Unshifted compute loadData centres can't shift work to cheaper, cleaner hours
Training runs, batch inference and analytics are scheduled around deadlines and cluster availability, almost never around what electricity costs or emits at that hour. Much of this work is genuinely movable, a few hours either way or a different region entirely, but the information needed to move it lives in a different system from the scheduler, and in most organisations nothing joins them. So flexible load behaves like inflexible load.
Platform and infrastructure teams · sustainability leads with compute in their inventory · grid operators courting flexible demand
Why now
The IEA's April 2025 Energy and AI report puts data centre electricity use at around 415 TWh in 2024, about 1.5% of global consumption, and projects roughly a doubling by 2030 with accelerated servers accounting for almost half the increase: a load growing fast and, unusually for industrial demand, substantially time- and place-shiftable.
Why it matters
A large and rapidly growing load is being treated as though it had no flexibility, when a meaningful share of it does. Making compute schedulable against grid signals is a software problem sitting between two industries that do not yet share a vocabulary.
From public reporting
Framed from public reporting. Here is what it was drawn from, so the framing can be checked against it:
The first mass-market electric cars are reaching the end of their first ownership with batteries that have years of useful service left. Battery-health reports exist, but no grade is yet trusted by buyers, insurers and second-life users alike, so buyers discount heavily to cover what they cannot see, insurers guess, and packs that would work well in a home or on a grid for another decade are recycled early or not moved at all.
For whomUsed-car buyers and dealers · fleet operators disposing of vehicles · home and grid storage developers
The constraint on the energy transition moved from manufacturing to administration, and no tooling followed it. Projects sit in interconnection queues longer than they took to design, studies are redone by hand each time the queue ahead reshuffles, and developers cannot see where they stand or what would move them. Most of the capacity that enters the queue is eventually withdrawn, often after years of waiting.
For whomIndependent power producers · utility interconnection teams · the financiers whose models assume a connection date
Loss happens at the handovers: field to packer, packer to distributor, distributor to shelf. Every party measures only its own side of each one. So each set of numbers looks defensible while the total is enormous, and the interventions that would work are precisely the ones nobody can see, because they sit between two businesses rather than inside one.
For whomGrowers and packers · distributors and independent retailers · redistribution charities working the same corridors
