A net-grid-benefit test for interconnecting AI data centres

Artificial intelligence data centres are expanding rapidly and risk imposing large, inflexible electrical loads on already-stressed power systems. This Perspective argues that grid access for such facilities should be conditional: interconnection should be granted only if developers commit to verifiable, operator-relevant contributions that offset incremental system burdens relative to a passive-load counterfactual. I propose a net-grid-benefit test linking interconnection rights to three measurable obligations: (1) hour-by-hour clean-power matching using auditable accounting and deliverability screens where relevant; (2) certified fast frequency-response and regulation services delivered via controllable IT load, UPS or battery systems, and cooling controls with telemetry and fail-safe guardrails; and (3) recovery and productive use of low-grade waste heat where feasible. Grounded in peer-reviewed evidence and compatible with existing regulatory frameworks, the test converts voluntary sustainability claims into enforceable performance duties, aligning AI-infrastructure growth with grid reliability, emissions integrity, and local energy-integration goals.