Insights

    GPU & Compute

    Hyperscalers are buying gigawatts directly. That tells you what they expect the grid to do.

    Devence Lab

    · 2 min read

    Share
    Hyperscalers are buying gigawatts directly. That tells you what they expect the grid to do.
    Photograph · Unsplash

    Microsoft contracting 10.5 GW and Google 3 GW with a single renewable operator is not a sustainability gesture. It is a hedge against the public grid being unable to supply.

    Microsoft's agreement with Brookfield Renewable Partners for 10.5 GW, and Google's 3 GW deal with the same operator, are usually filed under corporate sustainability. Read them as capacity procurement instead and they say something sharper.

    Vertical integration is a signal of scarcity

    Companies buy upstream when they stop trusting the market to supply. These are firms with the best grid relationships and the strongest negotiating position in the industry, and they are contracting directly with asset owners at gigawatt scale rather than relying on utility supply.

    That is not a bet that power will be cheap. It is a bet that power will be unavailable at any price in the places and timeframes they need it.

    When the largest buyers stop queueing and start buying the factory, the queue is the problem.

    What it does to everyone else

    Two consequences for organisations that are not hyperscalers. Contracted capacity is removed from the pool available to everyone else in those regions, which tightens the market that mid-sized operators and colocation customers are drawing from. And it establishes a procurement norm — multi-gigawatt, multi-year, direct — that most organisations have neither the balance sheet nor the horizon to match.

    The practical consequence is that access to accelerated compute increasingly depends on a relationship with someone who has already secured power. That is a different competitive dynamic from buying instances, and it favours incumbency in a way the cloud era largely did not.

    The planning implication

    If your roadmap assumes accelerated capacity will be available in a specific region on a specific date, that assumption now depends on a power market you are not participating in. Worth testing with your provider: not whether they have capacity today, but whether their expansion in your region is contracted or hoped for.

    The answer is rarely in a datasheet, and it increasingly determines whether a multi-year deployment plan is realistic.

    Sources

    1. AI Data Center Power: Grid Limits Reshape Energy in 2026Enki.AI
    2. 2026 Predictions: AI Sparks Data Center Power RevolutionData Center Knowledge
    3. Power-Bound, Not GPU-Bound: AI Data Center Power Constraints Are the Real 2026 BottleneckSpheron

    Written by the Devence Lab research team.

    Share

    Collaborate

    We share findings with partners operating in the same constraint space.

    Get in touch