AI Infrastructure Energy Disclosure

What energy data should regulators require AI infrastructure providers to disclose?

Measuring the environmental impact of AI requires energy data. But the organisations that hold that data (GPU manufacturers, cloud providers, data centre operators) disclose it at different levels of granularity, in different formats, and with different boundaries. This inconsistency makes it difficult for researchers, regulators, and software teams to measure and compare the energy footprint of AI systems.

The GSF's measurement specifications (SCI, SEE, SCI for AI) all depend on energy data from upstream infrastructure providers. Without consistent disclosure, these frameworks cannot be applied reliably across providers or regions. The question is not whether more transparency is needed, but what specifically should be required and by whom.

This assembly, run jointly with the Policy Working Group, will explore what a minimum energy disclosure framework for AI infrastructure could look like from a regulatory perspective. What data should be reported at the chip level, the instance level, the facility level? What granularity is practical? What existing regulatory frameworks (the EU Energy Efficiency Directive, the Climate Neutral Data Centre Pact, the German Energy Efficiency Act) already address parts of this, and where are the gaps?

The deliverable is a position paper that maps the current disclosure landscape, identifies the gaps that prevent effective measurement, and sets out recommendations for what regulators should consider requiring. It is positioned upstream of the GSF's existing measurement specifications, addressing the data availability problem that those specifications need solved to work in practice.

If you work in AI infrastructure, cloud platforms, energy policy, or regulatory affairs, this assembly will benefit from your perspective.