Microsoft has announced a series of deals totaling tens of billions of dollars to rent computing capacity for its artificial intelligence ambitions. Meta has secured nearly $30 billion in financing to build a massive data center in Louisiana without taking on the debt directly. Google has also committed to renting computing power from a small company and then reselling some of that to OpenAI.
These deals had something in common: They allowed companies posting huge quarterly profits to reduce their financial exposure to the frenetic global expansion of data centers.
Also read:
Continues after advertising
They also signaled new ways for the biggest technology companies to maneuver to shift some of the risk of the AI boom onto the shoulders of upstarts eager for market share.
The strategies allow companies like Meta and Microsoft to quickly add computing capacity and then wait to see how demand for AI shapes up before committing to projects that could last decades.
Trillions of dollars are at stake as technology companies try to predict how much computing power AI will require years ahead.
If the big companies decide they ultimately don’t need all that computing after the agreements expire, smaller companies and their funders will be left with the consequences.
“Risk is like a tube of toothpaste,” said Shivaram Rajgopal, an accounting professor at Columbia Business School. “You press it here, it will come out somewhere else. It is always in the system, the question is where.”
These deals also add a level of mystery to data center financing because many of the companies that operate the centers for the tech giants are far from household names in Silicon Valley.
Continues after advertising
Some are privately held, do business with large startups, and borrow from private lenders—all of which provide less transparency about their strength.
Meta’s Louisiana data center project mixes many of these creative financing elements into a multibillion-dollar plan that is taking shape among rural areas in the state’s northeast.
Meta created a special purpose vehicle called Beignet Investor LLC and worked with Blue Owl Capital, a private credit manager, to borrow money for the project.
Continues after advertising
Meta was responsible for building the data center, but Blue Owl assumed 80% of the financing.
As part of the arrangement, Meta agreed to “lease” the data center from Beignet through a series of four-year contracts. This allows the tech giant to classify the financing as an operating cost rather than debt, according to financial documents.
As part of the deal, Meta is paying a premium to Blue Owl so it doesn’t have to borrow the money directly, said Solomon Feig, private credit lender at Pinnacle Private Credit. “Instead, Meta is renting risk,” he added.
Continues after advertising
Blue Owl primarily financed the project, called Hyperion, through a bond issue from Pimco, an asset manager. Pimco, in turn, sold so-called “Beignet bonds”, which mature in 2049, to clients such as insurance companies, pension funds, foundations and financial advisors. BlackRock also purchased part of the bonds.
“The key part of Meta’s strategy, in my view, is that they’re going to build as much of this as possible with what the industry calls OPM: other people’s money,” said Andrew Rocco, equity analyst at Zacks Investment Research.
If the AI boom slows, Meta could pull out of the deal as early as 2033. How much it might have to pay depends on the circumstances.
Continues after advertising
Blue Owl could find a new customer or sell the project, although the data center’s value could depreciate if demand for AI disappoints.
Meta promised to contribute enough money to, in practice, pay off the underlying debt without formally placing that debt on its balance sheet, according to S&P Global, the risk rating agency.
Other technology companies have expressed interest in similar financial structures. But because this was an unprecedented deal, Meta offered more protections than future deals might require, according to two people involved in the deal who spoke on condition of anonymity.
Rajgopal warned that the arrangement is reminiscent of other investment booms that have turned to private credit and special purpose vehicles, less transparent ways of raising funds than the traditional banking sector.
“I thought we solved the problem off-balance sheet,” he said, referring to the accounting methods used by banks before the dot-com bubble of the 2000s. “This is like having Groundhog Day all over again.”
Meta declined to comment.
Big tech companies are also striking giant deals with a new generation of data center providers known as neoclouds. Typically with three- to five-year contracts, these agreements give them more computing power quickly, without locking them into decades-long commitments.
By agreeing to shorter contracts, large companies are able to obtain computing power that appears in their financial reports as a day-to-day operating expense rather than a long-term capital investment, which can scare off investors.
Industry experts say the costs of expanding AI have become so high that it is no longer possible to mitigate most of the risk. So tech giants are spreading this risk.
“This is very cunning of them,” said Alex Platt, an analyst at investment bank DA Davidson. “There are only a handful of companies that can even do this.”
c.2025 The New York Times Company
