Private Credit Is Now Financing the AI Data Center Boom as Banks Pull Back

In late April 2026, Oracle needed $16.3 billion to finance a single AI data center campus in Saline Township, Michigan — and the banks that have historically underwritten this kind of debt largely stayed on the sidelines. PIMCO, the world's largest bond fund manager, ended up anchoring roughly $10 billion of the deal after Bank of America sold the rest to a wider pool of investors. That single transaction is the clearest evidence yet that the AI buildout has outgrown the traditional bank lending model, and private credit is now the financing backstop for an industry that measures its capital needs in the hundreds of billions.

The thesis here is simple: the quality of AI data center debt is deteriorating just as the lenders willing to hold it are shifting from regulated banks to less-regulated private credit funds — a combination that concentrates risk exactly where oversight is thinnest.

Why banks are stepping back

TD Cowen reported that US banks have been actively retreating from Oracle's data center financing specifically, citing doubts about whether AI infrastructure demand will hold up at the scale Oracle is projecting. That skepticism isn't abstract: S&P Global Ratings has Oracle at BBB-negative and Moody's has it at Baa2 Negative, both below-investment-grade-adjacent outlooks that make regulated banks, which face capital charges tied to credit ratings, increasingly reluctant to hold this exposure on their own balance sheets. The Michigan deal isn't an outlier — it follows a $38 billion debt package for Oracle facilities in Texas and Wisconsin and an $18 billion deal for a New Mexico site, part of roughly $72 billion in total partner debt Oracle has assembled across four states for the Stargate joint venture with OpenAI and SoftBank.

The common thread across all of these deals: a meaningful share of the revenue backing this debt ultimately traces back to OpenAI, a company that has never posted an annual profit. Banks are pricing that risk by syndicating it away rather than carrying it.

Private credit is filling the gap — and it isn't free capacity

Private credit managers have record amounts of fundraised capital they're under pressure to deploy, and AI data centers offer exactly the kind of large, long-duration deal that fits their mandate. Ares is reportedly targeting more than $8 billion specifically for data center facilities spanning London, Japan, and Brazil. But this isn't risk-free capacity absorbing risk-free demand: Fitch's tracked US private credit default rate hit 5.8% for the twelve months through January 2026, the highest level since Fitch began tracking the metric. Voya Investment Management has already responded by explicitly limiting its data center credit exposure, citing doubts about whether AI demand will sustain the debt load being issued against it.

That's the mechanism worth watching: private credit is less regulated and less transparent than bank lending, which means the AI data center financing gap isn't closing so much as moving into a part of the credit market with weaker disclosure requirements and, per Fitch's own data, a rising default rate.

What this means for anyone exposed to the AI buildout

If you hold private credit fund exposure through a pension, insurance product, or direct allocation, data center lending is now a meaningful and growing slice of that book, and its performance is tied directly to whether OpenAI, Anthropic, and other AI labs can sustain the revenue growth needed to service debt issued against their compute contracts — not to the health of the broader economy. If you're evaluating a vendor or partner whose infrastructure depends on one of these financing structures, ask specifically whether their data center capacity is financed through bank debt or private credit, because a credit-market seizure in the private credit space would hit private-credit-financed capacity first and hardest. And if you're a policymaker or regulator, the Oracle deal is a useful case study in how capital that would have once required bank-level transparency is now flowing through vehicles that don't carry the same disclosure obligations — exactly the kind of blind spot that showed up in previous credit cycles before it became visible in a crisis.