AI capex is spilling into Wall Street's credit plumbing

As big technology borrowers raise ever larger sums to build AI infrastructure, Wall Street banks are trading more credit protection simply to keep the business on their books. Bloomberg reported that the borrowing tied to AI has already climbed to more than $250 billion globally, a scale large enough that dealer balance sheets are running into ordinary credit limits even when the names involved still sit near the top of investment-grade rankings. What began as an equity story about Nvidia multiples and venture valuations is now turning into a plumbing story about who can warehouse hyperscaler risk, for how long, and at what price.

This shift matters because the financing chain has changed. A bank that underwrites a hyperscaler bond or loan now has to think not only about fees and client relationships, but also about how quickly it can distribute the paper, what exposure remains on its own book, and whether a hedge in credit-default swaps will cost more next month than it does today. The AI buildout has not become a default scare. It has become a concentration trade. Small shift, large consequence.

But the analyst view is less about trading opportunity than about absorption. Reuters reported in January that Bank of America expected hyperscalers to drive about $140 billion of US corporate bond supply in 2026; a March follow-up lifted that forecast to $175 billion after Amazon’s bond sale. By May, Barclays told Bloomberg that big tech’s debt binge was already testing how much fresh supply the high-grade market could comfortably take. That does not answer every question about spreads, but it does answer one of the most important ones from the sell side: the market is no longer debating whether more AI debt is coming. It is debating how much extra hedging the market needs to digest it.

Why the hedging trade is growing

The bank-desk perspective is straightforward. When a handful of huge borrowers keep returning to the market for data-centre funding, underwriters do not just earn fees, they accumulate recurring exposure to the same cluster of names. A credit-default swap, or CDS, is the cleanest way to trim that exposure without stepping away from the client. Bloomberg’s reporting and its March report on JPMorgan’s AI CDS basket suggest the trade is increasingly less about making a grand macro statement on an AI bubble and more about keeping risk budgets flexible enough to stay in the next deal.

Mandell’s explanation matters more than the headline number on issuance.

“They’re trying to avoid being constrained by credit limits.”
— Matt Mandell, head of US single-name CDS at Bank of America, in Bloomberg

At first glance, that sounds mundane. On another level, it is the whole story. If banks were simply nervous about credit quality, they could scale back exposure and move on. Instead, they are looking for ways to preserve capacity. That implies a market where client demand is too lucrative to ignore, but too concentrated to leave unhedged. Bloomberg said monthly hyperscaler CDS volume at Bank of America is running 10 times higher, while notional traded on Meta CDS reached $534 million in the first quarter. Those are not recession-era distress numbers. They are flow numbers, the kind produced when balance-sheet management becomes a daily operating issue.

Bank plumbing explains much of the current flow. Bloomberg’s February report on new AI-linked derivatives described desks building products for investors who want to hedge or express a view on the buildout, but the current wave looks driven first by banks and counterparty-risk desks trying to keep room open for the next financing. The answer to the insider question, how much of this demand is bank hedging rather than outright bearish positioning, appears to be: a lot of it. The hedge fund community is happy to meet that demand if the pricing is attractive enough.

Andrew Weinberg of Saba Capital Management put the appeal bluntly.

“It’s the best opportunity in AA credit default swaps in a very long time.”
— Andrew Weinberg, portfolio manager at Saba Capital Management, in Bloomberg

For hedge funds, that is not a prophecy of imminent trouble at the largest tech borrowers. It is a comment on market structure. When demand for protection rises faster than the usual supply of sellers, even top-tier names can begin to trade like special situations. The opportunity is not that hyperscalers are about to crack. It is that the hedging need itself can cheapen or distort CDS pricing, especially around names that are still widely treated as pristine credits.

Seen through the skeptic’s lens, the signal is that AI financing is beginning to stress the mechanics around the borrowers before it stresses the borrowers themselves. That distinction matters. Markets rarely wait for a credit event to reprice a theme. They often reprice the carrying costs, the distribution channels and the liquidity first.

Where the spillover lands next

The next question is where that repricing shows up. Reuters reported this month that hyperscale US corporate bond issuers had already flooded the market with $18 billion of new paper in a single stretch, while Bloomberg reported that Barclays saw the high-grade market struggling to accommodate all of the financing needs. In other words, the credit market is still working. It is just working harder.

Convergence between the analyst and skeptic views starts there. Sell-side strategists worry about absorption and spread pressure. Credit investors worry about what gets financed at the margin once the obvious, top-quality names have taken their share. The largest hyperscalers can still fund themselves cheaply, even after a heavy run of issuance. The weaker link may be the ecosystem around them: suppliers, data-centre builders, power and networking plays, and private-credit structures that assume AI demand stays both strong and orderly.

Morgan Stanley strategists Vishwas Patkar and Joyce Jiang captured the risk in a sentence.

“Quality deterioration remains a risk.”
— Vishwas Patkar and Joyce Jiang, Morgan Stanley strategists, in Bloomberg

Placed beside Fast Company’s argument on Oracle’s debt-heavy expansion and Bloomberg’s discussion of AI borrowing moving up Wall Street’s risk list, that warning becomes more specific. The core issue is not that Microsoft, Amazon or Meta are suddenly fragile. It is that a financing regime built around repeated jumbo issuance from a narrow group of borrowers changes pricing for everything around them. Higher yields for longer would sharpen that effect, because every extra basis point makes warehousing, distributing and hedging the exposure more expensive.

A second spillover fits neatly with scramnews’s recent IPO coverage. Bloomberg’s March report on JPMorgan’s CDS basket showed Wall Street already trying to package the risk more efficiently. The Financial Times argued that AI has penetrated every corner of financial markets, while another Bloomberg report tied the next wave of mega-capital raises to companies such as SpaceX and, by context, the private AI names orbiting OpenAI and Anthropic. If that pipeline keeps growing, banks are not just underwriting more AI risk. They are helping turn it into a permanent asset class, complete with its own hedges, baskets, basis trades and balance-sheet constraints.

One ring out is probably where the weak link sits if rates stay higher for longer. The hyperscalers are large enough to keep tapping public markets. The strain shows up in the financing margins around them, where spreads are thinner, structures are less standardised and private lenders have done more of the incremental work. Bloomberg video analysis with Danielle Poli of Oaktree made the same point from another angle: hidden risk in the AI boom is less about a single issuer missing a payment than about how much debt the market is being asked to absorb at once.

For now, the signal to watch is not an equity multiple. It is the cost and volume of protection. If banks have to keep buying more CDS to stay in the next hyperscaler financing, then AI infrastructure funding has moved decisively beyond the stock market narrative that carried it through 2025. The money is still chasing compute. But the risk is now being priced, sliced and redistributed through Wall Street’s credit machine. Once that happens, the AI boom stops being just a tech story. It becomes a market-structure story.