity (7MW owned + 48MW leased) is about 55MW; at the same time, there is another 1.3GW of capacity under development. This new capacity is mainly to meet the customer contracts that have been signed in the future.

In general, the scale of Nscale's orders is already very large, but the business that has actually built data centers, deployed GPUs and started to contribute revenue is still only a small part of it.

In terms of revenue, Nscale is indeed in a stage of rapid growth. In the first half of 2026, the company achieved revenue of 140.6 million U.S. dollars, a year-on-year increase of 1252%.

But at the same time, the company's net loss in the same period reached 1.02 billion U.S. dollars, and the cost of revenue was 189.6 million U.S. dollars, which alone exceeded the revenue for the period. At the gross profit level, the company still failed to achieve profitability in the first half of the year.

What really needs attention is the scale of capital expenditure that Nscale will have to bear next.

Data centers, GPUs and power facilities are inherently capital-intensive businesses, and most of Nscale's long-term contracts still need to be fulfilled through continuous construction and deployment. The company also clearly stated in its prospectus that a large amount of capital is still needed to support business operations and expansion in the future.

It is more noteworthy that Anthropic's 44.6 billion U.S. dollar contract itself is also subject to financing conditions. Nscale needs to obtain eligible financing for the relevant GPU and data center infrastructure, and the relevant financing is still in progress.

Nscale's IPO narrative is driven by order-driven growth expectations on one hand, and continuous and huge capital expenditure pressure on the other.

The same computing power race, two different paths

Nscale is not an isolated case.

Over the past few years, a number of companies that have shifted from cryptocurrency mining to AI infrastructure have emerged overseas. CoreWeave initially did Ethereum mining and later turned to GPU cloud computing; IREN entered AI data centers and computing power services from Bitcoin mining; Crusoe also used energy resources for Bitcoin mining in the early days, and now its business focus has shifted to AI infrastructure.

The paths of these companies are not exactly the same, but their transformation logic is very similar: mining and AI computing have different uses, but both are highly dependent on electricity, land, data centers and large-scale computing equipment. After the demand for AI rises, the infrastructure accumulated in the past has new uses.

China has taken a different path.

The more obvious change in China is that traditional data center companies have begun to extend from IDC to AIDC and computing power services. With the growth of AI demand, data centers no longer only provide computer rooms and cabinets, but further expand to high-density intelligent computing infrastructure and computing power services.

One type of enterprise still focuses on intelligent computing infrastructure. In the past, IDC mainly provided computer rooms, cabinets, power and networks. After the rise of AI, data centers need to carry AI servers with higher power density, which puts forward higher requirements for power supply, liquid cooling and networks.

The other type continues to extend to computing power services, where enterprises purchase and operate AI servers to directly provide customers with computing power such as training and inference. This layer is closer to the model of Nscale and CoreWeave, which also means higher capital investment and operational requirements.

Several leading domestic data center enterprises are accelerating their layout in these two directions.

GIOC (GDS) proposed this year that it plans to invest 300 billion to 50 billion yuan in the domestic AI infrastructure field in the next three years to meet the rapidly growing demand for AI data centers.

Aofei Data is also promoting data centers and intelligent computing projects in Langfang, Zhangjiakou, Wuxi and other places, and some key projects are constructed in accordance with the intelligent computing requirements for power supply and other infrastructure from the very beginning of design.

Runze Technology has gone a step further. The company began to layout AIDC computing power business in 2023. By the first half of 2026, the revenue of this part had reached 1.995 billion yuan, a year-on-year increase of 126.24%, accounting for 53.26% of the total revenue, exceeding the traditional IDC cabinet business.

Data Harbor has also added intelligent computing capabilities on the basis of its original data center business, extending to computing power services and intelligent computing operations.

The several companies have different paths, but their directions are the same: data centers are evolving from simply carrying servers to carrying and providing computing power.

This is the same problem that Nscale is facing, but the industrial paths are different.

Nscale extends all the way from electricity, land and data centers to GPUs and cloud services; in China, the existing data centers, cloud computing, operators and computing power enterprises extend upstream and downstream from their original links respectively.

One obvious difference is the chip and computing power ecosystem. Nscale builds its infrastructure highly around NVIDIA GPUs, and NVIDIA is also its shareholder and financing participant; the chip sources and technical routes of domestic AI infrastructure are more diverse, and the deployment scale of domestic AI chips is also continuing to expand. Different hardware and software ecosystems have also led the two sides to take different paths in data center design, cluster construction and computing power service models.

The customer and financing structures are also different. Nscale locks in demand through multi-year contracts with customers such as Microsoft and Anthropic, and then continuously conducts equity, debt and equipment financing; large domestic intelligent computing projects are more constructed by multiple entities such as large Internet manufacturers, cloud vendors, operators, data center enterprises and local state-owned assets, with more diversified financing methods.

Therefore, China may not see a "Chinese version of Nscale" or "Chinese version of CoreWeave".

A more likely pattern is: data center enterprises extend to computing power services, computing power enterprises lock in computer rooms and power upstream, cloud vendors and operators continue to sink into infrastructure, and several forces gradually converge at the AI computing power link.

In the past, when measuring an IDC company, the market paid more attention to the number of cabinets, cabinet occupancy rate and power cost. Entering the AI era, electricity, chips, data centers and customer demand are tied together, and competition has evolved from competing for a single resource to competing for the ability to organize these elements together.

Conclusion

From a coal mine in Australia to AI data centers, what Payne has done in the past few years actually has a constant main line: find cheap energy, and then convert energy into computing power.

Now, Nscale holds 103.4 billion U.S. dollars in long-term contracts on one hand, and the value of contracts actually in performance is only about 2.6 billion U.S. dollars on the other. In the first half of 2026, the company's revenue was 140.6 million U.S. dollars, but its net loss was as high as 1.02 billion U.S. dollars.

For Nscale, the real test may have just begun.

This article is from the WeChat official account "Thecapital" (ID: thecapital), author: Li Bingzhi, authorized for release by 36Kr.