Technology

One reactor. One physics AI. Zero carbon.

Our hydrogen plasma reactor melts, reduces, atomizes and quenches in a single continuous step, turning scrap and oxides into spherical premium powder, with water as the only by-product.

The process, step by step

Every stage below is observed, modeled and corrected in real time by our own physics AI model, from plasma temperature in the melt zone to quench rate at the nozzle exit.

Melt

Plasma at ~5,000 K melts the feedstock; hydrogen dissociates into atomic H.

Reduce

Atomic hydrogen strips the surface oxide. Oxygen leaves as water, not CO₂.

Atomize

Supersonic gas expansion breaks droplets to target particle size.

Quench

Adiabatic cooling freezes chemistry and spherical shape in milliseconds.

Collect

An inert chain from reactor to sealed vessel. Powder never sees air.

In motion

See the process in 30 seconds

Watch the full film (2:32) →

GREEN14 engineers working with multiphysics simulation
Physical AI

The reactor is the flywheel. Every run sharpens the next.

Frontier AI has read the internet. It hasn't run a hydrogen plasma reactor. The physics that matters here lives in experiments, not in text, and high-fidelity simulation is too slow to steer a live process. So we built our own loop:

Physics AI model

Our own fast physics AI sweeps several thousand reactor configurations in seconds, grounded in conservation laws.

Multiphysics simulation

Promising candidates are checked in COMSOL: plasma-particle tracing, heat diffusion, and reactive transport at high fidelity.

Reactor runs

The reactor executes only the configurations the AI judges most informative. Every run is a real encounter with the plasma.

ML retraining

Each run produces gigabytes of plasma, particle and thermal data. Models retrain on it, sharpening the physics AI model and the simulation priors. Better data, better models, better reactor.

GREEN14 engineers reviewing simulations and reactor data

The result is a digital twin that finds the settings for target spec, picks the next most valuable experiment, and de-risks every scale-up step before hardware is built.

Why this beats conventional routes

Figures below reflect our Ti64 validation case; economics and footprint vary by material.

ConventionalGREEN14
FeedstockVirgin bar, €50+/kgCircular scrap, €1–5/kg
ProcessMulti-step, 1920s atomization physicsOne continuous step
Carbon~100 kg CO₂e/kg powderTarget ~18 kg CO₂e/kg
Waste30–50% off-spec, exported or downcycledClosed loop, zero waste powder