What Is an AI Data Center SPV — and Why Are Tech Giants Using Them?

When Meta and BlackRock announced a $14 billion AI data center campus in El Paso, Texas, the striking detail wasn’t the price tag — it was that Meta will own only 20% of it. The rest belongs to BlackRock-managed funds, financed through a special-purpose vehicle (SPV): a standalone legal entity created solely to build, own, and lease the facility back to Meta. It’s how a growing share of the AI industry’s data centers now get built — with the debt and the ownership pushed off the tech giant’s own books and onto outside investors.

How the structure works

An SPV is a shell company set up for one narrow purpose — here, to own a single data center. In the El Paso deal, Meta contributed land and partly finished construction (valued around $2.3 billion) for its 20% stake, while BlackRock’s funds put in roughly $4.9 billion in cash. The vehicle — reportedly named Sopaipilla Investor — then borrowed the remaining $12.5 billion from bond investors, with debt maturing as far out as 2048. Meta didn’t take out this loan itself. Instead, it signed a long-term lease to rent the finished data center capacity back from the SPV, and the rent payments are what service the debt.

Because Meta neither owns a majority of the entity nor guarantees its debt outright, accounting rules generally don’t require it to fold the SPV’s finances into its own balance sheet. The $12.5 billion in borrowing exists, but it shows up in the venture’s books, not Meta’s — an arrangement finance professionals call off-balance-sheet financing.

Why tech giants are doing this

Building AI infrastructure at the pace the industry wants costs more than any one company can comfortably borrow directly. The four largest AI spenders — Amazon, Alphabet, Meta, and Microsoft — are projected to spend over $600 billion combined on capital expenditure in 2026, and Morgan Stanley estimates hyperscalers could issue roughly $400 billion in corporate bonds in that year alone. Taking on that much debt directly would swell a company’s reported leverage and could pressure its credit rating, raising borrowing costs across the whole business.

SPVs solve this by inviting in investors who specifically want long-term, contracted income — asset managers like BlackRock, Blue Owl Capital, Pimco, and Apollo — to fund the entity, while the tech company simply pays rent. On the income statement, a capital expenditure effectively turns into a lease payment, which can make cash flow and profitability metrics look healthier to public investors. Meta has used this exact template more than once: an earlier Louisiana project, financed with Blue Owl Capital, gave Meta the same 20% stake alongside roughly $27 billion in debt. Elon Musk’s xAI has raised financing through a comparable structure, borrowing against its Nvidia chips rather than a building.

Why it worries economists and regulators

The practice is legal and, unlike the fraud at the center of the Enron scandal, it’s disclosed in financial-statement footnotes and regulatory filings rather than hidden. As one analyst put it, Enron’s failure wasn’t having special-purpose entities — it was concealing them. Still, three concerns keep coming up.

The first is opacity: the obligations are real, but investors have to dig through footnotes to see how much leverage a company is actually carrying. The second is risk transfer — if demand for AI computing ever slows, it’s the lenders and asset managers behind these vehicles, not the tech giants, who are first in line to absorb losses, and some of that private credit ultimately flows from insurers and pension-linked funds. The third is scale: JPMorgan estimates that AI and data-center borrowers already make up about 14.5% of its $10 trillion investment-grade bond index, roughly $1.5 trillion in exposure, and that number is growing quickly. Credit-rating agencies have already taken notice — S&P Global cited data-center lease commitments as a factor in downgrading Oracle’s outlook.

None of this means an SPV-financed data center is a bad investment or a hidden fraud. But it does mean that when a company says it’s “only” a 20% owner of a $14 billion facility, the other 80% — and the debt behind it — hasn’t disappeared. It’s simply sitting on someone else’s balance sheet instead, in exchange for a steady rent check.

In the news

The Meta–BlackRock El Paso venture is the latest example of this pattern, and it isn’t the only recent one: Nvidia is reportedly weighing a $250 billion guarantee to help back OpenAI’s own Ohio data center financing — another sign of how much of the AI buildout now runs through guarantees, leases, and financing vehicles rather than straightforward corporate spending. For the related practice of chipmakers and AI labs funding each other in circular deals, see our explainer on circular AI financing; for the broader spending pressure driving all of this, see what AI capex is doing to Big Tech’s cash.

Sources: Forbes, Bloomberg Tax, Construction Review Online, BNN Bloomberg, OilPrice.com.