CuspAI brings together Nvidia, Meta and Samsung to create materials with AI

What happened?

On Monday, British artificial intelligence lab CuspAI announced the launch of the AI Materials Foundry, a coalition bringing together more than 45 companies and research labs from around the world. Founding partners include tech giants Nvidia, Meta, Samsung, Hyundai Motor and the Lawrence Berkeley National Laboratory, according to Reuters. The initiative has initial funding of $450 million, a figure that reflects the industry's growing bet on AI applied to materials science.

CuspAI, founded just two years ago in Cambridge, UK, aims to build a 'search engine' for materials that don't yet exist. The platform will combine generative AI models to propose new molecular structures, computational simulations to predict their physical and chemical properties, and autonomous robots to synthesize and test the most promising candidates in the lab. This is intended to dramatically accelerate the materials discovery cycle, which traditionally can take decades from conception to commercialization.

The AI Materials Foundry is not the first project of its kind, but it is the one with the largest financial backing and the broadest partner base. Projects like the Materials Project, led by MIT, or Google DeepMind's research into crystal structure prediction have demonstrated the potential of AI, but until now efforts have been fragmented and lacked a unified infrastructure. CuspAI seeks to integrate all stages of discovery into a single platform, from candidate generation to experimental validation.

Why is it important?

The discovery of new materials is a critical bottleneck for innovation in key sectors such as energy, electronics, construction and sustainability. For example, improving solid-state batteries for electric vehicles, creating more efficient semiconductors for quantum computing, or developing biodegradable materials that reduce plastic pollution requires materials that don't yet exist and whose properties must be discovered through trial and error. Generative AI applied to materials science has shown promising results in recent years: in 2023, DeepMind published a model in Nature capable of predicting the structures of more than 380,000 hypothetical materials, some of which could have applications in superconductivity or energy storage. However, experimental validation of those candidates remains a slow and costly process.

The AI Materials Foundry aims to create a shared infrastructure that democratizes access to these tools. According to statements by CuspAI's CEO reported by Reuters, the platform will be open to the scientific community and industry, though founding partners will have priority access to results and computing capacity. This model resembles OpenAI's strategy with ChatGPT, which initially offered free access to attract users and then launched paid versions, but focused on materials science. The key difference is that here most generated data will be public, which could accelerate global scientific progress.

Moreover, the initiative comes at a time when the demand for new materials is urgent. The energy transition requires batteries with higher energy density and lower cost, the electronics industry needs semiconductors that overcome silicon's limitations, and construction seeks concrete with a lower carbon footprint. AI could be the tool that breaks the stagnation in materials innovation observed in recent decades.

What consequences will it have?

If the AI Materials Foundry succeeds, it could reduce the development time for new materials from decades to months. This would have a direct impact on multiple industries: cheaper and more efficient batteries would accelerate the adoption of electric vehicles and renewable energy storage; faster and more efficient chips would boost high-performance computing and artificial intelligence; and biodegradable or recyclable materials would help mitigate the waste crisis. A report by consulting firm McKinsey estimates that AI applied to materials could generate up to $200 billion in annual economic value by 2030, mainly through reduced R&D costs and accelerated commercialization.

However, significant technical challenges remain. Property prediction using generative models is still imperfect: algorithms can propose structures that are theoretically stable but cannot be synthesized in practice, or whose real properties differ from simulations. Additionally, robotic synthesis is still limited: current robots can handle a limited number of chemical reactions and cannot replicate the versatility of a human chemist. Experimental validation will remain a bottleneck until automation advances sufficiently.

From a competitive standpoint, the consortium strengthens Nvidia's position in the AI hardware market, as it will provide the GPUs needed for simulations and model training. Meta and Samsung, for their part, seek to secure key materials for their future products, such as foldable displays, high-performance batteries or virtual reality components. Hyundai Motor, which is heavily investing in electric vehicles and fuel cells, hopes to obtain materials that improve battery efficiency. The Lawrence Berkeley National Laboratory, one of the most prestigious research centers in the US, will contribute its expertise in materials characterization and advanced synthesis.

The initiative could also have geopolitical impact. Currently, China dominates the supply chain for rare earths and other critical materials for technology. If the AI Materials Foundry manages to discover synthetic alternatives or more abundant substitutes, it could reduce dependence on Chinese imports and reshape the geopolitical map of strategic resources. This would explain the interest of governments and large corporations in supporting the project.

What should readers know?

• Timeline: The platform will be operational in phases. First results from the generation and simulation phase are expected by the end of 2026, while experimental validation with robots will begin in 2027. The first materials discovered through the platform are expected to reach the market by 2030.
• Access: There will be a free version for academic researchers and startups, with limited access to models and computing capacity. Companies will be able to purchase subscriptions with tiered access levels, from basic queries to premium packages that include priority experimental validation.
• Risks: Reliance on synthetic data generated by AI and lack of thorough experimental validation could lead to false positives, i.e., materials that look promising in simulation but don't work in reality. Additionally, there is a risk that models reproduce biases present in training data, such as overrepresentation of certain material types.
• Competition: Similar projects like MIT's Materials Project (which has a catalog of over 140,000 materials), Google DeepMind's initiative (GNoME, which predicted 380,000 stable materials) or the European Materials Modelling Council already exist, but CuspAI bets on a more integrated approach, combining generation, simulation and experimentation in a single platform, with much higher funding. Moreover, the participation of large corporations gives it privileged access to computational resources and real industrial needs.

“We want to do for materials what Google did for information: organize them and make them accessible,” CuspAI's CEO told Reuters. “But also, we want to create them.”

In summary, the AI Materials Foundry represents an ambitious step toward automating materials discovery, with the potential to transform entire industries. However, its success will depend on overcoming technical challenges and the coalition's ability to stay united and focused on common goals. The coming years will be crucial to determine whether this initiative becomes a milestone or just another failed experiment.