AI energy demands draw skepticism at 2026 Climate Week

New York Climate Week sessions centered on AI infrastructure expansion. Reports documented increased natural gas plant approvals to supply data centers, offsetting prior renewable gains. Attendees referenced IEA estimates showing data centers consuming 2% of global electricity in 2025, with AI workloads driving 40% of new demand. Congressional and industry statements confirmed no binding efficiency standards emerged from the events.

MIT Technology Review coverage and parallel IEA 2025 data center outlook both recorded net emissions increases from gas peaker plants tied to AI loads. Primary measurements showed average PUE values remaining above 1.3 in new hyperscale builds despite efficiency claims. Cross-referenced filings from utilities in Virginia and Texas listed 15 GW of gas capacity additions explicitly linked to AI projects through 2028.

Operational impact centers on verification gaps. Climate models used by attendees projected AI could optimize grids but required 10x more compute than current deployments deliver. Without enforceable disclosure of training energy per model, sector claims of net-positive climate effects remain untestable against measured consumption curves.

Next steps include proposed DOE reporting rules for large training runs above 10^26 FLOPs, with first filings due Q2 2027. Industry groups have signaled opposition citing competitive risk.