Artificial Intelligence (AI): The credit boom worries economists

Coin Encyclopedia Improve your knowledge of cryptos

Alphabet, Amazon, Meta, Microsoft and Oracle together forecast more than $800 billion in investment spending in 2026, according to high forecasts from the companies included in the bulletin. This amount represents approximately 2.5% of US GDP.

To support this expansion, these five groups issued $222 billion in bonds as of September 22, mainly with long maturities. Added to this, according to the market sources cited, is between $75 and $100 billion in bonds and asset-backed securities linked to off-balance sheet vehicles.

The Hyperion data center illustrates these arrangements. The joint venture between Meta and funds managed by Blue Owl notably financed the project with $27.3 billion in private bonds. Meta does not consolidate this structure, but supports its financing through rental commitments and limited protection of the residual value of the assets.

However, off-balance sheet does not necessarily mean secret. The authors specify that rentals not yet started are the subject of published information, without being recorded as debt. The problem is mainly due to the difficulty of measuring total exposure.

These structures are in fact based on the solidity of a small number of companies. An investor can thus simultaneously hold the bonds of a technological giant and those of several vehicles dependent on its rents, without fully measuring this concentration.

Banks also remain exposed: they sometimes finance the construction of data centers and lend to funds that buy their debt. At the same time, chips are fragile collateral, as their value can decrease with new generations or simultaneous resales.

However, the bulletin does not note a general destabilization of the bond market. Over 18 days of announcements related to AI financing in 2026, the ten-year US term premium increases by around 1.5 basis points. This effect disappears in two sessions and the yield spreads of highly rated corporate bonds do not change significantly.

The authors emphasize that these are statistical associations, without proof of causality. A continuation of issues at the same pace could nevertheless exert more lasting pressure on long-term rates.

Machine-translated from French