The AI-Native S.M.A.R.T. Generator Master Blueprint
One architecture, eight layers, wired end-to-end into two fusion machines — the breeder (Hyperion) and the burner (Aegis / MetroVolt).
What the Master Blueprint is
The Master Blueprint is the single control document for how Kronos applies artificial intelligence, machine learning, and quantum computing to a fusion power plant. It is not a generic computing stack laid beside a reactor; it is an AI-native architecture in which every layer terminates in real hardware — REBCO magnets, cryogenic feedthroughs, the divertor, the direct-energy-conversion train — and in real plasma physics.
The blueprint spans eight layers, L0 through L7. Reading upward, L0 Foundation provides offline compute, L1 the Control Plane provides microsecond determinism, L2 the Data Fabric carries and curates telemetry, L3 Twin Modeling & AI turns that telemetry into predictions and actions, and L4 through L7 orchestrate, apply, present, and integrate the whole plant with the outside world.
Both machines, one stack
The same eight layers run the breeder (Hyperion), a D-T spherical tokamak with Q_sci 3.076 and 85.0 MW of fusion power, and the burner, a D-3He tandem-mirror generator with a 26.49 T end plug. What changes between machines is not the architecture but the physics the top layers reason about: equilibrium and disruption for the tokamak, end-plug density and the ambipolar potential for the mirror.
How to read the blueprint
- Layers are horizontal bands from L0 (bottom) to L7 (top).
- Colors classify a block by role — governance, infrastructure, operations, AI-workflow, thermal, feedback, framing.
- Line styles classify a connection by latency class — microsecond real-time, standard telemetry, or offline batch.
- Every block names a layer, an interface, a machine, and the physics it touches.
These conventions are documented on the color grammar and line semantics pages. Kronos treats the diagram as an engineering contract, not an illustration.
Honest framing
Both machines are design and simulation studies today. Breeder construction begins Q2 2027 with first-of-a-kind first tritium targeted around 2030. No hardware net-gain claim is made before that milestone; the blueprint documents the intelligence that will run the machines, validated against physics codes rather than against an operating reactor.