In funding AI data centers, over $120 billion of off-balance sheet corporate spending has been shifted using complex financing structures.
This method relies on creating private entities (known as SPVs or special purpose vehicles), where the purchase of land, buildings, and equipment (such as chips) is financed through these entities, and then tech companies lease these assets instead of owning them outright. This way, large debts remain off the main financial statements of the companies.
These deals have attracted massive private capital from institutions like PIMCO, BlackRock, Apollo, and Blue Owl Capital, along with major U.S. banks like JPMorgan.
Previously, Silicon Valley companies relied on cash flow and low debt, but the race for AI computing power has completely changed that, as building data centers now requires tens of billions of dollars each time.
Meta began its strategy in October with a $30 billion private credit deal for its Hyperion data center in Louisiana, through a private entity called Beignet Investor created with Blue Owl Capital. This entity raised about $27 billion in loans from big investors and **$3 billion** in equity, without any of this debt appearing on Meta's balance sheet, allowing it to raise an additional $30 billion in bonds in November.
Oracle, on the other hand, leases computing power to OpenAI, and has therefore partnered with developers and financiers such as Cruz, Vantage, Related Digital, and Blue Owl to develop several data centers, each with its own separate private entity. Examples include: an entity in Abilene, Texas, that raised about $13 billion, and a $38 billion debt deal for sites in Texas and Wisconsin, and **$18 billion** for a site in New Mexico. In all cases, Oracle leases the facilities while lenders retain rights to the assets.
Elon Musk's xAI takes a similar approach, raising $20 billion (including up to $12.5 billion in debt) through a private entity that buys NVIDIA processors and leases them to the company.
CoreWeave also did this, creating a private entity in March to meet a $11.9 billion contract to provide computing power to OpenAI, then borrowing $2.6 billion in July to finance these commitments.
There remains strong interest from private capital, with estimates indicating that tech companies borrowed about $450 billion from private equity funds by early 2025, with around $125 billion flowing in project financing deals during the year.
Building data centers now heavily depends on the $1.7 trillion private credit market, amid growing concerns about valuations, liquidity, and borrower concentration.
Interestingly, companies like Google, Microsoft, and Amazon are still avoiding these structures so far, financing their data centers through cash flow and traditional bonds.