Can AI find the materials abundance needs?

Introduction

If AI is going to help deliver a world of cleaner energy, cheaper manufacturing, better batteries and less resource-constrained growth, materials science is one of the places where that promise has to become real. Modern civilisation depends on a surprisingly small set of known materials. New battery chemistries, efficient solar cells, advanced semiconductors, catalysts for clean fuels and stronger lightweight alloys all depend on discovering substances with the right physical properties.

Google DeepMind’s GNoME system is one of the most ambitious attempts yet to accelerate that search. In 2023, the company reported that GNoME had identified 2.2 million candidate crystal structures, including roughly 380,000 predicted stable materials - a dramatic increase in the known universe of potentially useful inorganic compounds.[Nature]nature.comScaling deep learning for materials discoveryby A Merchant · 2023 · Cited by 1595 — In particular, GNoME models found 2.2 million c…

For advocates of an AI-enabled scientific bloom, the result matters because it suggests that a major research bottleneck may be becoming searchable. But it also exposes another reality: finding a material inside a simulation is not the same thing as turning it into a battery, solar panel, power grid component or industrial product. GNoME shows both the promise of AI-accelerated discovery and the stubborn physical bottlenecks that still stand between prediction and abundance.

What GNoME changed about crystal search

Materials discovery has traditionally been slow because the number of possible atomic arrangements is enormous. Chemists and materials scientists have often relied on a combination of theory, experience and trial-and-error experimentation to identify promising compounds.

GNoME, short for Graph Networks for Materials Exploration, approaches the problem differently. It uses graph neural networks - AI systems that represent atoms and their bonds as connected networks - to predict whether a proposed crystal structure is likely to be stable. The model was trained using data from the Materials Project, a large open database of known and computationally evaluated materials. It then generated new candidate structures and repeatedly improved itself through an active-learning loop that combined machine learning with more computationally expensive physics calculations. Nature[Google DeepMind]biomimicryinnovationlab.comNew materials discovered with deep…Dec 7, 2023 — Of the 2.2 million new crystals predicted by GNoME, 380,000 are the most stable, maki…

The scale of the result was what attracted attention. DeepMind reported that GNoME identified 2.2 million structures predicted to be stable relative to existing databases. Around 381,000 of these appeared on the updated thermodynamic stability landscape known as the convex hull, meaning they were considered especially promising candidates for real materials. Nature[Berkeley Lab News Center]newscenter.lbl.govgoogle deepmind new compounds materials projectBerkeley Lab News CenterGoogle DeepMind Adds Nearly 400000 New Compounds to…Nov 29, 2023 — GNoME researchers ultimately produced 2.2 m…

That matters because stable inorganic materials are the foundation of much modern technology. Researchers hope that among these newly proposed structures could be:

• Better battery materials and solid-state electrolytes.
• Improved catalysts for industrial chemistry.
• New superconductors.
• More efficient photovoltaic materials.[deepmind.google]deepmind.googlemillions of new materials discovered with deep learningGoogle DeepMindMillions of new materials discovered with deep learningNov 29, 2023 — AI tool GNoME finds 2.2 million new crystals, includ…
• Novel semiconductor compounds.
• Stronger and lighter engineering materials.[Google DeepMind]biomimicryinnovationlab.comNew materials discovered with deep…Dec 7, 2023 — Of the 2.2 million new crystals predicted by GNoME, 380,000 are the most stable, maki… Science The broader significance is not any single crystal. It is the possibility that AI can transform discovery from a largely artisanal process in[science.org]science.orgMaterials-predicting AI from DeepMind could revolutionize…Nov 29, 2023 — Researchers report that with a new artificial intelligence (A… to something closer to large-scale search. In the same way that AlphaFold made protein structures dramatically easier to access, GNoME suggests that huge regions of materials space may become computationally explorable rather than remaining effectively invisible.

For the larger AI bloom argument, this is an important pattern. Intelligence is being used not merely to automate existing work but to widen the range of scientific possibilities humans can examine.

Why stability is not the same as usefulness

The most common misunderstanding about GNoME is that it discovered 380,000 immediately useful materials.

It did not.

The key claim is much narrower: the system predicted materials that appear thermodynamically stable according to established computational methods. Stability is a necessary condition for usefulness, but it is far from sufficient. Nature[axios]axios.comTraditionally, materials science relied on tweaking existing stable compounds, whereas GNoME uses deep learning to model atoms and molecu… A material can be stable and still be commercially worthless. What matters in practice depends on many other properties:

A battery electrolyte, for example, must do much more than merely exist. It must conduct ions efficiently, survive repeated charging cycles, remain safe under stress, work across temperatures, and be affordable to manufacture at scale.

This is why materials researchers often describe discovery as a funnel rather than a finish line. Millions of candidates may ultimately yield only a handful of transformative technologies.

Some researchers have also emphasised that AI systems can be much better at predicting stable structures than predicting whether those structures can actually be synthesised in a laboratory. Materials may require highly specific conditions, rare ingredients or difficult processing steps that make them impractical even if they are theoretically stable. ScienceDirect[cypris]cypris.aiHow Generative Models, Graph Neural Networks, and…The Experimental Validation Bottleneck persists as computational predictions f… This distinction matters for public understanding of AI abundance. The optimistic narrative sometimes jumps directly from”millions of new materials” to visions of abundant clean energy and revolutionary manufacturing. The actual path is much longer and more uncertain.

The synthesis bottleneck is now the central problem

If AI can generate candidate materials faster than humans can test them, the bottleneck moves.

The challenge becomes synthesis: actually making the material.

This is where many materials scientists believe the next major acceleration problem lies. Computational prediction has become dramatically faster over the last decade, but laboratory validation remains expensive, time-consuming and dependent on specialised expertise.[arXiv]arxiv.orgAccelerating computational materials discovery with artificial intelligence and cloud high-performance computing: from large-scale s… ScienceDirect The contrast is striking. An AI system can propose thousands of candidates in hours. Producing and characterising a single material may take[sciencedirect.com]sciencedirect.comAdvancing materials discovery through artificial intelligenceby M Otyepka · 2025 · Cited by 23 — The synthesisability, stabi… days, weeks or months.

Researchers at Lawrence Berkeley National Laboratory have been working on this problem through A-Lab, an autonomous materials synthesis system that combines robotics, machine learning and experimental planning. In demonstrations linked to the broader GNoME ecosystem, automated laboratories were able to synthesise substantial numbers of proposed materials without continuous human intervention.[Nature]nature.comGoogle AI and robots join forces to build new materialsNov 29, 2023 — Tool from Google DeepMind predicts nearly 400,000 stable subs…

The importance of this development is easy to miss. If AI-generated scientific hypotheses expand by a factor of one hundred but experimental testing remains unchanged, the real rate of discovery may barely improve. The bottleneck simply shifts from imagination to validation.

For AI bloom scenarios, this is one of the most important lessons from GNoME. Scientific acceleration is unlikely to come from a single breakthrough model. It requires an entire chain:

The slowest stage determines the overall speed.

From predicted material to public benefit

Even successful synthesis is only the beginning.

A material that works in a laboratory often faces years of additional development before it affects everyday life. Scaling a material from milligram quantities to industrial production can reveal entirely new problems.

Researchers must determine:

Many apparently promising materials never survive this process.

This is why the strongest case for GNoME is not that it instantly creates abundance. It is that it increases the number of potentially valuable starting points. Instead of searching a tiny fraction of possible compounds, scientists can search a much larger landscape.[Nature]nature.comAI-powered open-source infrastructure for accelerating…by M Salas · 2026 · Cited by 5 — The review further discusses the importance of…

Historically, transformative materials have often arrived unpredictably. Silicon enabled modern computing. Lithium-ion chemistry reshaped portable electronics and electric vehicles. Synthetic fertiliser changed agriculture. New catalysts transformed industrial production.

The difficulty has always been that useful materials are needles hidden in an enormous chemical haystack. GNoME’s contribution is reducing the size of the search problem.

What the early validation evidence actually shows

One reason GNoME attracted attention beyond the AI community is that the work did not stop at computer predictions.

DeepMind and collaborators pointed to hundreds of cases where predicted materials had already been independently synthesised or experimentally validated. Reports around the project noted that more than 700 predicted materials had been confirmed experimentally, while separate work with automated laboratories demonstrated successful synthesis of many AI-proposed compounds.[Venturebeat]venturebeat.comgoogle deepminds materials ai has already discovered 2 2 million new crystalsGoogle DeepMind's materials AI has already discovered…29 Nov 2023 — In 17 days, the AI system was able to identify 2.2 mill… Reddit That evidence is significant because it addresses a common criticism of AI-generated science: that the systems may produce mathematically pla[reddit.com]reddit.comExternal researchers have independently created 736 of GNoME's new materials in the…Read more… usible outputs that fail in the real world.

At the same time, the validation numbers illustrate the scale of the remaining challenge. Hundreds of confirmed materials sounds impressive. Yet it remains a tiny fraction of millions of predictions.

This is not necessarily a failure. Materials discovery has always involved high attrition rates. The point is that AI has dramatically increased the number of possibilities entering the pipeline. Whether society can process that pipeline efficiently becomes the next question.

What GNoME means for the AI bloom argument

GNoME is not proof that AI will create a post-scarcity civilisation. It is evidence for a narrower claim: machine intelligence can sometimes expand the search space of science far faster than traditional methods.

That matters because many of the largest constraints on human flourishing are ultimately physical.

Cleaner energy depends on materials.

Better computing depends on materials.

Advanced robotics depends on materials.

Carbon capture, fusion systems, superconductors, desalination infrastructure and next-generation batteries all depend on finding combinations of atoms that perform better than today’s options.

If AI systems become increasingly capable of exploring those possibilities, scientific progress may become less limited by human trial-and-error. The long-term significance is not that one model discovered millions of crystals. It is that scientific search itself may be becoming more scalable.

Yet GNoME also highlights why abundance is not automatic. The bottlenecks increasingly lie in experimentation, manufacturing, supply chains, institutions and deployment. Intelligence may become abundant before physical implementation does.

That is one reason the strongest versions of the AI bloom thesis focus on systems rather than models. Scientific abundance requires prediction, validation, engineering and diffusion to accelerate together. GNoME provides a glimpse of what happens when one stage of that chain suddenly speeds up. The unanswered question is whether the rest of the chain can keep pace.

Amazon book picks

Further Reading

Books and field guides related to Can AI find the materials abundance needs?. Use these as the next step if you want deeper reading beyond the article.

Book

Human Compatible

By Stuart Russell

Explores how powerful AI could advance humanity if aligned with human values.

See on Amazon

Book

The Coming Wave

By Mustafa Suleyman

Directly addresses whether advanced AI can deliver broad civilisational benefits while remaining governable.

See on Amazon

Book

Power and Progress

By Daron Acemoglu, Simon Johnson

Explains why public institutions matter for distributing technology gains.

See on Amazon

Book

The Entrepreneurial State

By Mariana Mazzucato

Strong conceptual fit for public compute infrastructure.

See on Amazon

Browse more on Amazon:Human CompatibleThe Coming WavePower and Progress

As an Amazon Associate I earn from qualifying purchases.

eBay marketplace picks

Marketplace Samples

Example marketplace items related to this page. Use the search link to explore similar finds on eBay.

Shop location

UsingUSA USA USAUKCanadaAustraliaIreland

Example eBay listing

Crystal Caves - Basing Kit/ Basing Bits/ Tufts/ Model Scenery/ Goon Master

SearcheBay.co.uk: crystal model kit

Browse similar oneBay.co.uk

Example eBay listing

BANDAI MG 1/100 UNICORN GUNDAM Luminous Crystal Body Base Limited Model Kit

SearcheBay.co.uk: crystal model kit

Browse similar oneBay.co.uk

Example eBay listing

Crystals 2 - Basing Kit/ Basing Bits/ Tufts/ Model Scenery/ Goon Master

SearcheBay.co.uk: crystal model kit

Browse similar oneBay.co.uk

Example eBay listing

MG 1/100 Gundam Base Limited Unicorn Gundam Luminous Crystal Body Model Kit

SearcheBay.co.uk: crystal model kit

Browse similar oneBay.co.uk

Browse more oneBay.co.uk

Example items shown for inspiration; availability and pricing can change. Branchoria may earn a commission if you purchase through outbound eBay links.

Endnotes

1. Source: nature.com
Link:https://www.nature.com/articles/s41586-023-06735-9

2. Source: deepmind.google
Title: millions of new materials discovered with deep learning
Link:https://deepmind.google/blog/millions-of-new-materials-discovered-with-deep-learning/

3. Source: time.com
Link:https://time.com/6340681/deepmind-gnome-ai-materials/

4. Source: axios.com
Link:https://www.axios.com/2023/12/02/ai-robotics-new-materials

5. Source: sciencedirect.com
Link:https://www.sciencedirect.com/science/article/pii/S2352940725003981

6. Source: cypris.ai
Link:https://www.cypris.ai/insights/ai-accelerated-materials-discovery-in-2025-how-generative-models-graph-neural-networks-and-autonomous-labs-are-transforming-r-d

7. Source: arxiv.org
Link:https://arxiv.org/abs/2401.04070

8. Source: nature.com
Link:https://www.nature.com/articles/d41586-023-03745-5

9. Source: venturebeat.com
Title: google deepminds materials ai has already discovered 2 2 million new crystals
Link:https://venturebeat.com/ai/google-deepminds-materials-ai-has-already-discovered-2-2-million-new-crystals

10. Source: reddit.com
Link:https://www.reddit.com/r/materials/comments/186u1vn/deepmind_millions_of_new_materials_discovered/

11. Source: sciencedirect.com
Link:https://www.sciencedirect.com/science/article/pii/S0262407923022145

12. Source: sciencedirect.com
Link:https://www.sciencedirect.com/science/article/pii/S2666546825001375

13. Source: arxiv.org
Link:https://arxiv.org/html/2508.03278v1

14. Source: law.berkeley.edu
Link:https://www.law.berkeley.edu/wp-content/uploads/2024/02/Millions-of-new-materials-discovered-with-deep-learning-Google-DeepMind.pdf

15. Source: reddit.com
Link:https://www.reddit.com/r/mlscaling/comments/187q6za/gnome_graph_nn_for_discovering_crystals_alab/

16. Source: reddit.com
Link:https://www.reddit.com/r/singularity/comments/186t59y/deepmind_millions_of_new_materials_discovered/

17. Source: nature.com
Link:https://www.nature.com/articles/s43246-026-01105-0

18. Source: newscenter.lbl.gov
Title: google deepmind new compounds materials project
Link:https://newscenter.lbl.gov/2023/11/29/google-deepmind-new-compounds-materials-project/

19. Source: science.org
Link:https://www.science.org/content/article/materials-predicting-ai-deepmind-could-revolutionize-electronics-batteries-and-solar

20. Source: wired.com
Title: Google Deep Mind’s AI Dreamed Up 380,000 New Materials
Link:https://www.wired.com/story/an-ai-dreamed-up-380000-new-materials-the-next-challenge-is-making-them

21. Source: github.com
Link:https://github.com/google-deepmind/materials_discovery

22. Source: ft.com
Link:https://www.ft.com/content/f841e9e0-c9c6-49ab-b91c-6d7bea2a3940?syn-25a6b1a6=1

23. Source: biomimicryinnovationlab.com
Title: Google Deepmind
Link:https://www.biomimicryinnovationlab.com/blog/google-deepmind-new-materials

Additional References

24. Source: medium.com
Link:https://medium.com/%40HitachiVentures/ai-is-powering-the-future-of-material-science-from-lab-to-real-world-breakthroughs-2f92cf56ed90

25. Source: discussion.fool.com
Link:https://discussion.fool.com/t/ot-ai-makes-mistakes-with-duplicate-structures-that-haunt-crystallography-databases/122599

26. Source: arstechnica.com
Title: googles deepmind finds 2 2m crystal structures in materials science win
Link:https://arstechnica.com/ai/2023/11/googles-deepmind-finds-2-2m-crystal-structures-in-materials-science-win/

27. Source: thelab.brookesbell.com
Title: googles deepmind ai tool makes material science breakthrough 158802
Link:https://thelab.brookesbell.com/about/news/googles-deepmind-ai-tool-makes-material-science-breakthrough-158802/

28. Source: pubs.acs.org
Title: Publications Artificial Intelligence Driving Materials Discovery?
Link:https://pubs.acs.org/doi/10.1021/acs.chemmater.4c00643

Published: November 2023

29. Source: mercatus.org
Title: future materials science ai automation and policy strategies
Link:https://www.mercatus.org/research/policy-briefs/future-materials-science-ai-automation-and-policy-strategies

30. Source: fanaticalfuturist.com
Title: googles deepmind ai just invented 380000 new compounds
Link:https://www.fanaticalfuturist.com/2023/12/googles-deepmind-ai-just-invented-380000-new-compounds/

31. Source: freethink.com
Title: google ai discovers 2 2 million new materials
Link:https://www.freethink.com/robots-ai/google-ai-discovers-2-2-million-new-materials

32. Source: linkedin.com
Title: google ai discovers new materials 5838108
Link:https://www.linkedin.com/news/story/google-ai-discovers-new-materials-5838108/

33. Source: linkedin.com
Title: ai tool discovers 22m structures 5558185
Link:https://www.linkedin.com/news/story/ai-tool-discovers-22m-structures-5558185/