Analysis
Two Information reports published Wednesday sketch the same story from opposite ends: major banks including SocGen and MUFG are pulling back from data-center loan deals, even as CleanSpark -- the company building a data center for Meta -- had to offer investors "big concessions" to land financing earlier this month. Between them, the two pieces describe a credit market that's still open to AI infrastructure debt, but increasingly at a higher price and with more selective lenders.
What's Actually Happening
According to The Information, some of the largest lenders to the AI data-center buildout are becoming choosier about which deals they take on, as market volatility and rising yields give them room to pick safer bets over more speculative ones. Separately, The Information reported that Meta's recent $12.5 billion bond offering for its El Paso data center priced at higher yields than a similar earlier deal -- a widening gap that reflects growing lender caution after months of heavy AI-linked borrowing, compounded by a selloff in AI stocks.
“Orders for CleanSpark's bond exceeding demand 4x over is a sign of a market still eager for yield, not one seizing up.”
The Twist: Demand Is Still There
What complicates the "credit is tightening" narrative is that demand for this debt hasn't actually dried up. The first junk-bond offering tied to a Meta data center -- CleanSpark's roughly $2.28 billion debt sale -- drew more than $10 billion in orders, over four times the deal size, Bloomberg reported, with high yields pulling in investors hungry for returns. That's the real mechanism here: it's not that lenders won't fund AI data centers, it's that the price of that capital is rising as lenders price in more risk, and borrowers like CleanSpark are paying for it in concessions and yield rather than being shut out entirely.
Numbers In Context
This is the same dynamic Pulse flagged when covering Nscale's $3.36 billion pre-IPO debt financing and the broader wave of pre-IPO financing rounds across the AI infrastructure sector: debt has become the preferred funding instrument for capital-intensive AI buildouts precisely because GPUs and data centers make collateralizable assets, but that only works as long as lenders trust the revenue backing them. GMI Cloud closing $445 million in debt financing the same week shows the market for this kind of paper is still functioning -- just with tighter underwriting than six months ago.
Counterweight
None of this signals an imminent AI infrastructure credit crunch. Orders for CleanSpark's bond exceeding demand 4x over is a sign of a market still eager for yield, not one seizing up. The real risk sits further out: if rising financing costs stack on top of the aggressive capacity buildout multiple hyperscalers have already committed to, the companies whose revenue projections don't materialize on schedule will be the ones that feel today's wider spreads hardest -- a distinction between announced capacity and actually-monetized capacity that matters more with every rate-driven basis point added to the cost of debt.
