How it works
From a requirement to a decision.
Our AI agents, GPU simulations, and partner labs work as one program that turns what your device needs into a signed decision.
The platform
What happens to your brief.
AI agents turn your requirement into exact limits. Our models then generate and simulate the candidate routes on GPUs, and partner labs test whatever could still change the answer.
The SuperMatics platform
AI agents run every step
- Plan the runs
- Call the solvers
- Read every result
- Draft the record
You bring
Your requirement
Current, field, temperature, size, and cost
Your device
CAD as STEP, IGES, or mesh
Your line
Tools, precursors, and furnace limits
On the platform · GPUs
LLM agents
Read
Your brief, drawings, and papers become exact limits.
Generative models
Generate
Our models propose new materials, architectures, and process routes.
GPU-parallel sweeps
Simulate
We simulate every route at your operating point.
Ranking algorithms
Rank
We grade each route against every limit and record its margin.
Would a test change the ranking?
If yes, the tests go to the bench and their results retrain the models. If no, the ranking stands.At the bench · Partner labs
Partner labs
Make
Partner labs make the samples, at bench or pilot scale.
Partner benches
Measure
Each sample is tested at your operating point.
Active learning
Learn
Every result, hit or miss, retrains the models.
You get
A decision record
A verdict on every limit, with what would have to change to clear it, signed by a founder.
A route to production
Pilot runs with CAN Superconductors
Made and measured with
Lab work
Measurement
Pilot production
The difference
We rank the whole route.
A general materials platform stops at the material. We also grade the substrate it sits on and the process that makes it, against every number your device needs.
- The material
- REBCO
- What it touches
- on Hastelloy
- How it is made
- by MOCVD
- Copper stabilizer 20 µm
- Silver cap 2 µm
- REBCO 1.5 µm
- Buffer stack 0.2 µm
- Hastelloy substrate 50 µm
Representative record · not a program result
1 km city-grid cable · 3 kA per phase · 77 K
Verdict
Near-feasible · does not clear the brief as written
- 3 clear
- 1 close
- 2 miss
| Constraint | Value | Against the brief · better → |
|---|---|---|
| Critical current, Measured, short tape | 460 A/cm-width · 77 K · self-field | Clears, with margin |
| Operating margin, Computed | 1.5× at 3 kA | Clears |
| Joint resistance, Measured, short tape | 14 nΩ, soldered | Clears against 20 nΩ |
| AC loss, Computed | 0.7 W/m per phase · 50 Hz | Inside the budget, with little room |
| Cost per kA·m, Computed | 1.3× your ceiling | Over the ceiling |
| Piece length, From the published route | 600 m | Short of the 1 km you need |
- Measured, short tape
- Computed
- From the published route
- Values at the operating point are interpolated, never extrapolated.
Our own materials
How we search trillions of compositions.
Most groups treat the search for new superconductors as a race for compute. We cut the problem down to the variables that decide it.
The Modulated Electron Lattice (MEL) framework, licensed exclusively to SuperMatics, predicts which few variables decide whether a material superconducts. GPU sweeps and models trained on real measurements screen only those, so we can search trillions of compositions at a fraction of the compute.
We plan to license the materials we own for a royalty on every unit made.
Trillions of possible compositions.
Brute forceFull physics, one candidate at a time.
Our pipelineThe critical variables, screened in parallel.
Five proposed, in synthesis
Five
proposed materials in synthesis
100 kelvin
Predicted to superconduct at 100 kelvin or higher, warmer than liquid nitrogen at 77 kelvin
At scale
Each was chosen because it can be manufactured at scale.
None has yet been independently established as superconducting.
An independent experiment at Stanford and SLAC reported a result consistent with a central MEL prediction, with no involvement from SuperMatics. It does not validate the entire framework.The paper, Physical Review Letters (opens in a new tab)
Your data and IP
We agree on data and IP before the first run.
Your inputs stay yours
Designs, materials, and confidential data remain your property.
Ownership of results is agreed up front
Use rights, field exclusivity, joint ownership, or a license are set before any work begins.
Each program has its own workspace
It is encrypted in transit and at rest, and your team signs in with your company’s single sign-on.
Bring us a requirement.
Tell us what your device needs and what is holding it back.
Investors: request the deck →