The biggest names in tech have been on a borrowing binge that makes their old spending habits look quaint. Amazon, Alphabet, Meta, Microsoft, and Oracle, the so-called hyperscalers, have collectively issued approximately $121 billion in new corporate bonds in 2025 to fund AI infrastructure. The previous annual average was roughly $28 billion.

In other words, these five companies borrowed in one year what they used to borrow in more than four.

The debt wall nobody saw coming

Over $90 billion of that $121 billion total hit the market in just the final three months of 2025. Over the past five years, the collective debt load across these five companies has roughly doubled, reaching an estimated $350 billion.

AI-related debt now represents around 30% of net new issuance in the US investment-grade bond market. When nearly a third of new high-quality corporate bonds are effectively bets on the same technological thesis, the word “diversified” starts losing its meaning.

What $121 billion buys you

The money is flowing into data centers, servers, and GPU clusters, the physical backbone of the generative AI race.

Investor sentiment is shifting

By mid-2026, investor concerns around longer-dated AI debt have intensified noticeably. Some mega-cap names are already facing sentiment pressure specifically tied to how they’re financing their capital expenditure programs.

A Chicago Fed analysis published in 2026 highlighted elevated bank commitments and tail-risk exposure to AI-adjacent industries. Delinquency rates remained low as of late 2025.

The real risk isn’t that any one of these companies fails. It’s that the entire investment-grade market has become a leveraged bet on AI infrastructure paying off within a specific timeframe. If you’re holding a diversified bond portfolio in 2025, there’s a decent chance you’re more exposed to data center economics than you realize.