A reactor wants to run flat. A fleet of machines does not draw flat. Everything interesting about a portable plant happens at the boundary between those two facts, and it is the same impedance problem as any other source meeting any other load.
A 1.2 MW core against a bursty fleet. You have one budget to spend on closing the gap. Which buys you more?
Storage returns energy at 88% round trip, so a joule stored is a joule
partly lost. Compute and thermal absorb surplus at full value but cannot give it back — they
are sinks, not buffers. That asymmetry is the whole design problem.
A 1.2 MWe core, Kaleidos-class, over 24 hours at hourly resolution against a stochastic embodied-AI load. Curtailment is core output with nowhere to go. Unmet is demand the site could not serve. Both are counted in MWh over the day. Round-trip efficiency and the load profiles are representative, not measured; this is a design bench, not a plant model.