New superconductors, found with AI.
Better superconductors would cut the cost of power grids and MRI machines. Predicting one is easy now. Making one is not.
We start from MEL, a physics framework built for these materials, generate candidates inside its rules, and rank them by what a lab can actually synthesize.
Built with
- UIUC
- UC Berkeley
- Georgia Tech
- Hyunsung TNC
- CAN Superconductors
- Eloi Materials
From a candidate to a material a lab can confirm.
Anyone with enough GPUs can generate candidates by the million; Google’s papers are full of them. The bottleneck is the furnace. Every stage below exists to get a candidate through one and in front of an instrument. Click any stage to step through.
Stage 01
Classify
Read the material's playbook
Feed the platform a candidate. It reads the atomic structure and lists the quantum orders that structure can host: superconductivity, charge order, spin order, orbital order. Four primitives, one vocabulary.
The platform, running.
A live trace of the discovery pipeline, end to end: classifying candidates, generating new ones, screening, synthesizing, handing them to a collaborating lab to measure. Hover any line to decode the physics behind it.
events
1,242
confirmations
47
p50 latency
1.8 s
±1 K
predicted vs measured Tc · six published superconductors
Run backward over six superconductors the field has already measured, MEL reproduces each critical temperature to within a single degree, with the same parameters for every material. This is retrodiction, the exam with the answers published. The prospective test, on candidates nobody has made yet, is what the partner labs are for.
An outside lab confirmed the mechanism we build on.
In 2026, a team at Stanford and SLAC published in Physical Review Letters, one of the most selective journals in physics. They measured the effect SuperMatics screens for: in a copper-oxide superconductor, charge order and superconductivity strengthen one another instead of competing. That cooperation is the signal our platform uses to rank candidates, and an outside group has now observed it directly.
What they measured
Soft x-ray scattering on a copper-oxide superconductor
Stanford and SLAC followed the charge order in a cuprate as it was cooled through its transition temperature, with enough resolution to see how that order held together below Tc.
What they found
Charge order and superconductivity grow together
Below Tc, the charge order strengthens in step with the superconducting state, and its periodicity locks onto the lattice rather than drifting.
Why it matters for MEL
The mechanism MEL was built around
This cooperative behavior is the one MEL writes down, and the signal our platform ranks candidates on. The full formalism is on the science page.
Who measures next.
Our collaborators run the techniques MEL was designed to be confirmed by: scanning tunneling microscopy at UIUC and Georgia Tech, synthesis at UC Berkeley, with further research discussions underway across leading correlated-electron groups.
Prof. Vidya Madhavan
University of Illinois Urbana-Champaign
University of Illinois Urbana-Champaign
STM / STS · FT-STS
Experimental collaboration · STM/STS, FT-STS
Georgia Institute of Technology
Atlanta · STM / STS spectroscopy
Atlanta · STM / STS spectroscopy
STM / STS
Scanning tunneling microscopy and spectroscopy on MEL candidates
QB3 / Berkeley Nanofabrication Center
UC Berkeley · Waqas Khalid
UC Berkeley · Waqas Khalid
Synthesis · Fabrication
Research infrastructure · synthesis and fabrication
- CAN Superconductors
Czech Republic · HTS manufacturing
Czech Republic · HTS manufacturing
HTS synthesis · Scale-up
Industrial HTS synthesis of MEL candidates · advisory on synthetic routines and scale-up
- Eloi Materials (EML)
Advanced alloys · metal powders · PVD targets
Advanced alloys · metal powders · PVD targets
PVD targets · Powders
Industrial synthesis of MEL candidates · sputtering targets and precursor powders for thin-film and bulk routes
Wilson Sonsini Goodrich & Rosati
WSGR
WSGR
Patent · Corporate
Patent strategy · IP and corporate counsel
What MEL predicted, an independent group has now measured. The platform ranks candidates on that same cooperative coupling, γ < 0.
See the formalismFrom the founder
Why I’m doing this.
I studied chemistry and physics at Berkeley, spent time in the Nanofabrication Center building and measuring devices, then worked on the cryogenics behind superconducting qubits at Rigetti. Two things stuck. Superconductors quietly set the ceiling on a lot of technology that matters. And we still find new ones mostly by accident.
At Rigetti that ceiling was concrete: the qubits were only ever as good as the superconducting films underneath them. James Kim spent the better part of two decades on the theory for why some of these materials work. The math was there long before anyone could use it at scale. What changed isn’t the physics. It’s that generative models, and the compute to run them, finally caught up to it. The search the theory implies is now something you can actually do. I worked on that math with him. My name is on both papers.
I won’t oversell it. MEL doesn’t explain everything about these materials, and a prediction is just a prediction until a lab measures it. So we measure every one, with groups that do this for a living. So far the physics has held.
Davis Rens, Founder
Get in touch
Who we want to hear from.
Three kinds of people move this forward. If you’re not sure you’re one of them, write anyway.
The first market is higher-Tc superconductors, built with CAN Superconductors; heat off the AI chip sits next door. The full map →
Or just contact@supermatics.io. It goes straight to a founder.
Investors
Deep-tech capital at the AI and physics interface. The brief is by invitation.
Read the investor briefResearchers
Condensed-matter theory, experiment, or ML at the frontier of unconventional superconductors. If that is your work, write to us.
research@supermatics.ioBuilders
Founding CEO and Solutions Architect, in Berkeley. Real ownership, foundational work.
See the open roles