Aegis Compute deploys containerised GPU nodes at curtailed renewable sites and schedules batch AI workloads across them — turning Britain's wasted clean electricity into lower-cost UK GPU capacity.
The Aegis Compute principle
Britain is building renewables faster than the grid can carry them. The result is clean power, generated and thrown away.
paid to switch off UK wind in 2025 — 94% of it in Scotland.
Curtailment · GB grid, 2025
of northern Scotland's wind actually reached the grid.
Stranded power · 2025
a year is the forecast cost of curtailment by 2030.
Trajectory · forecast
of grid queue avoided by siting behind the meter.
Speed · private wire
A datacenter that moves to the energy — not the other way round. Conventional datacenters wait years for a grid connection, then pay the highest industrial power prices in the developed world. We skip both by going to where clean power is already being thrown away.
We partner with wind and solar operators whose output is constrained off the grid, and connect behind-the-meter via private wire.
Power flows straight from the generator to our containers — no grid export, no reinforcement bill, and UK data residency by default.
A refurbished container, fitted with GPUs and efficient cooling, deploys in weeks — skipping the multi-year grid queue entirely.
Batch inference, fine-tuning, rendering and evals run on cheap curtailed power — pausing and checkpointing the instant the operator needs the energy back.
Batch inference
· queue-based
High-volume inference jobs where latency matters less than GPU-hour cost.
Batch fine-tuning
· checkpointable
Fine-tuning and training runs that can pause, resume and migrate across sites.
Rendering & simulation
· parallel
Parallel jobs that scale across available nodes and checkpoint outputs to object storage.
Evaluation runs
· repeatable
Model evals, synthetic data and embeddings, scheduled when power is cheapest.
The Aegis Compute control plane is being built to place each batch job where power, GPU capacity and network conditions are best. When a site ramps down, checkpointed jobs resume on another node — so wasted power becomes one elastic pool of compute, not a row of isolated boxes.
One bridge between stranded clean power and the teams who need affordable compute.
Turn constrained output into revenue with a flexible, fast-ramping local buyer that sits behind your meter and steps aside when you need the power.
For teams running jobs that can queue, pause or checkpoint, Aegis Compute offers UK-sited NVIDIA GPU capacity below mainstream cloud pricing.
The gap between wasted power and scarce compute is widening — and the policy and market conditions to bridge it have just arrived.
UK curtailment cost £1.5bn in 2025 and is forecast to reach £8bn a year by 2030 as renewables outpace the grid.
The UK's AI Growth Zones are steering compute toward constrained power, with grid priority and electricity discounts in Scotland.
Hyperscalers chase gigawatt sites. The smaller, stranded sites — too modest for them — are exactly where a containerised fleet fits.
Aegis Compute is preparing a £1m pilot to deploy the first behind-the-meter GPU container, run paid batch workloads, and prove the fleet scheduler that can route jobs across future renewable sites. The business is fundable once two numbers clear: a £/kWh from a power operator and a £/GPU-hour a customer will pay.
Secure one constrained UK renewable site with private-wire feasibility and a £/MWh that clears the model.
Deploy used NVIDIA-class GPU capacity in a single containerised node, behind the meter.
Schedule, checkpoint and resume paid batch jobs through the first version of the Aegis Compute control plane.
The thinking behind siting AI compute at curtailed renewable power — and what it means for operators and compute buyers.
Get early access
Aegis Compute is at pilot stage — securing our first behind-the-meter site with a UK renewable operator and lining up early compute partners. Tell us which side you're on.
Aegis ComputeClean, low-cost AI compute, sited where Britain's renewable power would otherwise be wasted.
Part of the Aegis family · Aegis Works →