Power orchestration for AI data centres and the grid.
The most consequential technology of our time is AI. The growth of AI is limited by one factor: access to power. Grid infrastructure is not able to keep up with data centre energy demand.
We set out to fix this bottleneck and created the fastest ever energy orchestration reinforcement learning agent that gets more power to more data centres, unlocking additional compute capacity and millions of dollars in additional revenue — instantly.
Stable Motion is a software layer that manages the power draw of GPU clusters as they run. A reinforcement learning controller sets power limits across the fleet, holding the site inside its grid limit while protecting the performance of the workloads on it.
More compute from the connection a site already has, and a data centre that can respond to the grid in under a second.
more GPUs on the same grid connection
average workload slowdown
response to a grid signal
less slowdown than uniform power capping
From our testing to date.
Reads live power, thermal and workload telemetry from every GPU in the cluster.
A learned policy sets each GPU's power limit, balancing the site's grid limit against the performance of the jobs running on it.
Re-optimises within half a second as conditions change, so a site can flex with the grid without stalling its workloads.
Serial founder. Leads the company.
Architect of the reinforcement learning controller.
Oxford, energy systems and power engineering.
Pilots with GPU cloud and data centre operators, and with network operators, are under way. We are speaking with a small number of partners and investors ahead of launch.