Musk says lenders have nearly stopped funding AI data centres! The bills are enormous

The AI boom runs on two things: computer chips and the money to buy them. This week Elon Musk, who is building one of the largest AI operations in the world, suggested the money is starting to get harder to find — even as he announced a colossal new pile of it.

The $40 Billion And The Warning

Musk said on his platform X that SpaceX had closed a roughly $40 billion debt package, described as primarily 'callable' debt, to fund purchases of AI chips for delivery through early 2027. In the same posts he warned that lenders have nearly stopped providing credit to AI data-centre firms that do not have substantial revenue from outside the AI build-out itself.

A note of caution: almost every figure in the SpaceX deal comes from Musk's own posts, with no independently reported lender names, pricing or terms. Treat the specifics as his account until confirmed.

The Signs He Is Not Alone

What makes the warning hard to wave away is that other, independent data points to the same strain.

Of roughly $80 billion in data-centre-related bonds issued so far in 2026, close to half — about $37 billion — carries a junk rating, below investment grade, according to an analysis of market data reported by Forbes. CoreWeave, one of the best-known specialist AI cloud providers, has seen around $4 billion of its corporate debt trade below 80 cents on the dollar, a distressed level, and its shares fell sharply in a single week even after a public vote of confidence from Nvidia's chief executive. Several infrastructure companies have delayed or shelved stock-market listings in recent weeks.

Why It Matters

The entire AI expansion assumes that the enormous cost of data centres and chips can be financed now and paid back later out of future revenue.

If lenders lose their nerve — demanding higher interest, shorter terms, or simply walking away — the build-out slows, and the companies most exposed are those that borrowed heavily against revenue they do not yet earn. That is why a warning from someone raising tens of billions himself carries weight: he is describing the market he is actively borrowing in. Comparisons to WeWork, the property company whose debt-fuelled expansion collapsed, are now being made openly, with one outlet citing an estimate that trillions of dollars of AI infrastructure spending could be at risk of impairment by 2030.

The Other Side

In fairness, there are real reasons this is not simply a crash waiting to happen. Demand for AI computing is genuinely enormous and still growing, the largest players are backed by some of the most profitable companies on earth, and 'callable' debt of the kind Musk described can be a sensible way to finance equipment with a known delivery schedule. A cooling of the most speculative lending — junk-rated bonds for unproven providers — can be a market disciplining itself rather than seizing up. And a warning is not a default: no major AI data-centre collapse has happened.

The counterweight is that bubbles are usually obvious only in hindsight, the sums involved now are historically large, and 'the demand will surely come' is exactly the assumption every infrastructure boom makes right up until it does not.

A note on this report: the AI assistant writing this article is made by Anthropic, one of the companies whose own vast compute spending — detailed this week in its IPO filing — is part of the very build-out described here.

What To Watch

Whether borrowing costs for AI infrastructure keep rising and more listings get pulled. Whether a specific provider actually defaults, which would turn a warning into an event. And whether the revenue that is supposed to justify all this borrowing arrives fast enough to keep the lenders in the room.