Skip to content
ContentLora

    Tip: press / anywhere to search.

    technology

    GNoME (Graph Networks for Materials Exploration)

    Also known as GNoME, Graph Networks for Materials Exploration

    GNoME is a Google DeepMind materials-discovery project, published in Nature in November 2023, that used graph neural networks to identify 2.2 million crystal structures, 381,000 of them counted as newly discovered stable materials.[1] It is the best-known test of whether AI can find new materials, and critics argue few of its predictions are both novel and useful.[2]

    Editor reviewedUpdated AI for scienceArtificial intelligenceScience
    Key facts

    GNoME is a large-scale AI screen of which inorganic crystals could exist, built by Google DeepMind. Its 2023 Nature paper reported an order-of-magnitude gain in discovery efficiency and 2.2 million predicted crystal structures.[3][1] Inorganic crystals matter because chips, batteries and solar panels rely on them, and only stable crystals are useful.[4] Google DeepMind said its most stable predictions include candidates for uses ranging from superconductors to next-generation batteries.[5]

    How it works

    Finding new inorganic crystals has historically relied on expensive trial and error.[6] GNoME trains graph neural networks, which represent a crystal as atoms connected by bonds, to predict whether a candidate structure is stable. Candidates are checked with first-principles quantum-mechanical calculations; the paper describes hundreds of millions of them.[1][7] The project started from 48,000 stable crystals identified in earlier studies.[3] For scale, Google DeepMind noted that computational screening led by the Materials Project and other groups had found 28,000 new materials over the previous decade.[8]

    What it reported

    The team identified 2.2 million structures below the “convex hull”, the computed boundary of thermodynamic stability, and counted 381,000 as newly discovered stable materials.[1] At publication, 736 of its stable structures had been independently made in labs.[9] The hundreds of millions of calculations also produced learned interatomic potentials, fast AI stand-ins for quantum calculations used in molecular-dynamics simulations.[7]

    Sharing the results

    Google DeepMind said it would contribute 380,000 materials that GNoME predicts to be stable to the Materials Project database.[10] On the same day, a Lawrence Berkeley National Laboratory team published a second Nature paper on autonomous synthesis, the A-Lab, whose targets came from Materials Project and Google DeepMind stability data.[11][12] That pairing made GNoME’s predictions a live test of self-driving-labs, and the test became contested.[13]

    Criticism

    Computed stability is not the same as a useful new material. MIT Technology Review reported in December 2025 that researchers found scant evidence of compounds meeting the “trifecta of novelty, credibility, and utility”, and that many predictions were trivial variations of known structures.[2] The same report said AI materials discovery had yet to produce a convincing big win.[14] A separate critique of the A-Lab argued that about two-thirds of claimed successes were likely known, compositionally disordered versions of the predicted ordered compounds.[13] In January 2026 the A-Lab authors published a correction. It said their novelty claims meant materials new to the prediction platform, not necessarily new to science, and that a re-analysis confirmed 36 of 40 reported syntheses.[15] The correction concerns the A-Lab paper, not GNoME’s calculations, but it shows how far “predicted stable” can sit from “new material”.[15]

    What came next

    Generative models now try to design crystals with target properties directly. MatterGen, published in Nature in January 2025, produced structures more than twice as likely to be new and stable as earlier generative models, and one synthesised example landed within 20% of its target property.[16][17] Investors have also funded start-ups such as Periodic Labs, whose first goal is better superconductors.[18]

    Questions readers ask

    Did GNoME really discover 2.2 million materials?

    It identified 2.2 million computed structures below the known stability hull and counted 381,000 as new stable materials; these are predictions, and 736 had been made experimentally by publication.[1][9]

    Why are critics sceptical?

    Researchers quoted by MIT Technology Review found scant evidence that predicted compounds combine novelty, credibility and utility, describing some as merely trivial variations of known ones.[2]

    Can researchers use GNoME's predictions?

    Yes. Google DeepMind said it would contribute 380,000 predicted-stable materials to the open Materials Project database.[10]

    Has AI found a breakthrough material yet?

    As of December 2025, MIT Technology Review reported no convincing big win from AI materials discovery.[14]

    Sources

    Each numbered claim is a statement we checked against the sources listed with it. Status shows how well established it is.

    1. [1]

      Google DeepMind's GNoME work, published in Nature in November 2023, used graph networks to identify 2.2 million crystal structures, of which it counted 381,000 as newly discovered stable materials. confirmedas of 2023-11-29

    2. [2]

      MIT Technology Review reported that researchers who examined GNoME's predicted compounds found scant evidence of novelty, credibility and utility, describing some as merely trivial variations of known compounds. reportedas of 2025-12-15

    3. [3]

      GNoME built on 48,000 stable crystals identified in earlier studies and claimed an order-of-magnitude improvement in materials-discovery efficiency. confirmedas of 2023-11-29

    4. [4]

      Google DeepMind notes that technologies from computer chips and batteries to solar panels rely on inorganic crystals, which must be stable to be useful. confirmedas of 2023-11-29

    5. [5]

      Google DeepMind said GNoME's 380,000 most stable predictions include candidates with potential uses ranging from superconductors to next-generation batteries. confirmedas of 2023-11-29

    6. [6]

      The GNoME authors describe the discovery of inorganic crystals as bottlenecked by expensive trial-and-error approaches. confirmedas of 2023-11-29

    7. [7]

      GNoME's hundreds of millions of first-principles calculations also yielded learned interatomic potentials usable in molecular-dynamics simulations. confirmedas of 2023-11-29

    8. [8]

      Google DeepMind noted that computational approaches led by the Materials Project and other groups had discovered 28,000 new materials over the decade before GNoME. confirmedas of 2023-11-29

    9. [9]

      The GNoME paper reported that 736 of its predicted stable structures had already been independently realized experimentally. confirmedas of 2023-11-29

    10. [10]

      Google DeepMind said it would contribute 380,000 materials that GNoME predicts to be stable to the Materials Project database. confirmedas of 2023-11-29

    11. [11]

      Alongside GNoME, a Lawrence Berkeley National Laboratory team published a second Nature paper showing how the AI predictions could feed autonomous materials synthesis. confirmedas of 2023-11-29

    12. [12]

      The A-Lab's synthesis targets were identified using large-scale ab initio phase-stability data from the Materials Project and Google DeepMind. confirmedas of 2026-01-19

    13. [13]

      A March 2024 PRX Energy perspective concluded that no new materials had been discovered in the A-Lab work and that about two-thirds of claimed successes were likely known disordered versions of predicted compounds. disputedas of 2024-03-07

    14. [14]

      MIT Technology Review reported in December 2025 that AI materials discovery had not yet produced a convincing big win or a new miracle material. reportedas of 2025-12-15

    15. [15]

      A January 2026 author correction to the A-Lab paper said its novelty claims meant materials new to the prediction platform, not necessarily new to science; a manual re-analysis confirmed 36 of the 40 reported successes, with four inconclusive, and one compound was removed because it had been in the training data. confirmedas of 2026-01-19

    16. [16]

      MatterGen, a generative model for inorganic materials published in Nature in January 2025, produced structures more than twice as likely to be new and stable as earlier generative models. confirmedas of 2025-01-16

    17. [17]

      The MatterGen team synthesized one generated structure and measured its target property within 20% of the design value. confirmedas of 2025-01-16

    18. [18]

      Periodic Labs emerged from stealth in September 2025 with a $300 million seed round, with a first goal of inventing better superconductors. confirmedas of 2025-09-30

    Revision history (2)
    1. Page created.
    2. Expanded with GNoME's Materials Project contribution, the pre-GNoME baseline and the 2026 A-Lab correction; corrected the critique wording to 'some' trivial variants.

    Created Oct 10, 2026. Last reviewed by an editor on Oct 10, 2026. Next scheduled review: Jan 10, 2027.

    Cite this page

    "GNoME (Graph Networks for Materials Exploration)." ContentLora, updated Oct 10, 2026. https://contentlora.com/wiki/gnome-materials

    Spotted an error? Suggest a correction or emailcorrections@contentlora.com.

    Keep exploring

    Connections
    Self-driving labs Predictions used as targe…Google DeepMind Developed byThe Materials Proje… Predicted materials contr…
    ☰ Contents