The frenzy to finance AI’s data centers and GPUs is jamming bond markets. As issuance surges, capacity limits designed to ensure diversification and reduce risks could turn the boom into a credit contagion.
The AI boom has lifted markets to one record high after another. Investors are piling in. Companies are building data centers at a feverish pace. Bankers, jaws agape, are drooling over the fees only a once in a generation debt binge can provide.
But the buildout comes with risks that ordinary people rarely think about. At the WSJ Tech Live conference in Laguna Beach this October, OpenAI finance chief Sarah Friar warned that the government might need to backstop the debt fueling the expansion. Such a step would protect corporations and investors while strong-arming the public into absorbing at least some of the risk.
Was OpenAI’s CFO asking for a future bailout because she was concerned about the hundreds of billions in debt coming to market in a race to fund the AI infrastructure buildout? Within a day Friar tried to walk back her remark, writing on LinkedIn that she meant “partnership” and not a government imprimatur to make the debt easier to swallow. But the slip revealed what many in finance already know: the bond market may have the sheer capacity to fund AI, but it may not have the risk tolerance for such a large narrow bet. Limits on how much funds can hold from any sector or single issuer, along with the risk of holding AI-related trades across dozens of companies, cap how far this can go.
The scale is immense. Analysts at JPMorgan estimate that AI-linked investment grade bond issuance alone could reach $1.5 trillion by 2030. That compares with the average of $1.9 trillion of total U.S. corporate bonds issued each year since 2020. The overall borrowing tied to AI infrastructure is expected to be far higher. U.S. companies have already issued more than $200 billion in AI-related bonds this year, roughly 10% of the corporate bond market. Amazon was the latest to enter the fray, announcing a $15 billion sale on Nov. 17. Alphabet issued $25 billion earlier in November, with maturities stretching as long as 50 years. Meta raised $30 billion in October while Oracle sold $18 billion the month before that.
These tech giants don’t necessarily need the loans. Most of these companies have tens of billions in cash reserves, pristine balance sheets and investment grade credit ratings. Meta, for example, is sitting on $44.5 billion in cash and short term securities, according to its most recent SEC filings, while Alphabet and Amazon have nearly $100 billion each in their coffers.
That explains why Friar’s comment struck a nerve.
If investor appetite for AI-linked bonds wanes, these “AI growth” issuers could decide to pay more or sweeten their covenants beyond the market norm. This will likely raise borrowing costs for everyone. The blue chippers are already upping the ante. Analysts at Janus Henderson Investors, a London-based asset manager, wrote that Alphabet and Meta both paid a “clear premium to access the debt market” of 10 to 15 basis points compared to their prior issues. After Meta’s bonds came to market in November, demand for Oracle’s fizzled, with its 2055 maturity bonds seeing their spread to 30-year Treasurys expand by 11 basis points in a single week. Funds may rotate out of old credits to make room for new ones, pushing up yields and tightening liquidity across the market. What looks like an AI boom could spill into a wider credit squeeze.
Gil Luria, head of technology research at D.A. Davidson says tens of billions of dollars a year in AI loans are manageable, but that hundreds of billions would start to crowd out other borrowers. “The debt market as a whole will become increasingly concentrated in this very, very specific investment,” he says.
Investors in corporate bonds assume diversification will protect them. But when a wide swath of issuers depend on the same business model, diversification breaks down. Directors at S&P Global Ratings, including David Tsui and Naveen Sarma, say that technology, media and telecom issuers could all move together if demand for AI computing capacity slows. A shock in one area could ripple through the rest. What looks like separate risk can merge into one large exposure.
Credit markets have seen waves like this before.
Todd Czachor, global head of fixed income research at Columbia Threadneedle Investments, points to the shale boom, which drew roughly $600 billion of investment and reshaped credit markets as capital surged into a single theme as a parallel. He says the AI cycle shares the pattern if not the scale. His team estimates total AI infrastructure spending could reach $5.7 trillion, describing the size of the buildout as “on a different planet,” bigger than anything he’s seen before.
Czachor expects investors to sort winners from weaker bets, as they eventually did in shale, but says the risk today comes from the volume of issuance hitting the market at once. He notes that even well-capitalized companies will test portfolio limits if they fund hundreds of billions of new projects through public debt. That strain can push spreads wider across technology, media, and telecom issuers and spill into other sectors as supply grows.
That’s because institutional bond investors often face limits on how much they can own within a sector or from a single issuer. Pension funds, insurers, and mutual funds set caps to keep portfolios balanced. Those limits become constraints when a single sector dominates new issuance. The same logic that keeps bond buyers safe can also shut out new borrowers.
Concentration shows up first in the indices that anchor the bond market.
Many investment grade benchmarks cap how large any single issuer can be. MSCI’s USD Investment Grade Corporate Bond Index sets a 3% ceiling. Fidelity’s All Maturity U.S. Investment Grade Corporate Bond Index limits each issuer to 3.5%. MarketAxess’s U.S. Investment Grade 400 Index uses a 4% cap.
Fund mandates follow the same pattern.
BlackRock’s largest corporate bond ETFs inherit the caps from their underlying indices. iShares iBoxx Investment Grade Corporate Bond ETF, ($33B in net assets) uses the Markit iBoxx investment grade index with a 3% issuer limit. Its iShares iBoxx High Yield Corporate Bond ETF ($18B) follows the high-yield version, which also uses a 3% cap.
Many pensions and university pools take a broader approach but reach the same result. For example, the Chicago Policemen’s Annuity and Benefit Fund ($4.6B in assets) limits any one industry to 25% in its fixed income strategies. These rules work as speed limits when a surge of paper from one issuer or sector tries to enter the market all at once. Once the limits are hit, investors must either sell older holdings to make room or pass on new issues.
For now, the limits haven’t been breached. In iShare’s investment grade bond ETF, which has a 3% single issuer limit, Oracle sits at about 2% of the fund and Meta is the only other major AI borrower with a weight above 1%.
Private credit was supposed to fund part of the AI buildout, but Brij Khurana, a fixed-income portfolio manager at Wellington Management, says that view is fading. Early defaults in unrelated sectors have made lenders more cautious, and diversification rules limit how much any one fund can put into data center exposure. He says his firm’s private credit analyst thinks “maybe $200 to $300 billion” of total AI spending can be absorbed privately, far short of what is needed. That pushes the burden back onto public bond markets, where spreads are already moving.
Khurana also says the structure of most portfolios hides how correlated the risk has become. Concentration guidelines typically target individual issuers not themes. A fund can own separate 3% holdings of Oracle, Alphabet, Meta, and Microsoft, diversified by issuer, but highly concentrated in AI risk. The rules treat each name as different, but the underlying exposure is the same: one massive wager on AI computing and infrastructure.
Khurana believes big tech’s AI borrowing binge is already taking its toll. “These are still high quality companies and they’ve been issuing at spreads you haven’t seen in years, and what has happened is those spreads have stayed wide and the rest of the market has repriced higher,” he says. He warns that a high quality issuer willing to pay even more would reset valuations across the entire corporate bond market. In that scenario, investors would dump lower quality credits first. “Why would you own anything else in the credit markets” if pristine AI borrowers are paying tantalizing yields, he says.
Who are the “lower” quality borrowers?
How about BBB S&P rated AT&T which currently has $150 billion outstanding in bonds, Comcast, rated A-, which has $100 billion and Verizon, rated BBB+, with $120 billion? These three credits alone are among the biggest and most active bond issuers in the market. These technology, media and telecommunications sector stalwarts will need to make room for more big beautiful bonds from the likes of Alphabet, Meta and Amazon. This is sure to be costly for many corporate borrowers, but the systemic risks for bond investors could be even greater.