The AI datacenter boom is a grid story before it is a climate story.

The framing of AI driven datacenter demand as primarily a carbon problem is understandable and, at least for the next five years, mostly wrong.

In the United States, the United Kingdom, Ireland and the Nordic grids most exposed to hyperscale build out, the binding constraint is not the availability of low carbon electrons. It is the availability of transmission capacity to move those electrons to the sites where operators want to build.

Interconnection queues in the PJM and ERCOT regions now extend well beyond five years. In Ireland, EirGrid has effectively paused new datacenter connections in the Dublin cluster. In the Nordics, siting has shifted north to where the grid has headroom, but the grid headroom itself is finite.

The response so far has been a rush toward behind the meter generation. Gas turbines at the fence line, small modular nuclear power purchase agreements and, in a small number of cases, purpose built solar at scale co located with datacenter campuses. Each of these has a different climate signature, and the aggregate carbon outcome depends on which mix dominates.

The reporting worth doing here is granular. National carbon accounting for datacenter load will not capture the actual outcome. The site by site energy strategy will.