Banks pull back from AI lending as the gap between build cost and revenue widens

Wall Street appears to be hitting its limit on AI, and financiers are starting to say the quiet part out loud, that all of this AI spending does not look like it can pay off, at least not for everyone, Axios reported¹. The gap between what it costs to build AI infrastructure and what companies are actually paying to use it is widening, and anxiety is surging as the debt piling up to fund data centres and chips looks increasingly disconnected from economic reality¹. AI providers are projected to spend $7 trillion on infrastructure through 2030¹.

CoreWeave stock was falling again on Friday even after Nvidia’s chief executive, Jensen Huang, expressed confidence in the cloud-computing company, Barron’s reported². Reuters reported that CoreWeave has relied heavily on private-credit investors, banks and bond markets to fund its rapid expansion⁵. CNBC asked whether the AI sector is losing its rose-coloured glow, and reported Huang saying his company must cooperate with Chinese AI businesses to manage potential risks⁸.

Why it matters to investors

The transmission mechanism is credit, not equity. Interactive Brokers reported fielding questions about the effect CoreWeave and its bonds are having on broader markets, and noted that in more normal times credit concerns at a single corporate borrower should not resonate dramatically through global equity and fixed-income markets⁷. The AI build-out has become such a key factor in markets and in the global economy that tremors in a relatively marginal player can resonate wildly⁷.

That is why the pricing of default protection has become a market-wide signal. Reuters published an explainer on why credit default swaps are unnerving AI investors, setting out that the instruments pay out if a borrower fails to meet its debt obligations, and that bond investors expect interest payments and repayment of principal at maturity⁴. The New York Times wrote about the fallout from what it called the neocloud wipeout³, and Bloomberg’s opinion newsletter argued that private markets are the new AI markets⁶.

What to watch

The question is whether the pullback Axios describes shows up in terms rather than talk: who funds the next data-centre build, at what spread, and whether banks step back while private credit steps in^1,5,6. Huang’s stated confidence has not stopped CoreWeave’s shares falling². Reuters’ account of CoreWeave’s contracts and the lenders financing its build-out is the map of who carries that exposure: private-credit investors, banks and the bond market⁵.