From Algorithms to Atoms: Our Investment in CuspAI

It’s often said that the next decade is the age of atoms rather than bits. We believe advances in the latter will unlock breakthroughs in the former.

Looking at the evolution of intelligent systems, we can identify three distinct eras:

1. First came the era of systems built on formal (mathematical) models and simulations, generating synthetic data and reasoning within well-defined, logic-driven, and largely deterministic representations of the world. Manual experiments by scientists persisted in this first era and were essential for validating and calibrating the models and simulations.

2. Next was the era of systems that learned directly from large-scale experimental data, using statistical and probabilistic methods to capture patterns and make predictions from observed reality.

3. The emerging era will blend these paradigms into agentic, closed-loop systems that can define goals, design and run simulations, select viable paths, commission physical experiments, interpret results, and adapt their strategies iteratively without human micromanagement. By tightly coupling in-silico design with real-world validation in rapid feedback cycles, these systems will accelerate computational discovery and extend intelligent problem-solving into complex domains of the physical world.

CuspAI is spearheading this emerging era in computational materials science, where novel materials can be generated, synthesized, tested and validated in months instead of the 10-20 year horizon the industry has learned to expect. Based in Cambridge, UK with teams across Amsterdam and Berlin, CuspAI has demonstrated exceptional vision and execution: building state-of-the-art models, partnering with industry leaders across different domains, and gathering a stellar team with more than 2 million citations collectively. The company’s innovative approach to computational materials science aligns perfectly with our investment philosophy in backing exceptional talent with a pragmatic approach to solving world-changing problems in high-impact industries. And that is why we are thrilled to have led their Series A financing round.

Why Materials Science?

Materials underpin nearly everything: the homes and infrastructure we build; energy generation, storage, and transmission; mobility and aerospace; computing, communications, and sensing; clean water and food systems; health care and medical devices; textiles and packaging; and national security. Advancements here ripple across the economy.

Historically, discovering a new material is slow and expensive – often a decade or more and tens to hundreds of millions of dollars from idea to deployment[1]. 

CuspAI’s platform uses inverse design – starting with target properties and working backward to propose candidates – then evaluates stability, performance, and manufacturability through fast feedback loops. In practice, that means high-fidelity simulations, learned surrogate models, degradation pathway modeling, and constraint-aware generation informed by experimental data.

The acceleration of materials discovery enables:

• Addressing emerging challenges. e.g., filtration of PFAS (“forever chemicals”) from drinking water and industrial discharge.

• Tackling persistent bottlenecks. e.g., safer solid-state electrolytes, longer-cycle batteries, low-loss power electronics, corrosion-resistant coatings, high-performance membranes for desalination and gas separation. 

• Anticipate future demand. e.g., lightweight, high-temperature alloys for aerospace; rare-earth-lean magnets; thermal interface materials for data centers; recyclable or bio-derived polymers for packaging and apparel.

Why CuspAI?

We believe CuspAI has amassed a set of unique resources and strategies that are unparalleled in this space: 

• Dr. Chad Edwards (Co-founder & CEO) was previously the Commercial Co-Founder of Cambridge Quantum Computing (CQC). He later served as VP of Strategic Partnerships and Global Head of Strategy at Quantinuum following CQC's merger with Honeywell.

• Professor Max Welling (Co-founder & CTO) is a Professor at University of Amsterdam, and previously VP Technology at Qualcomm AI Research and Distinguished Scientist at Microsoft Research. He is considered a pioneer in AI’s application to science, variational inference, probabilistic deep learning, and geometric deep learning. 

Focus on large scale, curated data collection: CuspAI recognizes that high-quality, large-scale data is foundational to building state-of-the-art models. The team has made early and deliberate investments in building proprietary datasets at scale, including MOFs, to enable models that are both high-performing and generalizable across material classes. In addition, CuspAI runs tight integrations with downstream experimental data pipelines for simulation, synthesis, and testing workflows. This is also complemented by academic and scientific literature through licensing agreements.

Partnering with industry leaders across various domains: CuspAI partners directly with commercially successful businesses and industry leaders to drive impact at scale - aligning closely with partners’ priorities, and building deep collaborations across sectors like energy, climate, automotive, and semiconductors. In addition, CuspAI has assembled a distinguished advisory board that includes Nobel laureate Geoff Hinton (Turing Award laureate, deep learning pioneer), Yann LeCun (Turing Award laureate, Chief AI Scientist at Meta), Lord John Browne (former CEO of BP), Martin van den Brink (former President & CTO of ASML), Verity Harding (former Global Head of Policy at DeepMind), and Prof. Kristin Person (a leading figure in materials science).

Achieving SOTA model performance: CuspAI’s core model stack is fully proprietary, designed to cover end-to-end materials discovery lifecycle from micro-scale design (molecular and atomic levels) to macro-level deployment (process and manufacturability). The CuspAI platform includes a suite of generative models like MOFGEN, a state-of-the-art autoregressive transformer for metal-organic frameworks (MOFs) that achieves a VUN (valid, unique, novel) rate of 49%, which outperforms by a large margin models from Microsoft (10%) and Meta (16%)[2]. Unlike simpler inorganic generators, MOFGEN produces highly complex, synthesizable structures validated against strict physical and chemical constraints and tested against experimental data generated from industry partners.

The Future of Materials

We believe that CuspAI will play a crucial role in shaping the future of materials discovery for generations to come, and will touch many aspects of our physical world from the chips powering our machines to ensuring the sustainability of our environment. 

With our investment, CuspAI will be able to accelerate its research and development efforts, expand its market reach, and further solidify its position as a leader in the domain. We are thrilled to partner with Chad, Max, and the entire CuspAI team. Their vision and ambition have the potential to reshape the world, and we can’t wait to be part of that journey.