The "gold rush era" of American data centers
```
The AI wave is propelling the U.S. data center industry into a capital frenzy. Massive inflows of funds and new players, along with a constant stream of innovative financing structures, have fueled the boom. But beneath the surface, a huge gap between profit expectations and reality, the fragility of circular dependencies, and the lack of experience among newcomers are accumulating systemic risks in this gold rush.
On October 20, according to technology media The Information, the atmosphere at last week’s data center industry conference in Las Vegas was completely different from a year ago. OpenAI, xAI, and Meta have pledged to invest hundreds of billions of dollars over the next decade, and the focus of discussion has shifted from "difficulty finding land or power" to "who can build the most gigawatt-capacity" data centers. An investor group led by BlackRock and MGX acquired Aligned Data Centers—a data center operator just 12 years old—for a record $40 billion.
The report states that optimism hides real challenges beneath the surface. Take Oracle as an example: its AI cloud business' actual financials over the past five quarters show that current profit margins for leasing Nvidia chips lag target values by 15-20 percentage points. Industry insiders privately express caution, warning about "excessively circular capital flows" or "overreliance on a single company" in deal structures.
Innovative Financing Structures Become the Norm
To support astronomical levels of investment, the industry is inventing all kinds of creative financing methods.
Leaseback deals have become a new favorite: xAI buys Nvidia chips from its main investor Valor Equity Partners, then leases them back for use. OpenAI is also discussing similar structures with Nvidia—developing and managing data centers itself, but using leasebacks to cut costs and avoid paying premiums to Oracle and Microsoft.
According to the report, the essence of these deals is a risk-sharing mechanism that blurs the boundaries among customers, suppliers, and financiers, allowing capital to keep flowing into data center construction. The Aligned Data Centers acquisition was like a shot of adrenaline, inspiring more operators to look for buyers.
Meanwhile, Nvidia is not only a chip supplier but is also deeply involved in financing—providing funding to chip customers and data center projects, with the money ultimately flowing back in the form of chip purchases.
Industry insiders worry whether such circular capital flows are distorting real demand, and whether Nvidia’s dual role as both referee and player could create a market bubble. However, OpenAI’s recent commitment to use AMD chips and jointly design chips with Broadcom shows an intention to break Nvidia’s monopoly.
AI Companies Crossing Boundaries to Challenge Industry Rules
The most notable phenomenon in this data center "gold rush" is role reversal.
According to the report, Poolside, originally an AI programming startup, now claims to be building a 2-gigawatt data center, planning to lease part of it to AI cloud service provider CoreWeave, and claims to have solved the industry's most pressing bottlenecks. Startups like Fermi have jumped directly into multi-gigawatt-scale projects, betting they can outperform cloud giants like Google and Microsoft in speed and performance.
These new entrants—lacking traditional data center development experience—are challenging existing industry rules. Traditional data center developers are increasingly skeptical of newcomers’ abilities. Microsoft executives once told OpenAI they doubted Oracle could deliver the promised multi-gigawatt capacity.
Several industry insiders have been recruited by new entrants to "solve pressing operational challenges." Many predict a shake-up is coming, with overly aggressive projects set to collapse due to delays, power shortages, or unrealistic timelines.
But Profit Reality Puts Business Models to the Test
Nevertheless, real challenges are hidden beneath the optimism. Oracle projected upbeat revenue and margin forecasts at its annual cloud conference, but actual financial data from the past five quarters reveals a harsh truth—current profit margins from leasing Nvidia chips lag target values by 15-20 percentage points.
AI cloud service providers are in a race against time: they must purchase expensive Nvidia chips upfront, but customers only start paying after projects are completed and performance standards met. Power supply issues, equipment delays, and other uncontrollable factors can at any time disrupt plans.
When the roles of supplier, customer, and financier overlap, systemic risks accumulate.
Industry leaders privately warn against the fragility of such circular dependencies. When Microsoft chose to have Oracle take on part of OpenAI's server needs, the industry’s shrewdest players spoke with their actions: either they're pessimistic about long-term demand, or unwilling to shoulder excessive risk.
Analysts point out that in this gold rush, Nvidia sits firmly in the "shovel seller" position; traditional cloud giants have technical reserves and the ability to withstand risk; while newcomers face the greatest uncertainties. Only those with real technical strength, ample capital, and risk management experience will still stand tall when the tide recedes.
Risk Disclosure and DisclaimerThe market involves risk, and investments need to be made cautiously. This article does not constitute personal investment advice and does not take into account the individual investment objectives, financial situations, or needs of any particular user. Users should consider whether any opinions, views, or conclusions in this article are suitable for their circumstances. Investing on this basis is at your own risk. ```