A commentary published by heise online's iX magazine on Wednesday takes aim not at AI's capabilities but at the financial engineering behind it. Author Axel Kannenberg frames Anthropic's planned listing — reportedly targeting a valuation around two trillion US dollars while the company posts heavy losses and plans on the order of half a trillion dollars in infrastructure spending — as the template for a broader mania.
The core argument is about how the build-out is paid for. Kannenberg writes that capital expenditure for AI infrastructure can no longer be funded from the hyperscalers' own cash flow, so enormous amounts are being raised on financial markets as debt. He speculates that persistently high US Treasury yields are partly a consequence of the AI economy drawing capital out of bond markets. Demand is unproven: he cites an estimate by the consultancy Bain that the market would have to be about six trillion dollars for the current investment level to ever pay off, and an Allianz Research estimate that around half of the 12 GW of US data-centre capacity planned for 2026 is already delayed or possibly cancelled — due to protests, lawsuits or difficulties securing power supply. Even a stalled data centre still services its interest, he notes.
The sharpest point concerns where the debt sits. When companies such as Oracle and Meta move AI obligations into Special Purpose Vehicles, he argues, they appear only as a tenant or minority owner and the liability stays off their own balance sheet. He acknowledges the warnings have already been issued — by the Bank for International Settlements, the IMF and the ECB — and concludes that the script for a sequel to The Big Short is essentially written. The piece is opinion, not reporting; its numbers are attributed to the sources named above rather than independently verified here.
