Skip to content
Advertisement
Advanced-computing debt issuance tops $680 billion, five times its 2023 level

Advanced-computing debt issuance tops $680 billion, five times its 2023 level

U.S. Government / Wikimedia Commons (Public domain)

Hyperscalers are borrowing at a historic clip to fund data centers and AI chips, even as bond yields climb toward 7%

Advanced-computing-related debt issuance climbed past $680 billion by October 2026. That is five times the level recorded in 2023.

Where the money is going

The surge is driven largely by the cloud hyperscalers. These are the giant tech companies that rent computing power to everyone else.

They are spending heavily on data centers, AI chips, and advanced computing capacity. A growing share of that spending is now financed through debt instead of operating cash flow.

Research estimates put AI-specific debt issuance for 2026 somewhere between $466 billion and $500 billion. When related sectors are counted, the broader total lands at approximately $680 billion to over $850 billion.

The width of that range matters. How you define “AI-related” changes the answer by hundreds of billions of dollars.

Either way, the trend points in one direction. Issuance is running roughly four to five times higher than it was in 2023.

Advertisement

The biggest borrowers

A handful of names account for a large slice of the activity. Amazon has issued over $54 billion in debt.

Alphabet follows at around $31.5 billion, with Meta close behind at around $30 billion.

Combined bond issuance from the major tech firms is estimated at between $200 billion and $370 billion for 2026. The money is moving through several channels at once, including public bonds and private credit deals.

Some of it is also flowing through off-balance-sheet structures. These are arrangements that let a company fund a data center without the full debt load showing up front and center on its own books.

Borrowing costs are rising too

Average yields on top data center bonds approached 7% by October 2026. Earlier in the year, they sat below 5%.

A jump of that size changes the economics of a project. A data center financed at 7% has to earn meaningfully more to justify itself than one financed at under 5%.

How we got here

In 2023, AI infrastructure spending was already ramping up, but debt issuance tied to advanced computing sat at roughly one-fifth of today’s level.

The shift is also expected to continue. About 35% of hyperscaler capital expenditures might be debt-funded in 2027, according to the research projections.

What this means

For fixed-income investors, the immediate story is supply. A steady flood of large tech bond deals gives buyers access to high-quality issuers, but it also means more competition for capital among borrowers.

The off-balance-sheet piece deserves close attention. Structures that keep debt out of plain view can make a balance sheet look healthier than the underlying commitments suggest, which complicates credit analysis.

Three signals will show whether the boom is stretching too far. The first is whether data center bond yields keep climbing past the 7% area. The second is whether new deals start needing bigger concessions to attract buyers.

The third is whether the 35% debt-funding projection for 2027 materializes or gets revised. If tech giants lean even harder on borrowing next year, the $680 billion figure may turn out to be a waypoint rather than a peak.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.