Key points
- Hyperscalers' fast-growing share of bond sales
- Data center bonds carry unfamiliar risks
- What AI borrowers pay versus junk bonds
The biggest AI spenders will become the investment-grade bond market's largest source of concentration risk within about two years, overtaking major U.S. banks, Loren Moran, a fixed income portfolio manager at Wellington Management, said in a Bloomberg Wall Street Week segment posted October 11. Investors are still buying the debt, but Moran said buyers are "demanding more compensation to lend to these issuers."
Hyperscalers' share of bond sales has jumped
Hyperscalers are the largest cloud computing companies, such as Microsoft (MSFT), Amazon (AMZN), and Alphabet (GOOGL). Moran said they will probably sell around $300 billion of new investment-grade bonds in 2026, out of about $2.25 trillion in total U.S. investment-grade issuance she expects this year. Investment-grade issuance is up around 35% from a year earlier, she said, and the growth is "largely coming from the hyperscalers at this point." In September, she told Reuters that the market was "a bit starved for anything ex-hyperscaler," meaning debt from companies outside that group.
A broader measure covers the whole tech sector, not just the hyperscalers. By some estimates, tech companies accounted for just 5% of investment-grade corporate bond issuance in 2024, said David Westin, who hosted the segment. That share is now close to 20%, he said.
A third measure counts bonds already outstanding rather than new sales. Scott Schulte, global co-head of investment-grade debt syndicate at Barclays, said the five biggest hyperscalers now make up about 5% of the investment-grade index, up from 1% to 2% two or three years ago. Banks such as JPMorgan Chase and Bank of America used to sit at the top of the index, and they tend to issue shorter debt, he said. Hyperscalers prefer longer debt and sell bonds that mature in 20, 30, or 40 years, and in some cases 100 years. Schulte said that is lengthening the index's duration, a measure of how much bond prices move when interest rates change.
Data center debt brings risks that are new to bond buyers
Some data center financing uses special purpose vehicles tied to individual projects. Whether that debt appears on a hyperscaler's balance sheet depends on the arrangement. Moran said data center financing carries construction risk, permitting risk, and electrification risk that the investment-grade market has not traditionally taken on.
Moran said the market now separates highly levered balance sheets, such as Oracle (ORCL)'s, from less levered ones, such as Microsoft's. "And so it is not all one trade," she said.
The bond sample behind AIStockWire's AI Bubble Index covers eight AI infrastructure issuers. Seven of them, including CoreWeave (CRWV), are speculative-grade borrowers held by the SPDR Bloomberg High Yield Bond ETF (JNK), and the eighth is Oracle, which is investment grade.
Each bond's spread is its yield minus the Treasury yield at the same maturity, using the same day's prices from State Street's fund holdings files. Each issuer's spread is the median of its bonds, so Oracle's 38 bonds count as one issuer. On October 8, the median of the eight issuer spreads was 334 basis points. The median spread across every bond in JNK, measured the same way, was 231 basis points. A basis point is one-hundredth of a percentage point. CoreWeave had the widest spread in the sample, at 848 basis points.
A SocGen strategist sees governments doing the crowding
Subadra Rajappa, head of Americas research at Societe Generale, said AI borrowing "is quite a lot," and some companies that need to borrow can wait and issue next year. The hyperscalers have a mandate to keep issuing this year and next, she said, and are "time agnostic."
Rajappa said some high-yield names have widened significantly in the last couple of weeks, but the risks are "really hard to see" in the data so far. "But as of now I think that the supply is being very well absorbed," she said.
She pointed instead to rising government debt and deficits in developed markets, from Japan and the U.K. to Germany, which is issuing more debt for defense and infrastructure. Spreads on French government debt have started to widen, she said. In her view, "government bonds are starting to crowd out" the broader bond market.
Rajappa said her broader concern is whether the largest companies earn back their AI spending over the next several years. She also said she is concerned about pushback against data center development heading into the midterm elections, which are on November 3.
I think the distinction that matters is between heavy borrowing and evidence that other borrowers are being pushed out. Hyperscaler issuance is growing rapidly, but Rajappa sees government borrowing as the stronger crowding-out force so far.
