AI spending is becoming a chain of promises

The model brings together the parties that build models, supply chips, rent compute, construct data centers, provide power, and finance the assets. Long contracts and guarantees can make an enormous project bankable even when the central AI customer is still investing ahead of proven demand.

That architecture can accelerate capacity and distribute risk. It can also concentrate exposure in less visible ways when the same companies invest in one another, guarantee one another’s obligations, and depend on one another’s projected growth.

Stress the revenue, not only the hardware

The core risk is not that every AI infrastructure project fails. It is that financing assumptions synchronize around the same expectations for model usage, pricing, and enterprise adoption. If those expectations weaken, chip leases, power contracts, data-center debt, and cloud backlogs can all reprice together.

Disclosure should show the full chain: who guarantees which payment, how much demand is committed versus forecast, what happens if construction is delayed, and where losses land if a model company cannot meet its obligations.

Primary trail

Go to the source

Read the evidence behind this analysis. External links open in a new tab.

Financial Times — Inside Google’s $200bn Wall Street finance machine for Anthropic