The L0 Foundation Layer
L0 is the offline, high-compute substrate that trains, simulates, and re-grounds everything the real-time layers do for both Kronos machines.
Where L0 sits in the S.M.A.R.T. stack
The Kronos AI-native architecture is an eight-layer stack. L0 Foundation is its base: the non-real-time compute substrate where offline multi-physics simulation, batch model retraining, and large-scale reproducible study campaigns run. It is deliberately separated from L1, the microsecond control plane. L0 is allowed to be slow, expansive, and thorough; L1 is not allowed to be anything but fast and deterministic.
Every predictive object that later runs in the real-time digital twin is born here. The Grad-Shafranov surrogates, the physics-informed neural networks, the neutronics response tables, and the anomaly-detection ensembles are all trained on L0 and only then compiled down toward the edge. L0 is the ground truth generator; the fast layers are its accelerated shadows.
What runs on L0
- Offline multi-physics Monte Carlo for neutronics and activation, both machines
- Batch retraining and refinement of every twin surrogate
- GPU-cluster training runs for PINNs, GNNs, and anomaly ensembles
- Large parameter sweeps: breeder triangularity and TBR levers, burner plug-density scans
- Petabyte-scale archival and replay of pulse histories
Both machines, always
L0 serves the breeder (Hyperion), a D-T spherical tokamak with R0 1.2 m, aspect ratio 2.5, peak field 16.84 T (8 T on-axis), Q_sci 3.076, 85.0 MW fusion power, 9.66 MA plasma current, and negative triangularity delta -0.30. It equally serves the burner, a D-3He tandem-mirror generator with a 26.49 T plug, 17 T throat, and 5.44 percent neutron fraction. The physics differs sharply, but both share this compute base.
Because construction of the breeder begins Q2 2027 and first tritium is expected around 2030, everything on L0 today is design and simulation. There is no hardware net-gain claim before FOAK. L0 is therefore where the machines currently exist most fully: as validated, reproducible numerical models rather than as steel and plasma.