The data center leased at an annual cost of 290 million US dollars was soon resold for 1.67 billion US dollars, uncovering the top-tier AI asset operator.

How creative can AI capitalization get? Just look at Volta

This company was founded only half a year ago, and its valuation has already reached 2.4 billion US dollars. The investors behind it are all top-tier names: a16z, NVIDIA, and Michael Dell's family office.

Even more surprisingly, it has secured a whopping 100 billion US dollar super order from Anthropic.

But that's not the most impressive part of Volta.

While a large number of AI cloud companies such as CoreWeave are still struggling to borrow money and purchase GPUs, Volta has taken an extremely asset-light path:

The data centers are leased from Bitdeer, the money for buying GPUs is provided by investors, and Volta is only responsible for bringing together customers, computer rooms, equipment and capital, and then landed the 100 billion US dollar big contract from Anthropic.

Despite its low capital contribution, Volta has made a huge fortune.

On June 29, Volta leased the data center from Bitdeer at a price of 290 million US dollars per year. After Volta's renovation, it was sold to Anthropic at a price of 1.67 billion US dollars per year in August. In just one round of operation, the value has quadrupled.

Today, Silicon-based Observer will break down this top "operator" in the AI industry.

The 100 Billion Dollar "Jigsaw Puzzle Game"

AI computing power is no longer a game that ordinary people can participate in.

According to the plan of OpenAI Stargate, 10GW of computing power requires an investment of 500 billion US dollars. Roughly calculated, 1GW corresponds to 50 billion US dollars.

Don't think that's all. Having money is only the admission ticket to the AI computing power track.

Building a huge data center is a chaotic and complex business. It not only requires purchasing sufficient land, steel and power supply, but also completing the long construction process from steel and concrete to the installation of industrial components and GPUs.

The huge investment and cumbersome procedures have also brought a problem to the AI computing power industry: except for the giants, no other player can afford the huge capital expenditure and execution risk of the entire industrial chain at one time.

Different from CoreWeave which has taken on a huge amount of debt with great difficulty, Volta has turned AI computing power business into a jigsaw puzzle game.

Simply put, Volta has built a platform that integrates investment institutions, land with stable power supply capacity, customer demands, computing resources, software and operation capabilities.

The Tydal project, for which Volta has just won a 100 billion US dollar order, is a very typical example. What Volta faced was a typical multilateral deadlock:

Anthropic needs large-scale Rubin computing power, but does not want to bear the risk of data center development.

On the other hand, Bitdeer has ready-made land, power and data center resources, but lacks strong long-term customers and credit support.

Banks and infrastructure funds have sufficient capital, but they will not make investments easily without stable cash flow.

And what Volta really does is to reorganize the originally mutually distrustful demands of these parties into a valid and executable transaction.

On one hand, when learning that Anthropic has an urgent demand for computing power, Volta first spent 4.7 billion US dollars to lease the mining data center Tydal Data Center from Bitdeer for a lease term of 16 years.

After signing the lease, Volta started the renovation work, deploying equipment provided by Dell and NVIDIA Vera Rubin systems in this data center.

On the other hand, Volta quickly signed a 6-year computing power contract with Anthropic for a total amount of about 100 billion US dollars, corresponding to the total capacity of about 133MW of the Tydal project. This step is very critical, as it provides a cash flow that can be recognized by financial institutions for the entire project.

After securing the supply and demand sides, there is the last part: capital.

An AI data center with a capacity of hundreds of MW usually requires billions of US dollars in capital expenditure. If all these funds are paid by Volta's parent company itself, or all raised through equity financing, the project will quickly drag down the company's balance sheet.

Volta claims that it has mastered a unique large-scale infrastructure financing channel that will not dilute the company's equity.

Specifically, it introduces the project financing model of the traditional infrastructure industry into the AI sector. Volta has established a 50 billion US dollar AI infrastructure plan with the Spanish asset management company Azora, set up a project capital pool, and shares the cash flow of subsequent projects.

This logic is actually very similar to the traditional infrastructure Project Finance. For example, when building a toll road, you first get the government franchise and the right to collect tolls in the future, and then use the cash flow of the next few decades to get financing.

At this point, the entire structure is fully closed-loop.

Anthropic provides orders, Bitdeer provides computer rooms and power supply, NVIDIA and Dell provide equipment, institutional capital provides funds, banks provide credit support, and Volta acts as the project initiator and general coordinator, responsible for integrating all parties together.

In this process, what Volta earns is the price difference after processing traditional computing power into high-value AI computing power services.

Calculated based on this 100 billion US dollar contract from Anthropic, on average, it corresponds to an annual revenue of about 1.67 billion US dollars. The 16-year basic lease that Volta pays to Bitdeer is about 290 million US dollars per year on average, and the annual difference between the two exceeds 1.3 billion US dollars.

In other words, through Volta's operation, the asset value has quadrupled. This is the most brilliant part of Volta's business model.

Of course, these differences are not all Volta's profits. Expenses including electricity bills, GPU and server costs, project capital returns, financing costs and cloud platform operation fees need to be deducted from the difference.

Computing Power Price Has a 30% Premium, What Makes Volta So Competitive?

In addition to not requiring large upfront capital investment, Volta's model has another major advantage — extremely fast speed.

Although Anthropic has locked in more than 10GW of computing power this year, most of these computing power resources will not be available until two or three years later. For example, for the 5GW cooperation with AWS, less than 1GW is expected to be launched by the end of 2026. The 5GW agreed with Google will not be delivered in batches until 2027.

Volta is completely different.

From the signing of the contract to the launch of the first phase of Tydal, it took less than 5 months. According to the plan, the first batch of GPUs will be put into operation on December 31, 2026, and all 133MW capacity will be delivered in March 2027.

133MW is certainly not a large figure compared with 5GW. But for Anthropic at present, 133MW that can be used next year may be more valuable than 5GW that can only be obtained three years later.

This point can also be reflected from the price. The 100 billion US dollar contract corresponds to an actual IT load of 121MW, with a cooperation term of 6 years. Roughly converted, the annual revenue per MW of IT capacity exceeds 13.7 million US dollars.

As a reference, combined with CoreWeave's financial reports and public contracts from companies such as IREN, the exclusive computing power packages including GPUs in the market are roughly priced at 8 million to 10 million US dollars per MW per year for top suppliers.

In other words, the price Anthropic pays to Volta is more than 30% higher than the common market level.

Why can Volta move so fast?

First, because it does not build the data center from scratch at all.

Tydal was originally a data center park of Bitdeer in Norway. After the AI demand exploded, Bitdeer started the transformation towards AI data centers since the end of 2025, including the preparation of power supply, computer rooms and related facilities.

So when Volta entered the project, what it faced was not a wasteland. The most troublesome work such as land acquisition, approval, grid connection and large-scale civil engineering has been mostly completed. What Volta needs to do is to bring in customers, capital and equipment, install GPU servers, and get the network and software running smoothly.

In this way, a large amount of time is saved directly.

Second, Volta prepares all the capital in advance.

The financing scale of a 100MW-level AI project is billions of US dollars. Under normal circumstances, for each project, you need to re-contact banks and funds, negotiate interest rates and guarantees, and conduct due diligence. It takes several months just to raise all the funds.

Volta did not do that. It has pre-built a 50 billion US dollar AI infrastructure plan with Azora. This is equivalent to preparing the capital pool first, and then going out to find projects.

This is a bit like a real estate developer. If every time you acquire a plot of land, you need to temporarily find LPs and negotiate bank loans, the speed will definitely not be fast. But if you already have a fund of tens of billions of US dollars in hand, you can push the project forward immediately once you find a suitable target, which is a completely different level of efficiency.

Volta has turned the slow work of "financing for each individual project" into a reusable financing channel.

To sum up, Volta's speed advantage actually comes from two points: using existing assets as much as possible, and preparing capital as much as possible in advance.

It doesn't sound complicated, but the vast majority of companies simply can't do it. This is largely related to the background of Volta's founders.

Ricardo Boada, co-founder and CEO of Volta, was previously a senior vice president at Brookfield, the world's largest alternative asset management company. He participated in large European data center projects such as Data4, and the scale of infrastructure investment involved in his career has exceeded 500 billion US dollars.

In the field of AI computing power capitalization, Brookfield is almost one of the asset management companies that have gone the farthest.

At the end of 2025, Brookfield launched a dedicated AI infrastructure plan, aiming to leverage up to about 1 trillion US dollars of assets through funds, co-investment and debt, with investment scope covering from energy and land to data centers.

To a certain extent, Volta almost transplanted this set of logic into a startup company.

The other co-founder, Sofía Gumucio, makes up for the capability on the capital side. She worked as a financial analyst focusing on mergers and acquisitions at the Spanish asset management company Azora in her early years. This early work connection also explains why Volta was able to get 50 billion US dollars of support from Azora shortly after its establishment.

The last piece of the puzzle is cloud technology. Having project operation experience and financing capabilities does not mean you can really run hundreds of thousands of GPUs smoothly. Therefore, Volta also acquired the technology and team of Genesis Cloud this year. Genesis Cloud has been committed to GPU-first cloud since 2018, and has operated GPU clusters for many years in the past.

Back to Volta, a16z's evaluation of it is very accurate:

Many new AI cloud companies are either good at capital and physical assets, or good at GPU operation and software, but there are not many teams that have the capabilities of power sourcing, financing, data center development and production-level cloud platform operation at the same time.

Looking back, the real value of Volta lies in integrating three groups of people who are good at "operating infrastructure assets", "raising capital" and "cloud operation" into the same company. After the AI infrastructure sector enters the era of 100-billion-dollar projects, the ability to get land, capital, customers and computing power delivery all done at the same time is a very scarce capability in itself.

This article is from the WeChat Official Account "Silicon-based Observer Pro", author: Silicon-based Observer, published with authorization from 36Kr.