Computing and Discovery

Clean power and large-scale computation form a loop: energy runs the models, and the models accelerate the science, including fusion itself.

Vision, not specification. This page describes a future that fusion could help make possible — not a product promise, forecast, or performance claim. The Kronos machines today are design and simulation studies: the breeder (Hyperion) and the burner (Aegis / MetroVolt). No hardware net-gain has been demonstrated; construction of the first breeder is planned to begin in Q2 2027, with first-of-a-kind first tritium targeted around 2030.

Discovery has become compute-bound

Whole fields now advance at the pace of available computation. Protein structures, new materials, climate projections, and drug candidates are increasingly found by simulation and machine learning before they are ever made in a lab. The rate of discovery is, in part, an energy question: more clean compute means more searching of the space of possible answers.

Fusion sits inside this loop in an unusual way. The same class of computation that fusion could power is also what makes fusion itself tractable: surrogate models that run design studies orders of magnitude faster than first-principles solvers, and control systems that learn to hold a plasma steady. Kronos builds and uses this stack directly.

A quantum thread

The loop extends into quantum computing, where the breeder’s helium-3 output is relevant: helium-3 dilution refrigeration reaches the millikelvin temperatures that superconducting and spin qubits need. Fusion could thus supply both the power and a key material for the next generation of computers — which in turn help design better fusion.

The honest edge

• The surrogate and control stack is real and in use; the plant it will run is still a study.
• The helium-3 figure is a frozen design number, not delivered supply.
• No economic claims accompany the compute loop.

See the power side in powering advanced intelligence and the material side in fusion and space.