Kemira and CuspAI have applied generative AI end-to-end to the design of new materials targeting the removal of PFAS from water at trace concentrations — becoming the first commercial partnership to do so.

The project explored a design space of approximately 300 trillion possible material structures and delivered more than 5,000 novel material designs with property data for three priority PFAS molecules: GenX, PFBS and PFOS.

These were narrowed to around 20 selected priority candidates, now advancing to the next phase of development and testing.

This process used to take years — our discovery project took six months.

Kemira set us a demanding industrial brief: materials needed to target PFAS at sub-parts-per-billion concentrations, using chemistry that is water-stable, environmentally compatible, synthesisable and cost-effective.

We’re proud to be making progress on an AI application as vital and urgent as clean water.

Read the announcement in full.

Technical blog

Molecular simulations are the backbone of drug discovery and materials science, but they’re notoriously complex.

For example, the pharmaceutical industry spends 15% of its revenue on R&D — more than 20% of the total R&D spend across all other sectors combined — and molecular dynamics is one of the heaviest computational line items in that budget.

Modelling different phenomena demands fundamentally different techniques. Characterizing a single material, for example, can involve chaining several computational methods together, and the number of possible combinations grows quickly as new techniques emerge. For scientists and machine learning researchers, this can create a laborious, clunky workflow: combining several complex software packages, spending hours debugging inconsistencies, and potentially using hundreds of lines of configuration and glue code.

Today, we are excited to open-source kUPS, a molecular simulation engine for the AI era.

kUPS is a toolkit that unifies diverse molecular simulation techniques behind a single composable interface. It gives computational scientists and ML researchers a shared, Python-native platform to run a broad range of simulations that are GPU-native and highly efficient, scaling seamlessly from a single laptop to data center GPUs.

In practice, using kUPS, a researcher can write a custom simulation for a novel use case in an afternoon, rather than taking weeks of effort.

Simulation tools weren’t built for the AI era

Discovering materials requires combining hundreds of thousands of accurate molecular simulations with machine learning to select the most promising candidates for laboratory synthesis. High-throughput computational screening means exploring vast chemical spaces across thousands of simulation runs.

However, a one-size-fits-all approach fails when dealing with diverse phenomena. Researchers use different techniques including molecular dynamics for transport properties like diffusion, Monte Carlo sampling for adsorption equilibria and to predict gas uptake, and geometry optimization for predicting stable structures. Real-world applications rarely call for just one of these in isolation.

This means researchers juggle separate software packages for each method (LAMMPS, GROMACS, RASPA, and others), each with its own input formats, data structures, and assumptions. This fragmentation is costly. Experimenting with a new method typically means writing complex C++ code from scratch for every framework.

The rise of AI has made things harder still. Modern methods like machine-learning force fields, Boltzmann generators, and other learned components need to be deeply embedded in simulations for optimal performance, but traditional simulation packages were built to optimize well-known algorithms for CPUs, rather than GPUs. Their support for GPU-accelerated machine learning is bolted on through rigid interfaces that cannot keep pace with rapid developments in the field. Integrating a PyTorch model into a C++ engine means implementing custom bridges, cross-language memory management, and sacrificing the flexibility that made the model easy to develop in the first place.

Meanwhile, the tools powering AI development (Python, automatic differentiation, GPU-native computation) remain foreign to traditional simulation codes. We built kUPS to close that gap.

A modern foundation for molecular simulations

kUPS is built on three design principles: composability, structured data, and batched execution. Every operation flows through JAX, so simulations run natively on CPU, GPU, and TPU with no code changes.

Working in collaboration with NVIDIA, we have optimized for GPU performance in particular, achieving up to 49× throughput over widely adopted tools like RASPA for certain simulations. And, because many models today live in PyTorch, we also open-source Tojax, a library for transforming PyTorch functions into JAX, meeting developers where they are.

This work is in addition to our ongoing collaboration with NVIDIA on ALCHEMI, their platform for accelerating AI-driven simulation in chemistry and materials science.

“We’ve been working closely with the CuspAI team to help push the boundaries of GPU-accelerated molecular simulation. The ALCHEMI collaboration with CuspAI demonstrates what’s possible when modern simulation software is built natively for accelerated computing. This approach accelerates scientific discovery in unprecedented ways, unlocking the kind of throughput gains that let researchers move from thousands of simulations to millions, and thus from hypothesis to discovery, faster than ever before.”

— Ian Buck, Vice President of Hyperscale and HPC Computing, NVIDIA

Composability

To fully characterize a material, scientists often need to link multiple techniques into a single workflow. In traditional tools, each combination means writing new glue code from scratch. kUPS is designed so that any method can be combined with any other by construction.

Every operation follows a single abstraction: propagate a state to its next state. The next state depends only on the current state and a random key. This has practical consequences. Simulations are bitwise reproducible, because replaying the same sequence of keys from a given state always yields identical trajectories. Checkpointing is trivial, since the state is a complete snapshot with no hidden history, and any propagator can be tested in isolation by feeding it a single state.

For scientists, this means fewer hours spent debugging subtle nondeterminism, and more confidence that results will hold up when shared, reproduced, or scaled.

The shared interface is what makes it all composable. In kUPS, an integrator, a thermostat, a Monte Carlo move, a force field evaluation are all propagators. Because they share the same signature, they snap together freely. A new potential works with every existing sampler, and a new sampling method works with every existing potential.

This extends naturally to learned components: a machine-learning force field, a generative model serving as the proposal distribution in a Markov chain Monte Carlo (MCMC) simulation, or a learned large-timestep integrator all plug in through the same abstraction. This could reduce the timeframe for wiring a custom simulation for a novel use case from weeks, to a single afternoon.

Data structure

Molecular simulations involve data at different scales: per-atom positions and forces, per-molecule charges and topologies, per-system temperatures and pressures. Organizing such data in memory and ensuring their efficient use is a challenge for GPUs. When thousands of threads need to read positions simultaneously, scattering that data across separate objects means scattered memory access and idle hardware.

kUPS stores simulation data in Tables, columnar containers where every property shares the same leading dimension. Each row carries a unique key, analogous to a primary key in a SQL database, and tables reference each other through index arrays that act as foreign keys. This gives us structure-of-arrays layout (one contiguous array per property) rather than arrays-of-structures (one object per atom). With many checks and strict type checking, this significantly reduces implementation errors and the manual work of remembering data layouts when implementing new simulations. On GPUs, where performance depends on coalesced memory access across thousands of threads, this layout is the difference between saturating the hardware and wasting it.

Performance

Traditional simulation tools process one system at a time, leaving most of a GPU’s capacity idle. kUPS is built around batching: running many independent systems in parallel as a single vectorized computation. The columnar table layout makes this natural: stacking systems along the leading dimension turns a batch of simulations into a single dense array operation. This is the backbone of our high-throughput workflows and where the performance gains come from.

As an example, we benchmark two representative workloads on a single NVIDIA L4 GPU, using metal-organic frameworks — porous materials with downstream uses including carbon capture — as a test case. We are actively expanding our suite of benchmarks to include further examples across diverse material classes.

First, we explore NVT molecular dynamics of a single 114-atom metal–organic framework (MOF) using the MACE neural network potential. Even without batching, kUPS achieves 1.8× the throughput of ASE, the standard Python interface for running MACE, purely from lower overhead in the simulation loop. Batching multiple systems would increase performance further by saturating the GPU across independent runs.

Second, we look at grand-canonical Monte Carlo (µVT) for CO₂ adsorption in a 3×3×3 supercell of the same framework (3,078 framework atoms, ~30 CO₂ molecules at average loading). RASPA, the most widely used tool for this task, runs on a single CPU core. kUPS replaces it with a fully vectorized GPU implementation, achieving 49× throughput on the same workload. Using kUPS, an entire class of simulations that was architecturally locked to serial CPU execution can now run natively on accelerators.

Putting it together

In a traditional workflow, setting up a molecular dynamics simulation with a machine-learning potential might involve hundreds of lines of configuration spread across multiple files and languages, plus careful glue code to connect the model to the simulation engine. With kUPS, the same simulation takes a few lines of Python. Users can define a data layout, wire up a potential and integrator, and run:

class MyAtoms:
    positions: Array
    atomic_numbers: Array
    ...

class MySystems:
    unitcell: UnitCell
    ...

class MyFirstState:
    atoms: Table[ParticleId, MyAtoms]
    systems: Table[SystemId, MySystems]
    mace: MACEModel
    ...

atoms = MyAtoms(...) # populate atoms
systems = MySystem(...) # populate systems
state = MyFirstState(atoms, systems, ...) # populate state
chain = key_chain(jax.random.key(0)) # random key chain

potential = make_tojaxed_potential_from_state(
    identity_lens(MyFirstState), # lens: tells the potential where to find its inputs in the state
    compute_position_and_unitcell_gradients=True
)

propagator = make_md_propagator(state_lens, "verlet", potential)

for _ in range(1000):
    state = propagator(next(chain), state)

Swapping MACE for UMA is a single line change, adding a Monte Carlo move alongside the dynamics — or inserting a learned large-timestep model between integration steps — takes just a few lines more. With traditional tools, changes like these would typically mean switching to an entirely different package and rewriting the simulation from scratch.

Methods and potentials

kUPS ships with a broad set of methods and potentials covering the most common molecular simulation tasks, all built using the same composable abstractions available to any user.

On the sampling side, this includes molecular dynamics ensembles (NVE, NVT, NPT), Monte Carlo methods (NVT, µVT), and geometry optimizers (FIRE, L-BFGS). For interactions, kUPS provides classical potentials such as Lennard-Jones, Coulomb (via Ewald summation), harmonic bonds and angles, and Morse, alongside ML force fields including MACE and UMA. Every built-in method was written with the same public API you would use to build your own, and anything you add will work with every existing component from day one.

Bridging the gap: Tojax

Rather than asking users to rewrite their PyTorch models, we additionally open-source Tojax, a library that converts any PyTorch function into a JAX function with a single line of code.

model = SimpleModel() # any PyTorch nn.Module
jax_model = tojax(model) # → callable in JAX

x = jnp.ones((128, 32, 10))
output = jax.jit(jax.vmap(jax_model))(x) # JAX transformation compatible

A pre-trained MACE, NequIP, or custom PyTorch potential can be brought into kUPS with minimal code changes. With Tojax, the model composes with jit, vmap, grad, and all other JAX transformations, the same ones kUPS relies on for batching and automatic differentiation.

There is no performance penalty: Tojax traces through the PyTorch module and emits a native JAX computation graph, compiled and optimized end-to-end by XLA with no Python-level back-and-forth between frameworks at runtime.

Get started

kUPS and Tojax are open-source. We built them because we needed a simulation toolkit that could keep up with the pace of AI research: one where trying a new idea means writing a few lines of Python, not weeks of C++. One where simulations run natively on GPUs without compromise, and where classical methods, machine-learning potentials, and generative models all live within a single composable framework.

Whether scientists and machine learning researchers are screening millions of materials for carbon capture, semiconductors or catalysts, exploring drug-protein interactions, or developing entirely new simulation methods, we hope kUPS will give them the foundation to move faster.

github.com/cusp-ai-oss/kups
github.com/cusp-ai-oss/tojax

Thanks to: Moubarak, Elias; Morrow, Joe; de Haan, Pim; Openshaw, Hannah; Tait, Rhiannon; Valdez, Jessica; Barnett, Ruth; Welling, Max; CuspAI Team

The problem is the size of the space

The central challenge in materials discovery is the scale of the space it operates in. Even when restricted to chemically plausible compounds, conservative estimates place relevant chemical space well beyond 1⁰⁶⁰ candidates. If each possible material were a point, chemical space would be like a universe. Even scanning a vanishingly small fraction of it would take longer than the lifetime of any lab.

Progress depends on learning where not to look, and how to move efficiently toward the regions that matter.

From the outset, our goal at CuspAI has been to build a system that can navigate this space continuously and coherently. Not as a one-off optimisation, but as a persistent learning system that can be deployed, integrated, and run over time.

Over the past year, we completed the first full iteration of that system and began operating it end to end with partners. Targets are specified in terms of function and constraint. Candidate materials are generated and evaluated using physics-based simulation. Experiments are selected based on their expected contribution to reducing uncertainty. Results are fed back into the system and influence future decisions.

From research to deployment

Materials discovery sits at the intersection of simulation and experiment. Large parts of the search happen in silico, where simulations are used to explore design spaces and rule out implausible candidates early. The challenge is ensuring that these simulations remain tightly aligned with experimental reality, so that each design–build–test cycle meaningfully updates the system’s internal models.

This has shaped how the system behaves in deployment. Simulations are used aggressively, but always in service of deciding which experiments to run and why. Experimental results are used to recalibrate models, refine decision policies, and improve how uncertainty is handled across the system.

Over time, this leads to more selective searches and faster convergence on materials that survive real constraints. The system improves because the decision logic governing the loop becomes better informed with continued use, with several orders of magnitude speed ups being accomplished already.

We demonstrated this behaviour through SkyVault, where carbon capture materials moved from generative design through synthesis and experimental validation in six months. More important than the timeline was what it showed about the system’s ability to operate continuously and improve with experience.

The impact is now

In 2025, we deployed the system across several industrial environments where materials performance sets hard limits on progress. These deployments are long-lived by design: the system is embedded within partner workflows and continues to learn as programmes evolve.

In mobility, we started work with Hyundai Motor Group on next-generation materials for future vehicle platforms, where durability, manufacturability, and lifetime are tightly coupled. In water treatment, we deployed with Kemira to develop materials capable of selectively removing PFAS under real operating conditions. In catalysis, we began deployments with Topsoe, where incremental improvements in materials efficiency translate directly into large-scale industrial and energy-system impacts.

We also deepened our deployment with Meta, including joint work with the Georgia Institute of Technology, to expand the space of candidate materials for carbon capture and to release datasets that support broader research in the field.

Alongside these efforts, we significantly expanded deployments in semiconductors and advanced compute. As device scaling slows, progress increasingly depends on materials choices across logic, memory, interconnects, packaging, power delivery, and thermal management. These problems are characterised by narrow feasibility windows and strong coupling between physical performance and manufacturing constraints.

In this setting, systems that accumulate knowledge about which materials are both performant and compatible with real processes develop a structural advantage. Learning carries forward across programmes, rather than resetting each time. We’ll share more of this work in 2026.

We were also grateful for public recognition from Jensen Huang, who mentioned CuspAI as one of the AI companies he’s excited about when speaking alongside UK Prime Minister Keir Starmer.

Scaling

In September, we closed a $100M+ Series A, led by Temasek and NEA with strong backing from NVIDIA, alongside our Seed investors and strategic angels. Their support reflects a shared belief that deployed learning systems for the physical world will define the next phase of AI.

Our team has grown across Cambridge, Amsterdam, Berlin, and Tokyo, and we welcomed Professor Aron Walsh as Chief Scientific Officer. We were also joined by advisors including Lord John Browne, former CEO of BP, and Martin van den Brink, former CTO and President of ASML, alongside existing advisors Geoffrey Hinton, Yann LeCun, Kristin Persson, and Verity Harding.

In January, we will open our London Kings Cross office.

20x in 21 Months

Both the size of the CuspAI team, and the value of our contracts, have grown twentyfold in just 21 months. Our team is operating at a pace that would typically require a company several times larger. That is a direct consequence of being AI-native from the start, building learning and decision systems into the core of how work gets done.

As those systems continue to learn in deployment, the pace only further accelerates.

In 2026, we’ll continue to extend the reach of this materials intelligence layer across domains and geographies. We’ll also share more of our work in semiconductors, where the interaction between deployed learning systems and manufacturing reality is becoming increasingly central.

Our focus for the year ahead remains unchanged. Deliver.

— Dr Chad Edwards & Prof Max Welling, Co-founders, CuspAI

Here’s a summary of CuspAI’s 2025 milestones:

Partnerships:

• At the AI in Science Summit, CuspAI announced a collaboration with Topsoe and the Technical University of Denmark for creating more efficient and sustainable catalysts.
• CuspAI announced a strategic partnership with Hyundai Motor Group, working to enhance the efficiency, durability, and stability of next-generation materials for future smart mobility solutions.
• CuspAI partnered with Kemira to target the removal of PFAS (forever chemicals) from water.
• As part of our carbon capture partnership with Meta, we collaborated with them and the Georgia Institute of Technology on a research paper and dataset to increase the number of atomic structures for potential carbon capture materials by orders of magnitude.

Global footprint and talent:

• We hired preeminent materials scientist and Imperial College London Professor Aron Walsh as our Chief Scientific Officer.
• We’re opening our first site in London.
• This adds to our global footprint including Amsterdam, Cambridge, Tokyo and Berlin.
• We’ve welcomed new leaders onto our advisory board:

1. Lord John Browne, former BP CEO
2. Martin van den Brink, ex-CTO and president of ASML

• They join existing members, the Nobel Laureate Geoffrey Hinton; Meta’s outgoing Chief AI Scientist Yann LeCun; Berkeley Distinguished Professor Kristin Persson; and former Google DeepMind ethics lead Verity Harding
• We continue to hire — you can see our open roles here and follow updates from the team on our LinkedIn page.

Capital:

• We announced a $100 million+ Series A funding round, co-led by US fund New Enterprise Associates (NEA) and Temasek, with participation from NVentures (NVIDIA’s venture capital arm), Samsung Ventures, Hyundai Motor Group, and returning investors.
• This follows our $30 million Seed round in 2024 led by Hoxton Ventures, with significant participation from Basis Set Ventures and Lightspeed Venture Partners.

Recognition and Awards:

• CuspAI was one of the frontier AI companies named in the UK government’s recently published AI for Science strategy and we’re among the first to get significant access to the Isambard AI supercomputer.
• The World Economic Forum (WEF) selected CuspAI for its Technology Pioneers program
• TechCrunch Disruptors60 list
• Sifted ranked CuspAI No 1 on their AI 100 list of rising European AI stars
• Founders Forum named us an FF Rising Star
• We won Business of the Year at the Business Weekly awards in Cambridge
• Norrsken included CuspAI in their annual Impact/100, projecting our logo in New York’s Times Square
• Accel featured CuspAI on their list of top 100 AI and Cloud companies in its Globalscape report

Media highlights:

• Fortune’s Jeremy Kahn wrote about the incredible AI/ML talent joining CuspAI
• Chad was interviewed by Robert Peston and Steph McGovern on The Rest is Money podcast
• Chad was interviewed on the Times Tech podcast
• Our fundraise was covered by outlets including Fortune, Bloomberg TV, Axios Pro Rata and more
• Chad was interviewed by Maddyness about co-founding CuspAI
• The Economist wrote about Cusp in their deep dive on AI for materials science
• Handelsblatt published an OpEd by Chief Strategy Officer Markus Hoffman on the physical AI opportunity
• Max was interviewed by De (Financiële) Telegraaf
• UK AI minister Kanishka Narayan called out Cusp’s “world-changing work” when he launched the AI for science strategy, and UK science minister Sir Patrick Vallance gave CuspAI a shout out on The Rest is Money
• Check out our CuspAI Newsroom page.

2026

March

• CuspAI featured by Maeil Business (South Korea) as an example of Europe’s leadership in artificial intelligence

February

• CuspAI featured in the Wall Street Journal in a feature on the experienced founders behind billion-dollar startups
• Co-founder Max Welling featured on the Latent Space podcast

January

• CuspAI recognised in The Sunday Times as an up-and-coming AI company

2025

December

• CuspAI hires materials scientist and Imperial College London Professor Aron Walsh as our Chief Scientific Officer, featured in Sifted
• CuspAI announces new 2026 London base
• Feature: CuspAI on The Rest is Money podcast

November

• Hyundai Motor Group and CuspAI Partner to Accelerate Material Innovation Using AI. See also: Just Auto, Automotive World
• CuspAI featured in Accel’s list of top 100 AI and Cloud companies in its Globalscape report

October

• CuspAI listed in the Greenfield and TechCrunch AI Disruptors 60 list
• CuspAI listed as Europe’s standout AI player by Sifted

September

• Securing our $100M+ Series A to Revolutionise Materials Discovery with AI; see also: Fortune, Bloomberg TV, Axios Pro Rata
• Kemira and CuspAI Forge Strategic Partnership to Pioneer AI-Driven Materials Innovation; see also: Forbes, Business Weekly, Chemical Today
• Martin van den Brink (ex-CTO and president of ASML) and Lord John Browne (former BP CEO) join the CuspAI advisory board
• CuspAI listed as Business of the Year in Business Weekly awards in Cambridge
• CuspAI included in Norrsken’s annual Impact/100

August

• CuspAI publishes Open DAC 2025 with Meta and Georgia Tech

June

• CuspAI listed as a World Economic Forum Technology Pioneer
• CuspAI profiled as one of tech’s rising star at Founders Forum

March

• Prof. Yann LeCun, Prof. Kristin Persson and Verity Harding join Nobel Laureate Prof. Geoffrey Hinton on the CuspAI advisory board
• Feature: Economist — AI models are dreaming up the materials of the future

2024

June

• CuspAI Secures $30M to Combat Climate Change with AI-Designed Materials, partners with Meta

Clean air and water, next-generation AI compute, and sustainable energy — these multi-billion-dollar global challenges share a common barrier: materials. CuspAI’s mission is to change that. In just a year, we’ve gone from a concept, to active partnerships with world leaders in automotive, semiconductors, energy, and climate.

Today, we reached another significant milestone, with the closure of our $100 million+ Series A funding round. The round was co-led by US fund New Enterprise Associates (NEA) and Temasek, with participation from NVentures (NVIDIA’s venture capital arm), Samsung Ventures, Hyundai Motor Group, and returning investors.

Other firms participating include Basis Set Ventures, FJ Labs, Giant Ventures, LocalGlobe, Northzone, Prosus Ventures, Tiferes Ventures and Touring Capital.

Angel investors include Durk Kingma (OpenAI Co-Founder), Zoubin Ghahramani (Google VP Research), Arash Ferdowsi (Founder of Dropbox), Thomas Wolf (Founder of Hugging Face) and Victor Riparbelli (Founder and CEO of Synthesia).

Lila Tretikov, Partner and Head of AI Strategy at NEA and former Deputy CTO at Microsoft, who will join the CuspAI board, said:

“AI’s most profound promise emerges when it moves beyond the everyday and becomes a catalyst for discovery, deepening our command of chemistry, biology, and physics, and accelerating the design of materials that redefine what is possible. CuspAI harnessed this power to shutter long-accepted engineering limits, enabling breakthroughs across industries that can elevate the quality of life for generations to come in months rather than decades.”

Our platform acts like an AI ‘search engine’ for materials, enabling customers to specify the exact properties they need and generating new, synthesisable candidates up to 10x faster than traditional discovery methods. Our technology is materials agnostic, with potential for impact across multiple industries where breakthroughs can unlock billions in value.

This funding allows us to scale our platform and deliver transformative materials to market faster. We’re proud to have forged multi-sector partnerships in just one year spanning automotive, semiconductors, water purification and climate tech, including:

• Hyundai: A previously unannounced collaboration to work on sustainable energy applications
• Kemira: the Helsinki NASDAQ-listed public chemicals company, to work on PFAS removal
• Meta: to work on carbon capture — with a collaboration on the world’s largest direct air capture (DAC) database, ODAC25
• Additional partnerships still yet to be announced, including in the semiconductor sector

As part of CuspAI’s rapidly expanding commercialisation efforts, we’re also welcoming two globally recognised business leaders in semiconductors and energy to our board:

• Martin van den Brink, former President and CTO of ASML, a key architect of Europe’s semiconductor leadership
• Lord Browne, former BP CEO, Chair of the Francis Crick Institute, and current Chair of the UK Government’s Council for Science and Technology

They join our existing advisors Prof. Geoffrey Hinton and Prof. Yann LeCun — Turing Award-winning “godfathers of AI” — alongside Prof. Kristin Persson and Verity Harding.

The closure of our Series A is a moment for us to reflect as founders on how far we’ve come in a remarkably short timeframe. This financing comes just one year after we came out of stealth with a $30M Seed round.

We’re excited about this milestone, and grateful for the support of our investors and the strategic backing from our customers. With eyes on the future, we know that success is built on more than capital — and now the hard work continues.

We can only make real-world impact by continuing to interate our platform, collaborating on meaningful use cases with industry, and delivering next-gen materials to market faster. We’ll also be scaling our operations in the US and Asia to serve increasing customer demand.

When we started Cusp just a year ago, we knew building both a frontier AI company and a new materials platform from scratch wouldn’t be easy. This round is a reflection of what our team has achieved in record time, but also a reminder of the work ahead. The next breakthroughs in climate technology, batteries, semiconductors, and water purification will be achieved through iteration, partnerships, and a team that believes the future can be better. That’s what drives us.

— Dr Chad Edwards & Prof Max Welling, Co-founders, CuspAI

Read what our investors and customers have to say…

A Samsung Ventures spokesperson said:

“CuspAI is developing a foundational platform at the critical intersection of artificial intelligence and materials science. We believe their approach has the potential to unlock significant innovation across a range of industries, and we are pleased to support them as they scale their vision.”

Brooke Seawell, NEA Venture Partner, said:

“AI isn’t just promising a revolution in materials science, it’s already delivering one. By decoding the complexity of our material world, it’s starting to unlock breakthroughs from climate mitigation to faster, more efficient compute. With their unique approach and bold vision, the CuspAI team is well-positioned in this transformative space and I’m excited to be a part of it.”

Keith Noh, Vice President Head of ZERO1NE Group at Hyundai Motor Group, said:

“Hyundai Motor Group is advancing human progress through innovation and this requires embracing convictions that challenge the status quo. With CuspAI as our partner, we will leverage the power of AI to advance our performance and sustainability goals.

“We believe novel materials offer an extraordinary opportunity to accelerate progress towards a better, cleaner, more efficient future.”

Antti Salminen, CEO of Kemira, said:

“We’re proud to be a commercial partner of CuspAI — we chose them because we see the clear potential for AI to speed up the precision and impact of new materials discovery and they have fast emerged as the leaders in this approach.

“Our first collaboration together is focused on removing PFAS, so-called ‘forever chemicals’, from water. This partnership allows us to harness the power of AI to drive materials innovation, improve efficiency and deliver superior and sustainable solutions to our own customers.”

The world’s toughest challenges — from reducing carbon in the air we breathe, to developing next generation batteries and semiconductors — are materials science problems at their core. We spotted a massive opportunity: to upend the slow, expensive, hit-and-miss materials discovery process with the power of AI, so we could tackle some of society’s biggest challenges head-on.

Take pollutants in water, including PFAS: so-called “forever chemicals” which linger in our bodies and our environment without breaking down easily. Linked to health risks and under increasing regulatory scrutiny, PFAS and other micropollutants need to be removed from our water as a priority.

That’s why we’re thrilled that Kemira, a global leader in sustainable chemical solutions for water-intensive industries, has today officially announced its strategic partnership with CuspAI to revolutionise materials innovation within the chemical sector. Our first project with them will harness CuspAI’s AI-powered “search engine” for materials to target PFAS removal — though there are endless other possibilities for the future.

Acknowledging that the materials discovery process can be accelerated to as little as six months with AI, Kemira will integrate CuspAI’s AI-driven insights to fast-track the discovery and optimisation of new materials. Kemira will also work alongside Cusp’s team of AI, chemistry and engineering pioneers to enhance its R&D processes — developing digital, in silico molecules for its sustainable solutions.

“The AI materials development process is fast and precise, and we’re thrilled to be working with the leader in this area.” said Antti Salminen, CEO of Kemira. “The biggest global challenges, such as clean water, are fundamentally materials science problems and, as such, we see materials science as the next frontier for AI.”

The Cusp team is really excited to collaborate with such an established leader in sustainability and innovation, and to play a key role in their digital transformation journey. As well as co-creating sustainable solutions together, we’ll help them build a robust pipeline of AI expertise going forward.

“Our collaboration with CuspAI will provide our teams with invaluable exposure to cutting-edge AI technologies,” added Salminen. “This will not only enhance our innovation capabilities but also support the development of our AI talent pipeline, ensuring we remain at the forefront of industry advancements.”

There’s so much opportunity in the chemicals sector, but it’s not the only domain on the verge of disruption. Our materials discovery platform is materials agnostic, and we’re aiming for a ripple effect across industries, where even small breakthroughs can unlock billions in value.

We founded Cusp with a shared belief: that the most powerful way to build impact at scale would be through commercial partnerships, and close industry collaboration. Working alongside titans in their field, we have a shot at creating a safer, more sustainable path for humanity.

— Dr Chad Edwards & Prof Max Welling, Co-founders, CuspAI

Prof. Yann LeCun, Prof. Kristin Persson and Verity Harding join Nobel Laureate Prof. Geoffrey Hinton as advisors as we build our AI-powered search engine for discovering breakthrough materials.

When we started CuspAI last year, we knew we were tackling a challenge that could fundamentally transform how humanity addresses its most pressing problems. The stark reality is that the materials we need to solve such challenges simply don’t exist. From making carbon capture economically viable to creating next-generation computing materials, breakthroughs remain bottlenecked by a materials discovery process that takes decades, costs billions, and fails >90% of the time.

This disconnect became impossible to ignore. While generative AI was transforming entire industries, materials discovery remained largely untouched by such developments. That’s when we recognised our opportunity to build something truly transformative — combining frontier AI with deep materials expertise to fundamentally reimagine how humanity discovers and creates materials.

We knew from day one that the journey ahead would require not just breakthrough technology, but the guidance of people whose vision extends beyond the horizon of what most consider possible. That’s why we’re thrilled to announce the pioneers who’ll be joining Nobel Laureate and Turing Prize winner Professor Geoffrey Hinton on our Advisory Board:

• Professor Yann LeCun, Meta’s Chief AI Scientist, NYU Professor, and Turing Prize Laureate;
• Professor Kristin Persson, director of the Materials Project and UC Berkeley Professor, who was named Distinguished Scientist Fellow by the U.S. Department of Energy last year; and
• Verity Harding, AI policy expert, author, director of the AI and Geopolitics Project at the University of Cambridge, and one of TIME Magazine’s 100 most influential people in AI.

Yann says:

“The application of machine learning and modern AI to material science is one of the most promising approaches to addressing the biggest challenges faced by humanity. CuspAI’s combination of frontier AI and materials expertise is exactly what’s needed to tackle these challenges.”

This validation from one of AI’s foremost pioneers underscores what we’ve believed from the start — that we’re working at the intersection of two fields with extraordinary potential for impact. Their collective decision to join us isn’t just an endorsement — it’s practical support and a signal that the global community recognises the urgency of our mission.

At CuspAI, we’re building something that’s never existed before — a materials discovery engine that combines frontier AI with deep materials expertise to compress decades of traditional discovery into months. Our first focus areas span carbon capture, semiconductors and energy storage.

It’s why we’re delighted to welcome Kristin, the director and founder of the Materials Project, whose pioneering work aims to drastically reduce the time needed to invent new materials to serve societal needs.

Kristin says:

“I am thrilled to join the advisory board of CuspAI and collaborate with its outstanding team to advance AI-driven discovery and develop materials that support a sustainable future for our world.”

Any breakthrough on this scale needs careful consideration and engaging with relevant communities — just one of the reasons we wanted to engage Verity’s significant policy expertise from her work at DeepMind, the University of Cambridge and in government.

Verity says:

“For me, the true promise of AI has always been its ability to help humanity tackle our most difficult challenges. When I met Chad and Max, I immediately recognised that they had not only the talent but the clarity of purpose needed to put AI to work for that end. The potential for Cusp to revolutionise material design with AI is enormous and I’m delighted to support such brilliant and thoughtful founders with their mission.”

Since incorporating less than a year ago, we’ve assembled a world-class team of 27 experts across four jurisdictions, with collective scientific citations exceeding 2 million. We’ve secured $30 million in seed funding from leading European and US funds. And we’re already working with Meta on direct air capture, with more partnerships in the pipeline that we’ll announce soon.

We want to thank Jeremy Kahn at Fortune for covering this announcement, you can read his article here.

The road ahead is challenging but clear: we’re building the engine that will discover the materials humanity needs to thrive. What large language models did for AI, our engine will do for materials science: fundamentally transform what’s possible. With advisors of this calibre guiding us, we’ve never been more confident in our ability to deliver on that promise.

— Chad Edwards & Max Welling
Co-founders, CuspAI

CAMBRIDGE (UK), 0801 BST, TUESDAY 18 JUNE 2024: CuspAI, a transformational AI company building a platform for next-generation materials to tackle global sustainability and clean energy challenges has secured $30 million in seed funding from leading European and US venture funds.

The round was led by Hoxton Ventures, with significant participation from Basis Set Ventures and Lightspeed Venture Partners.

CuspAI leverages cutting-edge generative AI, deep learning, and molecular simulation to streamline the material design process. Their platform functions like a search engine for materials, allowing users to request specific properties for new materials on demand. This enables the rapid generation and evaluation of a vast number of novel structures, ultimately leading to the discovery of materials with precise functionalities.

One area in which the team believes AI-designed materials can have a significant near-term impact is carbon capture and storage, a critical technology for reducing greenhouse gas emissions and an industry expected to be worth $4 trillion by 2050.

CuspAI has been founded by Professor Max Welling, a renowned pioneer in AI and former Distinguished Scientist and VP at Microsoft Research and Qualcomm, and Professor at the University of Amsterdam. Co-founding the company with Welling is Dr. Chad Edwards, a chemist who has spent his career in deep-tech commercialisation including at Google and BASF and most recently quantum computing leader, Quantinuum.

Geoffrey Hinton, known as the ‘Godfather of AI,’ will also serve as a board advisor. Hinton said: “Humanity will face many challenges in the coming decade. Some will be caused by AI while others can be solved by AI. I’ve been very impressed by CuspAI and its mission to accelerate the design process of new materials using AI to curb one of humanity’s most urgent challenges — climate change.”

Professor Max Welling, Co-founder and Chief AI Officer at CuspAI, said: “Imagine a search engine not just for existing materials, but for all potential molecules and materials that could be created. Our AI can generate and evaluate new materials on demand. For example, you can request a material that selectively binds carbon dioxide under specified conditions — the AI then generates, evaluates and optimises the potential molecular structures that meet those exact criteria. Through careful process optimization and lab testing, we’re able to close the loop and ensure materials are synthesizable, stable and ultimately useful in production.”

CuspAI has partnered with Meta with a view to furthering its open science contributions focused on the discovery of new materials to address climate change.

Yann Le Cun, VP and Chief AI Scientist at Meta, said: “The Fundamental AI Research (FAIR) team is looking forward to collaborating with CuspAI in their use of AI, including our OpenDAC work, to accelerate the discovery of novel DAC sorbent materials. The world needs fast progress on affordable carbon capture, and we believe that CuspAI’s team is in an excellent position to apply AI-based materials discovery to this pressing problem.”

Dr. Chad Edwards, Co-founder and CEO of CuspAI, said: “The AI revolution is itself creating new challenges, including rapidly increasing energy consumption and carbon emissions from data centres. Our technology can help mitigate this impact by designing materials that efficiently capture carbon dioxide.

“Carbon capture is just the start — the material class we have in mind are well-suited for many other applications including energy storage, catalysis and gas and water purification. By harnessing AI for design and process optimisation we can create materials and solutions tailored to the specific needs of almost any industry. We are entering the age of ‘precision materials’.”

In addition to the founding team, CuspAI has appointed one of the world’s most accomplished computational chemists as Chief Scientist and assembled a world-leading research team across Cambridge (UK) and Amsterdam (NL), many of whom are joining from large tech players and are all motivated to ensure the latest advances in AI have a positive societal impact.

Other investors in the round include LocalGlobe, Northzone, Touring Capital, Giant Ventures, FJ Labs, Tiferes Ventures and Zero Prime Ventures. Prominent angel investors, including Mehdi Ghissassi and Dorothy Chou from Google DeepMind, also participated in the round.

Charles Seely, Partner at Hoxton Ventures, said: “I’ve known Chad for several years and recognized him as a truly gifted individual who had the ability to lead an extraordinary team. The partnership that he has forged with Max is one that I’m sure will help solve some of the world’s most critical problems.”

Lan Xuezhao, founding and managing partner at Basis Set Ventures, said: “The importance of new material discovery demands the mindshare of our absolute best and brightest. This Cusp team is exactly who you want dedicating their technical horsepower and business acumen to this essential effort.”

Paul Murphy, Partner at Lightspeed Venture Partners, said: “CuspAI is taking an entirely new approach to integrating efficient evaluation stacks, optimization, and generative models to create high-quality, economically viable materials that would have previously taken decades to discover. At Lightspeed, we believe the CuspAI team is leading the way into a new era where finding solutions for societally critical global challenges could become as easy as searching the web and we’re proud to partner with them.”

Hussein Kanji, Partner at Hoxton Ventures, said: “Our fund is premised on the idea that European founders would start and win the industry categories of tomorrow. It’s clear that AI has the ability to shake up how we design materials, and CuspAI has assembled some of the best people in the world to do this, starting with both Chad and Max.”

About CuspAI

CuspAl is an applied AI company dedicated to building a platform for next-generation materials that solve critical global challenges across sustainability and clean energy. By integrating advanced machine learning with material science and process design, CuspAl is unlocking the future of materials with AI. For more information visit: www.cusp.ai. Contact: press@cusp.ai