Skip to content
ContentLora

    Tip: press / anywhere to search.

    organization

    The Materials Project

    Also known as Materials Project, MP, materialsproject.org

    The Materials Project is a US Department of Energy effort, hosted by Lawrence Berkeley National Laboratory since 2011, that pre-computes the properties of materials and makes the data publicly available.[1][2] It became a major open repository for AI-predicted materials when Google DeepMind contributed nearly 400,000 predicted stable compounds from its GNoME model in 2023.[3]

    Editor reviewedUpdated Materials scienceScienceAI for science
    Key facts

    What it is

    The Materials Project describes itself as an effort from the US Department of Energy to “pre-compute properties of ‘materials’ and make this data publicly available”. Its aim is to speed up materials discovery.[1] Its scope covers inorganic crystals and molecules, with applications in batteries, solar energy, water splitting, optoelectronics and catalysis.[1] The project is hosted by Lawrence Berkeley National Laboratory and began in 2011. By November 2023 it had more than 400,000 registered users.[2]

    The idea is simple. Instead of making and measuring every candidate material in a lab, researchers compute its structure, stability and properties first, then test only the promising ones.[1]

    The AI moment: GNoME

    In November 2023 Google DeepMind published GNoME (Graph Networks for Materials Exploration). The deep-learning model predicted 2.2 million new crystal structures, of which about 380,000 were judged the most stable. DeepMind contributed nearly 400,000 of the new compounds to the Materials Project, the largest addition since the project began.[3][2] DeepMind said external researchers had already independently synthesised 736 of the predicted materials.[4] Berkeley Lab said the materials could be relevant to carbon capture, photocatalysis, thermoelectrics and transparent conductors.[5]

    The autonomous-lab test and its correction

    Alongside GNoME, Berkeley Lab reported that its robotic A-Lab had made 41 new compounds from 58 targets in 17 days of unattended operation.[6] The current version of the A-Lab paper reports a smaller result: 36 compounds realised from 57 targets over 17 days.[7] In early 2024 chemists from University College London and Princeton challenged the X-ray analysis. They argued that about two-thirds of the targets were ordered versions of already known disordered compounds.[8] In January 2026 Nature published an author correction. It acknowledged the concerns and clarified that “novel” meant new to the prediction platform, not necessarily new to science.[9]

    Why it matters

    The Materials Project is the shared infrastructure behind much computational and AI-driven materials work, and its data are publicly available.[1][2] The GNoME and A-Lab episode shows both the promise and the open question: predicted stability is cheap to compute, but whether a material can be made, and is truly new, still needs careful experiments.[4][9] Related pages cover GNoME, self-driving-labs and the AI materials discovery debate. For how the AI side works, see how AI discovers molecules and materials.

    How GNoME and the Materials Project fit together

    GNoME, built by google-deepmind, is a graph neural network trained with “active learning”. Candidate structures are checked with density functional theory calculations, and the results are fed back into the next round of training. DeepMind said this raised its stability-prediction discovery rate from around 50% to 80%.[10] Berkeley Lab said GNoME was trained on workflows and data the Materials Project had built up over a decade. Some GNoME computations were then used with Materials Project data to choose targets for the A-Lab.[11]

    Among the predictions, DeepMind highlighted 52,000 new layered compounds similar to graphene and 528 potential lithium-ion conductors.[12] The contributed dataset records how the atoms are arranged (the crystal structure) and how stable the material is (its formation energy). Berkeley Lab said that by November 2023 more than four papers citing the Materials Project were published every day on average.[13] Whether many of the predicted compounds are useful is still disputed; see the AI materials discovery debate.

    Questions readers ask

    What is the Materials Project?

    A Department of Energy effort to pre-compute the properties of materials and publish the data openly, to speed up materials discovery.[1]

    What did Google DeepMind add to the Materials Project?

    In November 2023 DeepMind contributed nearly 400,000 compounds predicted to be stable by its GNoME model, from 2.2 million predicted crystal structures.[3]

    Have the AI-predicted materials actually been made?

    Some have. DeepMind said external researchers had independently synthesised 736 GNoME predictions, but claims that an autonomous lab made dozens of new ones were later corrected.[4][9]

    Who uses the Materials Project?

    It had more than 400,000 registered users worldwide by November 2023.[2]

    Sources

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

    1. [1]

      The Materials Project is a US Department of Energy effort to pre-compute the properties of inorganic crystals and molecules and make the data publicly available to speed up materials discovery. confirmedas of 2026-10-10

    2. [2]

      The Materials Project is hosted by Lawrence Berkeley National Laboratory, began in 2011, and had over 400,000 registered users by November 2023. confirmedas of 2023-11-29

    3. [3]

      Google DeepMind's GNoME model predicted 2.2 million new crystal structures, of which about 380,000 were judged most stable, and contributed them to the Materials Project in November 2023. confirmedas of 2023-11-29

    4. [4]

      DeepMind said external researchers had independently synthesised 736 of GNoME's predicted materials. confirmedas of 2023-11-29

    5. [5]

      Berkeley Lab said materials first identified as promising leads from Materials Project data had been experimentally confirmed to show useful properties for carbon capture, photocatalysis, thermoelectrics and transparent conductors for solar cells, touch screens or LEDs. confirmedas of 2023-11-29

    6. [6]

      Berkeley Lab's autonomous A-Lab was reported in November 2023 to have synthesised 41 new compounds out of 58 targets during 17 days of unattended operation. confirmedas of 2023-11-29

    7. [7]

      The A-Lab, an autonomous lab for solid-state synthesis described in Nature in November 2023, realized 36 compounds from 57 targets over 17 days of continuous operation, according to the current version of the paper. confirmedas of 2026-10-10

    8. [8]

      In early 2024 chemists from University College London and Princeton argued that the A-Lab's X-ray diffraction analysis had systematic errors and that about two-thirds of its predicted compounds were ordered versions of already known disordered compounds. confirmedas of 2024-01-16

    9. [9]

      In January 2026 Nature published an author correction to the A-Lab paper acknowledging concerns about structure identification and clarifying that "novel" meant new to the prediction platform, not necessarily new to science. confirmedas of 2026-01-18

    10. [10]

      GNoME is a graph neural network trained with active learning, in which candidates are checked with density functional theory calculations and fed back into training; DeepMind said this raised its stability-prediction discovery rate from around 50% to 80%. confirmedas of 2023-11-29

    11. [11]

      Berkeley Lab said GNoME was trained using workflows and data the Materials Project had developed over a decade, and that some GNoME computations were used with Materials Project data to test the A-Lab. confirmedas of 2023-11-29

    12. [12]

      Among GNoME's predictions, DeepMind highlighted 52,000 new layered compounds similar to graphene and 528 potential lithium-ion conductors. confirmedas of 2023-11-29

    13. [13]

      Berkeley Lab said that, on average, more than four papers citing the Materials Project were published every day as of November 2023, and that the dataset records crystal structures and formation energies. confirmedas of 2023-11-29

    Revision history (2)
    1. Page created.
    2. Aligned A-Lab figures with the current paper (36 of 57); added how GNoME was trained and works, its highlighted candidates and Materials Project usage.

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

    Cite this page

    "The Materials Project." ContentLora, updated Oct 10, 2026. https://contentlora.com/wiki/materials-project

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

    Keep exploring

    Connections
    GNoME Largest contributor of pr…Self-driving labs Predictions tested by
    ☰ Contents