Helion's DC Power for AI Data Centers
Helion Energy
This points to Helion trying to sell a better electrical match for AI infrastructure, not just a new source of generation. Its reactor design is built to create electricity directly from changing magnetic fields, without the steam turbines used by most power plants. That matters because servers and batteries ultimately run on DC internally, so a co-located fusion unit could cut some of the conversion steps, equipment, and energy loss that sit between the grid and a rack of GPUs.
-
Helion’s machine is designed more like a pulsed electrical device than a conventional thermal plant. Plasma expansion pushes against magnetic coils, current flows back into capacitor banks, and power can be sent out without first making heat, steam, and turbine rotation. That is the core reason DC delivery is plausible in a way it is not for tokamak systems like CFS.
-
In a modern AI data center, power is repeatedly converted before it reaches chips. Grid power arrives as AC, is stepped through transformers and power distribution gear, then converted again for server power shelves and battery systems. Vertiv and NVIDIA have both framed high voltage DC architectures as a way to remove conversion stages and improve efficiency for dense AI loads.
-
The strategic fit is with behind the meter AI campuses that increasingly look like private power systems. Helion already has a 50 MW power agreement with Microsoft aimed at a first plant by 2028, while Crusoe is building a $12B Abilene campus for OpenAI and Oracle that is expected to bring 700 plus MW live by December 2026. As campuses get larger, even small percentage savings in power delivery become meaningful.
The direction is toward data centers buying generation that plugs more directly into their electrical stack. If Helion can prove commercial operation, its advantage will not just be clean baseload power. It will be a simpler path from reactor to GPU, which could make fusion especially attractive for the next generation of giant AI campuses.