CuspAI accelerates materials discovery with generative AI

Anyone searching for a new material tends to think in decades. Millions of molecular combinations are possible, yet every laboratory experiment costs time and money. CuspAI, founded in March 2024 and operating from Amsterdam, Cambridge and Berlin, aims to raise that pace by moving materials research largely into software.

The company's platform functions as a search engine for the material world. A researcher specifies desired properties, such as conductivity, porosity or chemical stability, and the system generates candidate molecular structures that meet those requirements. Those structures are then evaluated through physics-based simulations, leaving the most promising candidates for potential laboratory work. Applications range from battery materials and hydrogen fuel cells to PFAS removal from water and materials for chip manufacturing.

The company now employs around 65 people and had raised a total of $130 million by June 2026, including a $100 million Series A in September 2025 co-led by New Enterprise Associates and Temasek, at a valuation of $520 million.

Two founders, two fields of expertise

CuspAI grew out of the collaboration between Dr. Chad Edwards and Prof. Max Welling. Edwards is a chemist and was previously a commercial co-founder of Cambridge Quantum Computing, later renamed Quantinuum, where he was responsible for strategic partnerships. Welling is a professor of machine learning at the University of Amsterdam, a co-inventor of variational autoencoders, and previously worked as a Distinguished Scientist at Microsoft Research AI4Science. In 2025 he was elected a member of the Royal Netherlands Academy of Arts and Sciences.

Welling serves as CTO and Edwards as CEO. The combination of deep ML expertise and commercial experience in the deeptech sector defines the company. CuspAI also attracts a broad advisory board, including Geoffrey Hinton, Nobel laureate and pioneer in deep learning, Yann LeCun of Meta, and Martin van den Brink, former president and CTO of ASML.

What sets the system apart from conventional database research

Traditional materials research relies on existing databases: scientists search catalogues of known structures for something close to their target. CuspAI generates new structures based on desired properties, without being limited to what has already been synthesised or published. The system combines generative AI models with molecular simulations grounded in physical principles.

A benchmark for this type of system is the so-called VUN percentage, which indicates what share of the generated structures are valid, unique and novel. For metal-organic frameworks, a class of porous materials relevant for applications such as gas storage and filtration, CuspAI's own model MOFGEN achieves a VUN score of 49 percent. By comparison, a comparable model from Microsoft reaches 10 percent and one from Meta 16 percent.

In April 2026, CuspAI released the open-source simulation toolkit kUPS. According to the company, that toolkit can accelerate certain research tasks by a factor of 49 compared with existing methods.

Customers and application areas

CuspAI deliberately targets multiple sectors simultaneously. Partners the company works with include Meta, for carbon capture via the OpenDAC database, Hyundai Motor Group, for materials aimed at battery efficiency and hydrogen fuel cells, and Kemira, a Finnish chemical supplier, for materials that remove PFAS from water. ASML and NVIDIA are also named as partners, although the details of those collaborations are not public.

Angel investors include names such as Durk Kingma, co-founder of OpenAI, Zoubin Ghahramani of Google DeepMind, and Thomas Wolf, founder of Hugging Face. Institutional investors include NVentures, NVIDIA's investment arm, Samsung Ventures and Prosus Ventures. The fact that CuspAI attracts investors from both the semiconductor sector and the automotive industry is linked to the breadth of applications the platform is intended to support.

Amsterdam roots in an international company

CuspAI is legally registered in Cambridge but has a research team in Amsterdam at Amsterdam Science Park, in the University of Amsterdam's Lab42 building. There are also teams in Berlin and Tokyo, and in 2026 the company is opening an office in London King's Cross.

According to reports from June 2026, CuspAI is preparing a new funding round of $400 million, co-led by Bezos Expeditions and Kleiner Perkins. That round would value the company at $2.6 billion. Formal confirmation from the company has yet to be provided.

For the Dutch and European deeptech scene, CuspAI is an example of how academic ML research, in this case through the University of Amsterdam, can give rise to a commercial company that attracts international investors and industrial partners. The fact that Amsterdam Science Park serves as a base alongside Cambridge and Berlin shows that the city is a competitive location for this type of knowledge-intensive activity, even when the registered headquarters lies elsewhere.