The Bond Market Is Financing the AI Boom. Does It Know the Full Risk?

The AI infrastructure boom has a financing layer most people aren't watching — and the people who are watching it are starting to worry. Morgan Stanley estimates AI-related global debt issuance totaled $236 billion through May 2026, four times the pace of a year prior, and projects $570 billion by year end (Motley Fool, 2026). Amazon, Alphabet, Meta, Microsoft, and Oracle issued $159 billion in corporate bonds in the first five months of 2026 — more than those same companies borrowed combined over the entire previous five years (Congress.net, 2026). The bond market has been treating AI infrastructure debt as nearly risk-free. Signs are emerging that it is not.

The scale of the shift is structural. Hyperscaler bond issuance was roughly $28 billion per year in 2024 — their historical average. In 2025 it jumped to $121 billion. In 2026 it is tracking toward $300 billion. Alphabet became the first technology company in decades to sell a 100-year bond. Meta drew a $125 billion order book for a single $30 billion issuance. These companies are borrowing at scale because the data center buildout requires capital at a pace that even their enormous cash flows cannot absorb (IESE, 2026). The IMF has flagged the core risk: a "potential maturity mismatch" between the duration of physical AI assets and the debt used to finance them, if data centers and chips lose value before loans are repaid (Congress.net, 2026).

The regulatory picture compounds this. The Sanders-AOC AI Data Center Moratorium Act, introduced March 2026, would impose a federal freeze on new construction above 20 megawatts until AI safety legislation passes (Congress.gov, 2026). Separately, nine states have proposed statewide moratoriums, over 100 localities have acted to block or delay projects, and $156 billion in data center projects were disrupted in 2025 alone by community opposition (S&P Global, 2026).

The bond market is now beginning to register this. Morningstar reported signs of "buy-side indigestion" this week — investor appetite straining under the volume of new issuance (Morningstar, 2026). Forbes noted that data center bonds are increasingly being priced with more protections for investors — amortisation schedules, tenant commitments, tighter structures — that plain corporate bonds rarely require (Forbes, 2026). In other words: investors are starting to price in risks they previously ignored.

This is not a prediction of defaults. These are among the most cash-generative companies ever to exist. It is an observation about a correction in how this debt is priced — bonds that were originally issued as though the AI buildout were a certainty are now being scrutinised as though it is a bet. The market is catching up to what the regulatory disruptions, the maturity mismatch, and the demand uncertainty were always suggesting.

These bonds were priced during peak AI optimism, when investor demand for anything AI-adjacent was nearly unlimited. The scrutiny arriving now reflects a more sober question: what if the revenue these data centers are built to generate doesn't arrive on the timeline the debt assumes?