Key points
- Top-rated corporate credit has split into two markets
- AI borrowers face pickier buyers than their ratings suggest
- Investors say they are "starved for anything ex-hyperscaler"
The market for the highest-rated corporate bonds has quietly split into two worlds. Investors are treating bonds from AI-related companies warily, while competing hard to buy debt from traditional financial and industrial borrowers, Reuters reported on Tuesday.
The investors Reuters spoke with were focused less on near-term default risk than on the volume and unpredictability of borrowing needed to fund AI infrastructure. None of them suggested the hyperscalers were in danger of missing a payment. That supply is what is pushing buyers to demand bigger concessions on new deals and to rethink how much of any single borrower they are willing to hold.
The scale explains a lot. Gross debt issuance from hyperscalers is expected to reach a record $420 billion next year, up 60% from 2026 estimates, according to Goldman Sachs data cited by Reuters. Overall US corporate issuance through August reached $1.8998 trillion, up 29.8% from a year earlier, according to SIFMA. Gross issuance includes refinancing as well as new borrowing, but it still measures the supply investors must absorb. The whole market is busy. The AI slice is simply growing faster than buyers want to absorb it.
"We're being very selective in terms of how we invest within hyperscaler debt," said Colby Stilson, head of fixed income at Brown Advisory in London. His conviction has to be high, he said, "because of the coming supply and because of the lack of visibility into that return on invested capital."
A premium on the safest names
The gap shows up cleanly in spreads, the extra yield borrowers pay over government debt. Bonds from AI-related issuers have stayed persistently wider, at around 115 basis points, according to Goldman data, versus 78 basis points for the broader investment-grade market, according to ICE BofA. That is a real premium to lend to companies that, on paper, look about as solid as borrowers get.
Lon Erickson, a portfolio manager at Thornburg Investment Management, said bonds from big AI spenders such as Meta Platforms (META) and Alphabet (GOOGL) have consistently traded wider than similarly rated peers, even though both throw off enormous cash and carry strong balance sheets. "Investors are only able to digest so much, so fast," he said. The premium, he added, reflects an expectation that these borrowers will keep coming back to the market as AI capital spending climbs.
We have watched this play out deal by deal. When Alphabet launched a $25 billion bond sale in August, hyperscaler spreads widened even as GOOGL's stock barely moved. Alphabet had to offer a large concession to complete that August deal, BNY said in a research note cited by Reuters. Meta has been reaching for new pockets of debt too, including a won-denominated bond sale in South Korea to help fund its buildout.
Everyone else is getting a bidding war
Outside the AI complex, the picture flips. Insurance broker Aon's $13.5 billion acquisition financing this month drew $65 billion of orders, and pricing on its 30-year piece tightened by 35 basis points, a measure of how hard investors are competing for bonds that have nothing to do with AI. Loren Moran, a fixed income portfolio manager at Wellington Management, pointed to recent pharmaceutical and insurance financings that needed little or no concession because buyers were hunting, in her words, "ex-hyperscaler."
"There are a lot of investors that just want something other than hyperscaler debt for now," Moran said. "The market is a bit starved for anything ex-hyperscaler."
None of this means the AI borrowers are being shut out. They are being made to pay. Russell Brownback, deputy chief investment officer for global fixed income at BlackRock, said some AI bond deals carried double-A ratings but were pricing more like triple-B credits. He framed the widening as ordinary supply and demand rather than a worry about credit quality, and said the trade still works for both sides: companies borrow at wider spreads because they expect AI to earn back more than it costs, while investors collect yields usually reserved for lower-rated names.
Volume is not the only thing making buyers cautious. Some institutions are bumping up against single-name limits once all the debt tied to one technology company, including money raised through affiliated data-center financing entities, is counted back against the same parent, Moran said. Erickson said many investors also want to keep cash on hand in case the enthusiasm cools, so they can buy hyperscaler debt later if spreads widen further.
Nick Elfner, co-head of research at Breckinridge Capital Advisors, said some hyperscaler deals have drawn weaker demand than investors expect from marquee names, and a number have traded poorly after they priced. The trouble, he said, is surprise. Buyers can plan for a large borrowing program when management gives clear guidance, but a fresh deal only months after the last one, often at a wider spread, chips away at confidence and raises the price of the next sale.
Trading in frequent issuer Oracle's (ORCL) debt has at times shown all of these strains, Marty Fridson, publisher of Income Securities Investor, wrote this month. We saw a sharper version of the same stress when Oracle-linked data-center loans changed hands near 89 cents on the dollar as banks struggled to sell them.
For now, the bond market's message is not that AI borrowers have lost access to funding. It is that investors are becoming selective about the price and the amount of debt they will absorb from any one borrower.
