AI Finds Two New Superconductors Colder Than the Ones We Use

An international team of physicists says it has used machine learning to screen a nearly limitless number of material combinations and, for the first time, carried two AI-flagged candidates all the way from algorithm to a lab-confirmed superconductor. The materials, named YRu3B2 and LuRu3B2, were identified by the SuperC consortium, a global research collaboration led by Aalto University physicist Päivi Törmä, with the results published in Physical Review Research on June 17, 2026.

There’s a catch the press releases mostly buried. Both materials only turn superconducting below 1 Kelvin, colder than nearly every superconductor already doing useful work in hospitals and physics labs today. The result validates a search method Törmä hopes will eventually screen billions of candidate materials, and it also measures exactly how far the consortium’s own target, a room-temperature superconductor by 2033, still has to travel.

How the Algorithm Picked Its First Real Winners

Superconductivity has been a known phenomenon since 1911, when it was first observed in mercury cooled by liquid helium. More than a century later, finding new superconducting materials is still mostly a matter of luck.

Over the decades researchers have recognized over 7,000 superconductors, but mostly serendipitously

That’s Törmä’s own tally of the problem. The process of testing candidates is so computationally heavy, she says, that scientists have only ever managed to theoretically predict the viability of about 20 of those materials before someone actually found them in a lab. Even a material that looks promising on paper can still turn out to be impossible to synthesize or scale.

SuperC’s pipeline is built to flip that ratio. Instead of testing compounds one at a time, the consortium’s machine-learning-guided discovery of kagome superconductors hunts for a specific geometric signature: electrons arranged into flat energy bands inside a kagome lattice, a geometric arrangement inspired by traditional Japanese basket weaving patterns. The consortium screens suitable candidates from millions of possible materials using quantum geometry and AI, then hands the survivors off for deeper scrutiny. The full pipeline runs in five stages:

• Screen – machine learning models scan enormous numbers of elemental combinations for kagome arrangements likely to produce flat electronic bands.
• Rank – a specialized algorithm narrows that pool down to the handful of combinations most likely to superconduct.
• Calculate – researchers run detailed quantum mechanical calculations on the survivors to test the prediction on paper.
• Synthesize – Rice University’s Center for Quantum Materials, under Professor Emilia Morosan, combines the raw elements into an actual compound.
• Verify – the finished sample goes through magnetization, specific heat, and electrical transport measurements to confirm it really superconducts.

That last step is where YRu3B2 and LuRu3B2 stopped being predictions and became measurements. Morosan’s team arc-melted stoichiometric quantities of high-purity yttrium, ruthenium, and boron in argon atmosphere, then verified crystal structure using a Bruker D8 Advance powder X-ray diffractometer with Rietveld refinement, testing the resulting crystals until the data confirmed genuine, bulk superconductivity.

Two New Superconductors, Far Colder Than the Ones Already at Work

Here’s the detail most of the coverage left out. YRu3B2 becomes superconducting only below 0.81 Kelvin, and LuRu3B2 only below 0.95 Kelvin, according to the measurements in the published paper. That’s colder than liquid helium’s ordinary boiling point of 4.2 Kelvin, the standard coolant for the superconducting magnets inside a hospital MRI machine. Reaching below 1 Kelvin generally means a dilution refrigerator, lab equipment far more specialized than anything installed in a hospital basement. Set next to the rest of the field, the two new materials sit at the coldest end of a very wide range.

YBCO, discovered in 1987, was the material that first let physicists swap costly liquid helium for cheap liquid nitrogen as a coolant. In 1993, scientists discovered a mercury-based copper-oxide ceramic called Hg1223 that reached superconductivity at minus 140 degrees C, or 133 K, a record that stood for more than three decades until a University of Houston team reported 151 K at ambient pressure this past March using a technique called pressure quenching. Lanthanum hydride becomes a superconductor at 250 K under a pressure of 170 gigapascals, a squeeze that confines samples to microscopic volumes that preclude any device integration. YRu3B2 and LuRu3B2 don’t need that kind of pressure. They just need to be more than 150 times colder, measured in absolute temperature, than the ceramic that set this year’s ambient-pressure record.

One of the Two Superconductors Almost Got Left Out

The pipeline isn’t infallible, and its own results show it. LuRu3B2 was initially overlooked by the high-throughput predictions because its computed phonon spectrum showed weakly imaginary modes, suggesting a possible dynamical instability. It was subsequently investigated by analogy with the yttrium compound and confirmed as a bonafide superconductor.

That near-miss is arguably as informative as the discovery itself. The pipeline can miss candidates; domain expertise remains essential alongside the algorithm, as the SuperC research consortium documents in its ongoing work. Machine learning narrowed the haystack, but a person still had to notice the needle the software almost threw out. Coverage of the release drew the same distinction Törmä has been careful to make, calling the two confirmed compounds “a step toward this future,” a modest claim measured against a target still seven years out.

LK-99 and the Cost of Moving Too Fast

Physicists have reason to be cautious with this kind of announcement. In July 2023, a team of South Korean researchers published preprints claiming a copper-doped lead apatite called LK-99 superconducted at up to 400 K (127°C; 260°F) at ambient pressure. The claim went viral, briefly moved Korean and Chinese tech stocks, and triggered a global scramble to replicate it. By mid-August 2023, the consensus was that LK-99 is not a superconductor at room temperature, and is an insulator in pure form. A number of replication attempts identified non-superconducting ferromagnetic and diamagnetic causes for the behavior that had briefly looked like superconductivity.

Princeton University condensed-matter theorist B. Andrei Bernevig was one of the physicists who pushed back hardest during that episode, telling Physics World that “a lot of the stuff early on was rushed and statements from all sides were unchecked,” and adding that he hoped “we never do science like this again.” Bernevig is also a co-author on the new Physical Review Research paper describing YRu3B2 and LuRu3B2, a peer-reviewed result built on modest, specific, sub-1-Kelvin measurements.

Google and Microsoft Are Running the Same Race With a Different Engine

SuperC isn’t the only machine-learning effort hunting for new materials. Google DeepMind’s Graph Networks for Materials Exploration model, announced in Nature in late 2023, predicted 2.2 million new crystal structures, of which 380,000 were the most stable, making them promising candidates for experimental synthesis, and its haul included a compound called Mo5GeB2 flagged as a potential superconductor. Microsoft followed within days with MatterGen, a generative AI model for reverse design of materials, capable of directly designing the structure of new materials based on required properties.

The difference is scale versus follow-through. GNoME’s superconductor candidate remains a computational prediction; nothing in the public record shows it has been synthesized and measured in a lab the way Morosan’s team did with YRu3B2 and LuRu3B2. DeepMind and Microsoft are casting enormously wide nets across the entire periodic table. SuperC is running a narrower, physics-guided search aimed specifically at superconductivity, and it is the one that has, so far, walked a prediction all the way through a real sample and a real measurement.

A 2033 Deadline, Backed by Foundations and One Individual

SuperC’s funding list reads like a mix of institutional science philanthropy and one unusual private backer. The consortium’s work is supported by the Kavli Foundation, Klaus Tschira Stiftung, and Kevin Wells, as well as the Jane and Aatos Erkko Foundation, the Keele Foundation, the Magnus Ehrnrooth Foundation, and the Neste and Fortum Foundation.

Wells is executive director of the Stanford Institute for Theoretical Physics, with a career that included executive positions with Medtronic Diabetes and The Walt Disney Company before he led the networking teams in software engineering at Apple and later moved into funding physics research. Elsewhere he is described as a science philanthropist who provides seed funding to incubate new directions in scientific research, including fundamental physics, quantum science, and condensed matter physics. In the announcement of the joint grant behind this line of research, Wells said the collaboration’s mix of expertise left it “well positioned to make breakthroughs in next generation superconducting materials”.

What that money is chasing is a measurable shift in how much energy computing and power grids waste as heat.

• 0.81 K and 0.95 K – the critical temperatures of YRu3B2 and LuRu3B2, the two compounds SuperC’s pipeline confirmed.
• 7,000-plus – superconductors catalogued since 1911, the large majority found through decades of trial and error.
• 2033 – the year SuperC has set as its own target for a scalable, room-temperature superconductor.
• 2 to 4 percent – the share of global carbon emissions attributed to the information and communications technology sector, a problem a warmer superconductor could ease.

The ICT sector is estimated to contribute between 2 and 4 percent of all carbon emissions globally, and it is projected to continue its upward tick over the coming decades. Superconducting technology that worked at room temperature could theoretically deliver 1,000 times more energy efficiency and 100 times higher clock frequency than today’s conductors, while cutting the cooling costs out of quantum computers, fusion magnets and maglev trains entirely. Törmä’s team thinks its screening pipeline could eventually process candidate materials by the billion, several orders of magnitude beyond what any lab can test today. None of that changes what sits in Morosan’s lab notebook right now: two real, working, peer-reviewed superconductors that need to be colder than almost anything else in the field to do their job.

SuperC’s research goes on public display at Aalto University’s Designs for a Cooler Planet exhibition, running September 1 through October 30, 2026, in Greater Helsinki. Visitors will see the hexagonal kagome pattern that made two new superconductors possible. The label won’t mention that both of them need to be colder than deep space to work.

Frequently Asked Questions

What Is a Superconductor?

A superconductor is a material that carries electric current with zero electrical resistance, meaning no energy is lost as heat. The effect is a quantum phenomenon that, in every material discovered so far, only appears when the material is cooled to very low temperatures.

Why Does Room Temperature Matter So Much?

Every superconductor in practical use today needs expensive, energy-intensive cooling systems to reach the near-absolute-zero temperatures where the effect kicks in. A material that superconducted at room temperature could cut energy loss out of power grids, computers and data centers, which is why Törmä has said such a material would change how the world consumes energy.

What Did the SuperC Consortium Actually Discover?

SuperC identified and experimentally confirmed two new superconductors, YRu3B2 and LuRu3B2, using a machine-learning pipeline to screen candidate materials before Rice University synthesized and tested them. Both compounds only superconduct below 1 Kelvin, far short of room temperature.

What Is a Kagome Lattice?

A kagome lattice is a hexagonal geometric arrangement of atoms named for a traditional Japanese basket-weaving pattern. In YRu3B2 and LuRu3B2, that arrangement forces electrons into flat energy bands, an unusual electronic structure that researchers believe can support superconductivity.

How Close Is a Real Room-Temperature Superconductor?

Not close, by SuperC’s own timeline, which targets 2033. The current ambient-pressure record is 151 Kelvin, set by a University of Houston team in March 2026, and the warmest superconductor ever measured, lanthanum hydride at 250 Kelvin, only works under extreme pressure inside a diamond anvil cell.

Is This Connected to the LK-99 Controversy From 2023?

Not directly, though the two episodes make a useful contrast. LK-99’s claimed room-temperature superconductivity was debunked within weeks in 2023 after independent labs traced its behavior to magnetic impurities. One of the physicists who pushed back hardest against that hype, Princeton’s B. Andrei Bernevig, is a co-author on the new, peer-reviewed SuperC paper.