Tech companies have taken more than $120 billion in data center spending off their balance sheets using financing vehicles from Wall Street investors, sparking concerns about the risks of the big bet on artificial intelligence.
Elon Musk’s Meta, xAI, Oracle and data center operator CoreWeave are leading the way in complex financing deals that shield companies from the large loans needed to build AI data centers.
Financial institutions like Pimco, BlackRock, Apollo, Blue Owl Capital and US banks like JPMorgan provided at least $120 billion in debt and equity for the IT infrastructure of these tech groups, according to a Financial Times analysis.
This money is routed through special purpose holding companies called SPVs. The influx of funding, which doesn’t show up on tech companies’ balance sheets, could obscure the risks these groups are taking – and who will bear the consequences if demand for AI disappoints.
SPV structures also increase the risk that financial stresses facing AI traders in the future will spill over to Wall Street in unpredictable ways.
“Eighteen months ago, this would have been inconceivable. Today, it’s the norm,” a senior executive at one of the major financial institutions said of the tens of billions of dollars flowing into SPVs to finance data centers.
“The tech industry can access much more capital than anyone else because of its credit profile.”
Silicon Valley giants have traditionally generated a lot of money and had little debt, giving these companies excellent credit ratings and high investor confidence.
The race for computing power for advanced AI, however, has led technology groups to borrow larger sums than ever. The use of private equity financing through off-balance sheet structures protects companies’ credit ratings and improves their financial metrics.
Meta closed the largest private data center credit deal in October, worth $30 billion, for its proposed Hyperion facility in Louisiana. The project created an SPV called Beignet Investor, in partnership with the financing company Blue Owl Capital.
The SPV raised $30 billion, including about $27 billion in loans from Pimco, BlackRock, Apollo and others, as well as $3 billion in equity from Blue Owl.
The deal meant Meta could borrow $30 billion without any debt showing up on its balance sheet. This made it easier for the company to raise another $30 billion in the corporate bond market a few weeks later.
Oracle led the way by structuring large debt deals through third parties to pay for its massive data center power leasing commitments to OpenAI.
Larry Ellison’s technology group has partnered with builders and financiers such as Crusoe, Blue Owl Capital, Vantage and Related Digital to build numerous data centers that will eventually be owned by SPVs.
Its off-balance sheet financing deals include approximately $13 billion invested by Blue Owl and JPMorgan, including $10 billion in debt, in an SPV that owns its OpenAI facilities in Abilene, Texas; a $38 billion debt package to pay for two data centers in Texas and Wisconsin; and an $18 billion loan for a site in New Mexico.
In each case, Oracle agreed to lease the facilities from the SPVs. In the event of default, creditors would have recourse to the assets – the data center, the land it is located on and the chips that power it – and not the companies that run the sites.
Raising debt off balance sheet via SPV has become more popular as the amount of capital needed to finance AI infrastructure has skyrocketed, straining tech companies’ cash reserves. Morgan Stanley estimated that $1.5 trillion in external financing was needed to fund tech companies’ AI projects.
In many cases, investors in these data center deals have been convinced that the financial risk still falls on the tech company leasing the site if demand for AI services falls, causing the value of these massive computing facilities to decline.
In the case of Beignet Investor, Meta owns 20% of the SPV and has given a “residual value guarantee” to the other investors. This means the social media group would have to repay SPV investors if the value of the data center falls below a certain level at the end of the lease and Meta decides not to renew.
Musk’s AI startup, xAI, is seeking a similar structure as part of a $20 billion raise, including up to $12.5 billion in debt. The SPV will use the money to buy graphics processing units from Nvidia and lease them to xAI.
CoreWeave said in March it had created an SPV to fulfill an $11.9 billion contract to provide computing power to OpenAI, which would “incur debt to finance its obligations.” In July, it borrowed $2.6 billion to finance its contracts with OpenAI.
Private equity investors are eager to participate in the AI boom, increasing demand for these frameworks. Tech companies had borrowed about $450 billion from private funds at the start of 2025, $100 billion more than in the previous 12 months, according to UBS.
This year, about $125 billion was invested in “project finance” deals – long-term financing of infrastructure projects – like the Meta and Blue Owl deal, UBS said.
Data center construction has become largely dependent on private credit markets, a rapidly expanding $1.7 trillion sector that has itself raised concerns due to sharp increases in asset valuations, illiquidity and borrower concentration.
“There are already risky loans and underlying credit risk built up in private credit,” said a banker familiar with data center financing deals. “This creates a very interesting setup for the next few years, as there are two significant risks to the outlook that are becoming increasingly intertwined.”
The risk in these structures depends largely on their extension. If multiple AI companies use SPVs, the tension could spread simultaneously to the private credit funds that back them, with little transparency.
The rise of AI data centers also depends largely on a small group of customers. OpenAI alone has completed over $1.4 trillion in long-term IT commitments across most of the industry’s major players.
Lenders from several different data centers could therefore be exposed to the same risks in the event of the bankruptcy of one of the companies involved. They also face uncertainties regarding energy access, AI regulation, or technological changes that render the current generation of AI hardware obsolete.
Not all hyperscale data center companies have followed the trend. Google, Microsoft and Amazon already had large, established data center companies before the AI boom and continue to finance construction with their own money.
While Google and Amazon have recently turned to bond investors to directly raise more debt, all three companies have yet to disclose significant SPV financing.
Wall Street is also moving toward more obscure structures around data center transactions.
Several tech bankers said they have even seen AI debt securitization deals in recent months, in which lenders package loans and sell slices of them, known as asset-backed securities, to investors. Two bankers estimate that these transactions currently amount to single-digit billions of dollars.
These transactions spread the risk of data center loans to a much broader group of investors, including asset managers and pension funds.
Matthew Mish, head of public and private credit strategy at UBS, said most investors “believe it’s actually a good thing to end up with hyperscaler risk” given these companies’ strong balance sheets and credit profiles.
But Mish added that SPV financing “still adds unpaid liabilities” for tech companies, meaning “overall credit quality for the hyperscaler would be worse than currently modeled.”
About The Author
