CuspAI: The AI Materials Discovery Search Engine

CuspAI's AI materials discovery platform designs new molecules up to 10x faster than traditional methods, backed by $130M and advisors Geoffrey Hinton and Yann LeCun.

Photo source:

Cusp AI

CuspAI: The AI Materials Discovery Platform Redesigning What's Possible


Nature had a head start of about four billion years. CuspAI is closing the gap fast.

Every major human challenge — cleaner energy, better compute, drinkable water, breathable air — hits the same wall eventually. Not funding. Not willpower. Materials. The right molecule either doesn't exist yet, or takes decades to find through conventional research. CuspAI was built to change that timeline entirely.


Founded in Cambridge in 2024, the company has built an AI materials discovery platform that works like a search engine for molecules that don't yet exist. A researcher specifies the exact properties they need — a compound that binds carbon efficiently, a semiconductor that runs cooler, a polymer that pulls toxins from water — and the platform generates new, synthesisable candidates up to 10 times faster than any traditional lab method. The molecule arrives before the lab bench is even set up.

How the AI Materials Discovery Platform Thinks Differently


Traditional chemistry has always moved in one direction. Synthesise, test, fail, repeat. A single discovery cycle can eat months. A breakthrough material can take a generation. CuspAI reverses that process entirely, starting with target properties and using AI to identify molecular candidates before anything physical is created, cutting costly trial and error at the source.


The engine underneath combines generative AI with physics-based molecular simulation. It doesn't mine existing databases for what's already known. It designs materials that have never existed, then validates them computationally before a single gram is synthesised. That shift — from test-and-fail to design-and-confirm — is how decades compress into months.


Prof. Max Welling, co-founder and CTO, described it at launch: imagine a search engine not just for existing materials, but for every molecule that could ever be created. That's not a product roadmap. That's a different relationship with chemistry itself.

The Problems Already on CuspAI's Workbench


Within its first year, CuspAI secured partnerships with Hyundai on sustainable energy materials, Kemira on removing PFAS — the so-called forever chemicals — from water supplies, and Meta on carbon capture, contributing to ODAC25, the world's largest direct air capture database.

Each of those partnerships starts with the same question: what material would solve this, if it existed? Previously that question took years to answer. Now it's a query.


Even Geoffrey Hinton — the man who quit Google to warn the world about AI's dangers — agreed to advise CuspAI. His statement was characteristically conflicted: humanity will face many challenges in the coming decade, he said, some caused by AI, others solved by it. For Hinton, materials discovery sits firmly in the second column.

Why the AI Materials Discovery Bet Is Getting Bigger


One year after coming out of stealth with a $30 million seed, CuspAI closed a $100 million-plus Series A in September 2025. The round was co-led by NEA and Temasek, with NVIDIA's NVentures, Samsung Ventures, and Hyundai Motor Group joining alongside it. The angel list reads like a reunion of the people who built the modern internet: the co-founder of OpenAI, the founder of Dropbox, the founder of Hugging Face, and Google DeepMind's VP of Research.


Two new board members arrived with the funding: Martin van den Brink, former President and CTO of ASML, and Lord John Browne, former CEO of BP and current Chair of the UK Government's Council for Science and Technology. Neither joins early-stage startups without conviction. Both did.

The problems CuspAI is chasing are each worth billions on their own. Better battery materials. Cheaper carbon capture. Semiconductors that don't depend on contested supply chains. The common thread through all of them is the same: a better molecule is the answer, and nobody has found it yet. CuspAI is building the machine that does.

Lock

You have exceeded your free limits for viewing our premium content

Please subscribe to have unlimited access to our innovations.