The Intelligence Flywheel Across Design, Build, Operate
Each phase produces data that improves the models used in the next, so knowledge compounds from design through build to operation.
What a flywheel is
A flywheel is a system where each turn makes the next easier. Applied to a fusion program, it means the models built for design are refined by construction data and then by operating data, and those refinements flow back to improve future designs. Knowledge accumulates rather than resetting at each phase.
The three phases
- Design: models explore the configuration space and close the design point.
- Build: as-built measurements correct the models to match real hardware.
- Operate: operating data calibrates the models further and trains control and maintenance.
How the loop closes
Operating experience on the Hyperion breeder informs the design of later units and of the burner. A discrepancy found in operation becomes a correction in the design models, so each machine is designed with more knowledge than the last. Across a fleet this becomes fleet learning.
Why it needs the AI-native architecture
The flywheel only turns if data flows freely between phases, which requires the shared model backbone of the AI-native plant. Without it, each phase would use disconnected tools and the learning would leak away.
The timeline
Construction begins Q2 2027 and first tritium is targeted around 2030, so the design-phase flywheel is turning now and the build and operate phases follow. Building the data infrastructure early is what lets the later phases contribute.
Discipline
A flywheel amplifies errors as well as insights, so every correction passes verification before it updates a model.