AI Borrowing Is Reshaping the Corporate Bond Market
MarketWatch reported on November 3, 2026 that hyperscalers, the large cloud operators building out AI infrastructure, had burned through their cash and turned to the bond market. The article's headline places AI alongside war and elections as a force moving bonds, a framing that the rest of the reporting on AI borrowing has been moving toward.
The volume is already large. In an October report, MarketWatch put roughly $250 billion of AI-related debt on the high-grade corporate bond market this year. High-grade, or investment-grade, means bonds issued by companies rated strong enough that pension funds and insurers are required or inclined to hold them. Bloomberg reported that global bond sales shattered a record in October, fueled by an AI borrowing binge. Nvidia also returned to the bond market in June after five years of staying out, according to MarketWatch's coverage of its deal.
The forecasts have been moving up. Barclays expects U.S. corporate bond issuance to reach $2.46 trillion in 2026, up 11.8% from $2.2 trillion in 2025, and analysts told Reuters in January that AI hyperscalers would drive the increase. By August 10, BNP Paribas data showed hyperscaler debt issuance at $220 billion for 2026. Looking further out, Goldman Sachs data show gross debt issuance from hyperscalers expected to hit a record $420 billion in 2027, up 60% from 2026 estimates. Gross issuance is the total of new bonds sold, before any older bonds are repaid. A single bank's number can move quickly once a large deal prints.
The disclosed totals may also understate the exposure. S&P Global warned that top hyperscalers likely have more AI-linked debt than is currently known.
For anyone trying to size concentration risk in a corporate credit book, I would weight that caveat as heavily as any single headline figure. Undisclosed debt is the number that tends to surprise people later.
The pattern is not confined to dollar markets. The European Central Bank reported that hyperscalers accounted for 15% of the increase in domestic euro-denominated corporate bond holdings in the year to March 2026, meaning euro-priced corporate bonds held within the euro area.
The part I find most worth watching is what happened to Treasuries. U.S. Treasuries are government bonds, and investors tend to buy them when they are nervous, which is why they are called a haven. Bloomberg reported that U.S. Treasuries showed signs of reclaiming that haven role in October as concerns mounted over the outlook for leading AI firms.
In my view, two scenarios produce very different outcomes for a portfolio. In one, AI spreads widen while Treasuries rally, and the risk stays concentrated in corporate credit. In the other, supply and sentiment worsen together, and duration, a measure of how much a bond's price falls when interest rates rise, no longer offers the cushion it usually does. The October data point suggests the first scenario has been the one on screen so far.
The scale of the underlying spending is what keeps the supply coming. Bloomberg Intelligence expects a tracked group of hyperscalers and neocloud firms, newer companies built mainly to rent out AI computing, to invest almost $4.5 trillion.
Debt is one of the channels financing that capital spending, and it reaches ordinary savers less directly than an equity rally would. Investment-grade corporate bond funds, pension portfolios and insurance general accounts hold these bonds, so the cost of the build-out lands on people who never chose it. Whether that cost shows up as wider spreads, thinner new-issue concessions, or simply more supply to digest will depend on how much of the forecast volume actually arrives.
The honest reading of this story is that the direction is clear and the magnitude is not. Issuance has risen, the forecasts point higher, and the disclosed numbers may be incomplete. Treat the projections as a map of where the pressure is coming from, not as a schedule of what will be sold next quarter.
