Fusion for AI Compute
AI training clusters are becoming gigawatt-scale, constant, DC loads — precisely the firm, DC-native power a MetroVolt plant is designed to deliver.
- Load
- Large, constant, growing to GW scale
- Need
- Firm power; DC distribution inside racks
- MetroVolt
- 486 MWe (Mode D), DC-native
- Fit
- Co-located firm clean supply
Training and running frontier AI models is extraordinarily power-hungry, and the largest compute clusters are heading toward gigawatt-scale demand — large, constant, around-the-clock, and increasingly bottlenecked by electricity supply rather than chips. That's a firm-power problem of exactly the kind fusion targets.
A MetroVolt unit delivers 486 MWe of firm output at the Mode-D operating point (a three-unit station ≈1.5 GWe; 0.61–0.84 GWe per unit at the gated hot-ion ceiling), matching a hyperscale campus. Two features fit AI compute unusually well: it's firm, so it can anchor base load without weather risk; and it's DC-native from direct energy conversion, aligning with the DC distribution inside modern data centers and cutting conversion stages.
Co-locating firm, clean, DC-native generation with AI compute is one of the clearest near-term value propositions for firm fusion — developed through strategic partnerships. See firm power for data centers.