Lines of Thought

Power is the bottleneck

Who is paying for, and who gets to connect, the electricity that gigawatt-scale AI data centres need?

Labs have signed multi-gigawatt chip and data-centre deals that outstrip existing grid capacity. Hyperscalers are now funding reactors, regulators are rewriting interconnection rules, and the EU is using public money to anchor its own capacity.

  1. DealComputeUS
    OpenAI, SoftBank, Oracle and MGX launch $500bn Stargate data-centre venture

    Sets the gigawatt scale for frontier labs.

  2. DealComputeUS
    OpenAI signs 6GW AMD GPU deal and gets warrant for up to 160m AMD shares

    Chip deals are now measured in gigawatts, not units.

  3. Rule · In forceUS
    AI Action Plan

    Federal policy makes data-centre and grid build-out a priority.

    Also on Governments bet on AI for science

  4. DealComputeUS
    Meta signs nuclear deals with Vistra, TerraPower and Oklo for up to 6.6GW by 2035

    A hyperscaler underwrites new reactors directly.

  5. DealComputeUS
    Anthropic secures multiple gigawatts of Google TPU capacity with Broadcom from 2027

    Another multi-gigawatt commitment from 2027, mostly in the US.

  6. PolicyComputeUS
    FERC orders six US grid operators to fix or defend rules for connecting data centres

    The grid regulator forces each region to set rules for connecting large loads.

  7. LaunchComputeUS
    Emerald AI, Google and NVIDIA form alliance for grid-flexible AI data centres

    Industry offers flexibility in exchange for faster connection.

  8. ProgrammeComputeEU
    EU opens call for up to seven AI gigafactories in at least seven member states

    Europe's publicly anchored alternative, with bids due 12 Nov 2026.

Where this line could go next

Connected developments not on this line