The Bank for International Settlements (BIS) documented on October 27, 2026 that five US hyperscalers issued 222,000 million dollars in bonds through September, more than 10% of long-term Treasury issuance. The figure matters beyond that market: US yields are the benchmark for long-term financing in other economies.
The finding is in BIS Bulletin 140, "Building AI on credit: structures and spillovers," authored by Mathias Drehmann and Egemen Eren. The document estimates that capital spending by those five companies will exceed 800,000 million dollars by the end of 2026, close to 2.5% of US gross domestic product, and that a growing share is financed with long-term debt and off-balance-sheet structures. Bloomberg Línea reported on November 1 that the Basel-based body warned about pressures on long-term bond yields. In Mexico and Latin America the channel is the cost of external financing, which is benchmarked to long-term US rates.
The report breaks down the financing. Between 75,000 and 100,000 million dollars were placed in 2026 through off-balance-sheet special purpose vehicles, according to market sources cited by the BIS. Moody's estimated at 1.2 trillion dollars the data center lease commitments of the largest hyperscalers as of mid-2026, of which about 820,000 million correspond to contracts that had not yet begun. The BIS reviewed 18 trading days in 2026 with news about AI financing and found that the term premium on the 10-year Treasury bond rose about 1.5 basis points, a move that reversed within two sessions. Hyperscaler bonds of 10 years or more accounted for about 24% of corporate issuance at that maturity and 13% of gross Treasury issuance, compared with 3% and 1.5% in 2024.
The BIS warned that, if the pace of issuance continues, a concentrated group of investors will have to absorb a larger volume of correlated obligations, with more intense pressure on long-term yields. The next reading is in the Treasury term premium on days with artificial intelligence investment announcements.
This note was drafted with artificial intelligence assistance from verified sources and reviewed by a human editor before publication.