Bank of England warns of AI debt

The Bank of England's Financial Policy Committee (FPC) highlights that a substantial volume of AI-related investment is being financed through debt. This increasing indebtedness, coupled with opaque and sometimes 'circular arrangements' in financing, complicates risk assessment and could amplify losses if market expectations for AI's contribution to growth prove overly optimistic.

A key concern for the FPC is the potential impact on sovereign debt markets. Growth prospects and fiscal outlooks are partly tied to the expectation of significant productivity gains from AI. A reassessment of these expectations could therefore affect not only AI asset valuations but also government debt, implying a broader economic risk beyond just the tech sector.

The emergence of advanced open-weight AI models, narrowing the capability gap with closed-weight models, presents a complex dynamic. While this could democratize access to advanced AI and potentially reduce costs for technology buyers, it also introduces a competitive pressure for frontier AI firms. If cheaper, less hardware-intensive open models gain traction, it could challenge the premium associated with proprietary offerings from companies like Anthropic and OpenAI, affecting their projected returns.

Analyst firm Gartner projects that aggregate AI capital investment may not achieve a 12% return on invested capital (ROIC) – the investment hurdle rate – until 2032 at the earliest. This extended timeline suggests that IT buyers and CIOs face a real risk of some AI product providers not being viable in the long term, and that inference and machine learning costs could rise as publicly listed AI companies seek to fund further expensive hardware investments and increase shareholder value.