AI data centres are a grid problem before they are a software problem
Why utilities, planners and AI operators need a shared playbook for the new load, and what the consultancies are missing.
By the BuiltWorld AI research team
The new load
Public estimates from McKinsey, BCG and IEA all point to AI data-centre demand roughly doubling by 2030 in major markets. The headline figure hides three operational shocks: interconnection queues clogging in lead grids, mismatched siting between AI clusters and renewable build-out, and demand-response programmes that were never designed for this load shape.
Where consultancy framing falls short
Most consultancy work frames the problem as capacity expansion. From an operator standpoint, the harder issues are upstream: how to triage interconnection requests, how to encourage co-location with firm and renewable resources, and how to design tariffs and curtailment products that AI operators will actually accept.
These are not problems a vendor can sell into. They are policy, planning and information-sharing problems that need a vendor-neutral convener.
What we are doing
BuiltWorld AI is mapping queue and tariff treatments across leading utilities and large-load AI buyers. The goal is a shared, anonymised baseline that planners can cite in rate cases and that AI buyers can use in siting decisions. Output is targeted for late 2026.
© BuiltWorld AI · Independent foundation. Cite as: BuiltWorld AI Briefing, “AI data centres are a grid problem before they are a software problem”.
All briefings
