A startup called CuspAI has assembled a global network of over 45 organizations, including NVIDIA and Meta, to tackle one of the biggest constraints on industrial progress: the world needs materials that don't yet exist. The initiative, called the AI Materials Foundry, combines artificial intelligence with laboratory infrastructure and exclusive datasets to dramatically speed up the discovery of new compounds for semiconductors, clean energy, and advanced manufacturing.

Why Is Materials Discovery Becoming an AI Problem?

The challenge isn't engineering or design expertise. Companies and researchers understand how to build semiconductors, batteries, and manufacturing systems. The real bottleneck is materials. Engineers need compounds with specific properties that don't exist yet, and finding them through traditional trial-and-error methods takes years. CuspAI's approach uses machine learning to screen billions of candidate molecules at once, then routes the most promising ones to physical labs for validation.

The company has already demonstrated the potential. With Finnish chemicals company Kemira, CuspAI screened a search space of 300 trillion potential molecular structures and delivered 20 validated novel candidates in just six months, a process that previously took years.

How Does the AI Materials Foundry Actually Work?

At the heart of the network sits MIRA, CuspAI's proprietary AI platform. Here's how the system operates: a partner organization defines what they need, whether that's a compound with specific thermal stability, a semiconductor with a target bandgap, or a catalyst with a defined reaction profile. MIRA then generates candidate structures using generative models trained on the world's largest curated experimental materials datasets.

The platform doesn't stop at generating ideas. It runs property predictions across millions of candidates, selects the most promising ones, designs synthesis routes matched to available lab infrastructure, and routes the work to the right facility based on capability, geography, and throughput. As results come back from physical experiments, MIRA feeds them back into the model, continuously sharpening its predictions.

What Makes This Network Different From Solo Labs?

Rather than relying on the physical constraints of a single, isolated laboratory, the Foundry creates what CuspAI calls a "compounding intelligence loop." Breakthroughs achieved within the network have the potential to accelerate discovery timelines across the entire global value chain. The platform is also designed for industrial confidentiality, with partner data protected in private Foundry instances suitable for multinational and government operations.

The founding members represent a cross-section of industries that depend on new materials. They include semiconductor equipment makers like Applied Materials and Tokyo Electron, chemical companies like Henkel and Topsoe, automotive firms like Hyundai Motor Group, and research institutions like Cambridge University and the Technical University of Denmark.

Steps to Understand the Foundry's Technical Architecture

• Data Foundation: CuspAI has secured exclusive AI training rights to foundational experimental records including the Cambridge Structural Database and the Inorganic Crystal Structure Database, plus licensed access to materials science content from publishers like Wiley. This dataset compounds with every validated experimental result from Foundry programs.
• Simulation Layer: The Foundry runs on kUPS, an open-source molecular simulation toolkit built by CuspAI in collaboration with NVIDIA's ALCHEMI (AI Lab for Chemistry and Materials Innovation) team. It leverages Meta's Universal Model for Atoms (UMA), which enables fast, accurate simulation of atomic interactions across the periodic table.
• Compute Infrastructure: NVIDIA provides the accelerated computing infrastructure needed to screen at molecular resolution across billions of candidates, making it possible to evaluate possibilities that would be impossible to test physically.

The team behind CuspAI brings deep expertise to the problem. CTO and co-founder Professor Max Welling co-invented the variational autoencoder (VAE) and equivariant neural network architectures that now underpin generative molecular design. Chief Scientific Officer Professor Aron Walsh is one of the world's foremost computational materials scientists. John Giannandrea, who built and led AI research at Google before serving as Apple's SVP of Machine Learning and AI Strategy, is helping set up US foundry operations.

"If we don't make progress fast, the next 50 years of industrial progress will be constrained by a single challenge: the world needs materials that don't yet exist. That's what we're on a mission to solve, combining frontier agentic AI with deep domain expertise, exclusive data access and close customer partnerships," said Dr. Chad Edwards, CEO and Co-Founder of CuspAI.

Dr. Chad Edwards, CEO and Co-Founder, CuspAI

What's Already in Motion?

One Foundry project already underway is a multi-year partnership between CuspAI and A*STAR, Singapore's lead public sector research and development agency. The collaboration will combine AI-driven discovery with autonomous synthesis capability across semiconductors, carbon capture, and advanced electronics.

"As AI transforms the physical world, new materials will open up new frontiers across semiconductors, energy and advanced manufacturing. The AI Materials Foundry brings NVIDIA accelerated computing infrastructure together with world-class chemistry and materials expertise to help power the next generation of materials discovery," stated Ian Buck, Vice President of Hyperscale and HPC at NVIDIA.

Ian Buck, Vice President of Hyperscale and HPC, NVIDIA

The initiative also reflects a broader shift in how AI is being applied to scientific discovery. Rather than treating AI as a standalone tool, the Foundry treats it as the orchestrator of a complex ecosystem that includes data, compute, physical labs, and human expertise. This approach acknowledges that the bottleneck in materials discovery isn't any single component, but the integration of all four.

"We're proud to be longstanding partners with CuspAI and now as founding members of the AI Materials Foundry ecosystem. Our open source frontier models for materials science research will enable more teams to tackle previously intractable challenges, more precisely and more quickly than before," explained Rob Fergus, Vice President of AI Research and Head of FAIR at Meta.

Rob Fergus, Vice President of AI Research and Head of FAIR, Meta

For industries waiting on new materials, the Foundry represents a potential acceleration of timelines that have historically stretched across years. The combination of generative AI, simulation at scale, and coordinated experimental validation could reshape how quickly companies can move from identifying a materials need to having validated candidates ready for production development.