Big Tech’s AI Infrastructure Gamble and the Quiet Shift of Risk
As artificial intelligence reshapes the global technology landscape, the race to build the infrastructure that powers it has become one of the most expensive bets in corporate history. This fall, some of the world’s biggest technology companies announced a series of deals worth tens of billions of dollars to secure computing power for AI, without fully owning the assets or carrying the long-term financial burden on their balance sheets.
Microsoft revealed multiple agreements to lease massive amounts of computing capacity. Meta secured nearly $30 billion in financing to build a huge data centre in Louisiana while avoiding direct debt. Google committed to renting computing power from a smaller company and reselling part of it to OpenAI. While the deals differed in structure, they shared one crucial feature: they reduced financial risk for companies that already generate enormous quarterly profits.
These arrangements signal a fundamental shift in how big tech firms are managing the uncertainty of the AI boom. Rather than betting everything on long-term infrastructure projects that could last decades, they are increasingly pushing risk onto smaller partners, private lenders, and investors eager to participate in the AI gold rush.
The Trillion-Dollar Question: How Much Computing Will AI Really Need?
At the heart of these deals lies a massive unknown. No one can accurately predict how much computing power artificial intelligence will require five, ten, or twenty years from now. Trillions of dollars are potentially at stake as companies try to estimate future demand for AI models, cloud services, and enterprise tools.
If demand continues to grow at today’s pace, the investments may look visionary. But if the market cools or efficiency improves faster than expected, the industry could end up with vast amounts of underused infrastructure. In that scenario, the consequences would not fall evenly.
Large tech firms with flexible contracts can walk away or renegotiate. Smaller data centre operators, private credit firms, and institutional investors may be left holding assets that are difficult to repurpose and rapidly depreciating in value.
As Columbia Business School accounting professor Shivaram Rajgopal put it, risk does not disappear—it merely moves. When pushed out of one place, it emerges somewhere else in the financial system.
The Growing Opacity of Data Centre Financing
Another striking feature of this new wave of AI infrastructure deals is how opaque they can be. Many of the companies building or financing these data centres are not household names. Some are privately held, rely heavily on private credit, and operate outside the transparency typically associated with large public corporations or traditional banks.
This lack of visibility makes it harder for outsiders to assess the long-term stability of these projects. It also raises broader questions about systemic risk, especially as private credit plays an increasingly central role in funding AI infrastructure.
The case of Meta’s Louisiana data centre illustrates how complex and creative these arrangements have become.
Inside Meta’s Louisiana Data Centre Deal
Meta’s multibillion-dollar data centre project in northeast Louisiana combines several financial techniques designed to limit the company’s direct exposure. To structure the deal, Meta created a special purpose vehicle called Beignet Investor LLC. Through this entity, it partnered with Blue Owl Capital, a private credit firm, to secure financing.
While Meta took responsibility for constructing the data centre, Blue Owl provided approximately 80 percent of the funding. Instead of owning the facility outright, Meta agreed to lease it through a series of four-year rental agreements. This structure allows Meta to classify the payments as operating expenses rather than long-term debt, a distinction that matters greatly in financial reporting.
According to industry lenders, Meta effectively chose to pay a premium in exchange for avoiding debt on its balance sheet. Rather than borrowing directly, the company outsourced much of the financial risk to its partners.
Bonds, Private Credit, and “Other People’s Money”
Blue Owl financed much of the project, known as Hyperion, through a bond offering arranged by Pimco, one of the world’s largest asset managers. Pimco sold these so-called “Beignet bonds,” which mature in 2049, to institutional investors such as insurers, pension funds, endowments, and financial advisers. BlackRock also purchased a portion of the bonds.
For Meta, the strategy reflects a broader industry mindset. As analysts have noted, the company is eager to build AI infrastructure using what financiers often call “OPM”—other people’s money. This approach allows Meta to scale quickly while preserving financial flexibility.
If AI demand were to slow, Meta could exit the deal as early as 2033. While it has made commitments to ensure bondholders are repaid, the underlying risk remains tied to the long-term value of the data centre itself.
Echoes of Past Financial Booms
Not everyone is comfortable with how familiar these structures feel. Rajgopal and other experts have warned that the combination of private credit, special purpose vehicles, and off-balance-sheet financing resembles strategies used in previous investment booms.
In the years leading up to the dot-com crash, complex accounting methods helped obscure risk until market conditions changed abruptly. Critics worry that history may be repeating itself, albeit in a different technological context.
Although Meta has declined to comment publicly, the deal has drawn close attention across the tech and finance industries, particularly as other companies explore similar arrangements.
The Rise of Neocloud Providers
Beyond bespoke financing structures, big tech firms are increasingly relying on a new class of data centre operators known as neoclouds. These companies specialize in providing high-performance computing capacity on shorter contracts, typically lasting three to five years.
For companies like Microsoft, these deals offer speed and flexibility. Computing power can be added quickly and treated as a routine operating expense rather than a capital-intensive, decades-long commitment that might unsettle investors.
Over the past few months, Microsoft has signed a string of massive neocloud agreements. These include a $17 billion deal with Nebius, a company linked to the founders of Russian internet firm Yandex, and a $23 billion commitment to Nscale, a privately held British provider. Additional agreements with Iren, a former bitcoin miner, and Lambda further underscore the scale of Microsoft’s strategy.
Flexibility as a Core Strategy
Microsoft executives have emphasized that flexibility is central to their global infrastructure approach. The company is racing to meet surging customer demand while remaining cautious about locking itself into assets that may not be needed in the future.
That caution became evident when Microsoft quietly paused some of its own data centre construction last fall. The pause coincided with changes in its relationship with OpenAI, although Microsoft has denied a direct connection.
Previously, Microsoft had an exclusive arrangement to supply computing power to OpenAI. More recently, it has allowed OpenAI to source some of its infrastructure from other providers, including Oracle. The newer neocloud deals are expected to help support the enormous computing commitments OpenAI has made to Microsoft.
OpenAI, CoreWeave, and the Web of Dependencies
Microsoft and Google both have agreements to provide computing power to OpenAI, often through intermediaries such as CoreWeave, the largest neocloud provider in the market. OpenAI itself has committed to purchasing more than $22 billion in computing capacity directly from CoreWeave.
This web of dependencies highlights how interconnected the AI infrastructure ecosystem has become. While large tech firms remain at the center, much of the financial and operational risk is distributed across a network of smaller companies and investors.
A New Phase of the AI Boom
The rapid expansion of AI infrastructure marks a new phase in the technology industry’s evolution. Rather than simply building and owning everything themselves, companies like Meta, Microsoft, and Google are experimenting with financial structures that prioritize speed, flexibility, and risk management.
Whether this approach proves sustainable will depend on how AI demand evolves in the coming years. If growth continues unabated, these deals may be remembered as smart, strategic moves. If not, they could expose vulnerabilities in a system that has quietly shifted risk away from its most powerful players.
For now, one thing is clear: the AI boom is not just a technological revolution—it is also a profound financial experiment, with consequences that may take decades to fully unfold.
Click Here to subscribe to our newsletters and get the latest updates directly to your inbox