The Bleeding Edge

// energy

AI in Energy & Utilities

AI's energy story is now a grid-access story: interconnection queues, not chips, gate new data centre capacity — and the biggest loads have started buying their own generation.

standing summary updated August 14, 2026

Where AI in energy & utilities actually stands

For this sector, AI is not primarily a software story — it is a load story. The constraint that decides how much AI capacity actually gets built has moved off the chip supply chain and onto the grid: what gates a data centre now is a connection point, an energisation date, and a queue position. The clearest available measure of how concentrated that dependency has become is BloombergNEF's read on Texas, where a pause in grid interconnections initiated by the state's governor puts roughly 20% of the entire US data centre pipeline at risk. One state, one administrative decision, a fifth of the national build-out.

The visible response from the demand side is to stop waiting. Siemens Energy is supplying B&W with 1GW of steam turbine-generators for a US data centre power programme, for an operator that has not been named. Read alongside the queue problem, that looks less like a one-off equipment order and more like the largest loads concluding that self-procured generation is faster and more predictable than the interconnection process. Inference The awkward consequence for utilities and system planners is forecasting: demand that arrives with its own generation attached does not show up in the queue, and the queue is where most load forecasts are anchored. Inference

It is worth being explicit about what is not in the record. There is no interconnection reform, no indication the Texas pause has been lifted, narrowed or challenged, and no stated reason for it in the reporting available to us. No utility has publicly revised a load forecast in response. The Siemens Energy order is unverified and the end operator undisclosed, so nobody outside the deal knows whose gigawatt this is. And nothing this week touches AI applied to grid operations itself — forecasting, dispatch, outage prediction — which is a genuinely different story from AI as a customer of the grid.

This was a thin week: two stories, both about power for compute rather than compute for power. That balance is itself the signal. What would move the picture is a named counterparty on the Siemens/B&W gigawatt, a defined scope or end date for the Texas pause, or the first utility to publicly rebuild its load forecast around customers that bring their own generation.

// Big moves

2026-W33
  • BloombergNEF estimates that a pause in grid interconnections, initiated by the Texas governor earlier this month, puts roughly 20% of the US data centre pipeline at risk. One state's administrative decision is enough to stall a fifth of the national build-out.

    CorroboratedDataCenter Dynamicssource →

  • Siemens Energy will supply B&W with 1GW of steam turbine-generators for a US data centre power programme, DataCenter Dynamics reported. The end operator is undisclosed — a gigawatt of thermal generation procured against a single programme, with no named counterparty.

    UnverifiedDataCenter Dynamicssource →

  • The two stories are one story: when interconnection timelines become political and unpredictable, the largest loads stop queueing and procure generation directly. Neither report states that link — it is our read across them.

    InferenceDataCenter Dynamicssource →

// Daily stories

  1. 2026-08-15

    A data centre spanning up to 50,000 sqm has been approved for construction in west London. Planning documents put its electricity use at 64MW, supporting 48MW of IT load.

    why it matters

    Consents like this are the leading indicator of load growth: 64MW of draw for 48MW of compute. Worth tracking how many clear planning in your service area before they hit the connection queue.

  2. 2026-08-13

    Siemens Energy will supply B&W with 1GW of steam turbine-generators for a data centre power programme, DataCenter Dynamics reported. The units are destined for an undisclosed operator in the US market.

    why it matters

    A gigawatt of thermal generation procured for a single programme suggests large loads are increasingly buying their own power rather than waiting in interconnection queues. Watch what that does to load forecasts.

  3. 2026-08-12

    BloombergNEF estimates that a pause in grid interconnections, initiated by the Texas governor earlier this month, puts roughly 20% of the US data centre pipeline at risk.

    why it matters

    Interconnection queues, not chip supply, are now the binding constraint on AI capacity — and one state's policy can stall a fifth of national build-out.

// Track AI in Energy & Utilities

One email a week on what actually changed — in energy & utilities and across the frontier. Verified, labelled, no hype.