Bubble or Nothing (publicenterprise.org)

A new policy-focused report warns the current AI-driven data center boom could be a fragile bubble: if tech-sector conditions sour, the investment surge in compute and associated clean-energy projects could evaporate, stranding assets and regional plans. The report treats data centers as hybrid real estate/infrastructure assets and identifies four acute risks for the AI/ML ecosystem: rising inference-service costs amid weak pricing power; rapid GPU depreciation that undermines collateral value; an asset–liability mismatch as tenants cycle through expensive capex inside long leases; and “roundabouting” or circular financing among hyperscalers that concentrates liabilities. Combined with growing, sometimes opaque debt structures and off-balance-sheet vehicles, these trends could weaken developers’ creditworthiness and reduce available, reliable compute. Technically, the report argues current cash flows from AI inference are insufficient to service expanding liabilities, and GPU replacement cadence (now yearly) compresses residual values—raising counterparty and lender risk. Using Minsky-inspired contagion scenarios and T-chart asset/liability mappings, it outlines how a correction could cascade across markets and energy projects. Policy prescriptions: avoid one-shot tax incentives that don’t yield durable benefits, resist tying local budgets to this single industry, and proactively plan to acquire and repurpose distressed energy infrastructure to preserve regional clean-energy and compute capacity.