AI capital spending moves into foreign bond markets as model prices collapse

The largest cloud computing companies have grown too large for the American corporate bond market to absorb on its own, and are now borrowing wherever they can find capacity. The Financial Times reported that the volume of issuance from Amazon, Alphabet and their peers has pushed up borrowing costs in Canadian dollars, Swiss francs and sterling, markets that are small relative to the sums involved. Amazon raised C$14 billion in Canada in June, the largest corporate bond sale that market has seen, and Alphabet has set issuance records in yen, Canadian dollars, Swiss francs and sterling within the same year, according to IFR.

The revenue side of the same industry is moving the other way. The Financial Times also reported that OpenAI and Anthropic have released cheaper models as Chinese rivals take share. Chinese models whose underlying code is published rather than kept closed (open-weight models) run at a fraction of the price of the leading American systems, and US companies have routed a growing share of their token volume to them when a task does not require the strongest model available, as CNBC has documented.

Together, these two trends form a pattern familiar from past credit-financed capacity booms. Companies are financing fixed costs with long-dated debt denominated in currencies whose local investors have limited alternatives, while the price of the output those assets produce falls quickly under competitive pressure. The assets are real and the demand is real. The open question is the gap between what the capacity cost to build and what its output can be sold for.

For bond investors outside the United States, the arithmetic is more immediate. A pension fund in Toronto or Zurich that buys a domestic index now owns exposure to a single technology thesis it did not choose.