The problem, as Dr. Chad Edwards puts it, is older than most of chemistry itself: discovering new materials takes forever. A decade, maybe two. Tens of millions spent. Success, when it arrives at all, through expensive fumbling rather than elegant design.
Edwards, who spent years at Google and BASF before co-founding Cambridge-based CuspAI in 2024, thinks his startup can compress that timeline to six months using AI. It's the sort of claim that makes seasoned industry executives reach for their skepticism—and their checkbooks. In September 2025, NEA, Temasek, and NVIDIA's venture arm led a $100 million Series A round that valued the company, by some whispered estimates, near the $1 billion mark just 15 months after its seed.
Bold promises are cheap. Execution is another matter entirely. Yet CuspAI enters 2026 with something rarer than hype: partnerships already inked with Kemira, the Finnish water treatment giant (targeting materials to scrub PFAS "forever chemicals" from drinking water), Hyundai Motor Group (next-generation mobility applications), and Meta's FAIR lab (direct air capture sorbents). The company claims it moved a carbon capture material from generative design through synthesis and experimental validation in six months during 2025.
If that pace holds—a significant qualifier—the implications ripple far beyond climate tech. Semiconductors. Batteries. Advanced manufacturing. Any industry where material performance dictates what's possible.
Why Discovery Takes So Damn Long
Traditional materials R&D unfolds in a familiar, grinding rhythm. Researchers dream up candidate compounds. Simulate a handful using quantum mechanics software. Synthesize the most promising in the lab. Test against specifications. Repeat.
Each cycle: months. Scale that across thousands of chemical combinations and you grasp why bringing a novel battery electrolyte or carbon capture sorbent to market often requires a decade or two and budgets approaching $100 million—figures the Acceleration Consortium in Canada has documented in strategic assessments.
The past 18 months, though, have delivered a cascade of large-scale datasets and foundation models aimed squarely at this bottleneck. Google DeepMind's GNoME project predicted 2.2 million candidate inorganic crystals in 2023, work the company characterized as equivalent to roughly 800 years of traditional discovery. About 381,000 of those candidates exhibit thermodynamic stability (sitting on the Materials Project convex hull); more than 700 have been independently synthesized and verified in labs worldwide.
Meta's FAIR lab released OMat24 in 2024—over 110 million density functional theory calculations with pre-trained neural network models achieving roughly 20 meV/atom mean absolute error on formation energy predictions. In 2025, FAIR followed with Open Catalyst 2025 (7.8 million DFT calculations for solid-liquid interfaces across 88 elements) and ODAC25 (approximately 70 million single-point calculations across roughly 15,000 metal-organic frameworks for direct air capture scenarios).
These open datasets matter because they democratize what was once proprietary. Where a corporate R&D team once needed months generating simulation data, they can now fine-tune a publicly available foundation model in weeks. The Materials Project, a long-running Department of Energy program, surpassed 650,000 users by early 2026 and delivered 465 terabytes of data in two years. That surge reflects something fundamental shifting.
How CuspAI Says It Works
The startup positions itself as a "search engine for materials." Customers specify target properties—a sorbent that selectively binds PFAS molecules at parts-per-trillion concentrations, say, or a polymer that withstands 200°C while remaining flexible. CuspAI's generative AI models propose synthesizable candidates, validate them using physics-based simulations, then rank options by likelihood of meeting specifications.
Professor Max Welling, co-founder and an AI luminary who previously served as VP of Technology at Qualcomm, framed the vision in a 2024 interview: "Imagine a world where you can just search for materials with specific properties, and the AI will generate and optimize them for you."
Welling and Edwards secured a $30 million seed round in June 2024 led by Hoxton Ventures, with Basis Set and Lightspeed participating. The September 2025 Series A, fifteen months later, brought NEA and Temasek as co-leads, joined by NVentures, Samsung Ventures, Hyundai Motor Group, and others. The company employs roughly 30 people across Cambridge, Amsterdam, Berlin, Tokyo, Edinburgh, and Lausanne.
The advisory board reads like someone raided the upper echelons of AI research and industrial engineering: Geoffrey Hinton and Yann LeCun, alongside Martin van den Brink (former ASML CTO and President) and Lord John Browne (former BP CEO). That combination—deep learning pioneers plus industrial execution veterans—signals ambitions well beyond academic proof-of-concept.
The Partnerships That Actually Matter
Credibility in this space hinges on real-world execution, not press releases. In July 2025, CuspAI announced a partnership with Kemira to develop materials removing per- and polyfluoroalkyl substances from drinking water. Kemira framed the collaboration as cutting development timelines from "up to 10 years to as little as 6 months."
The motivation is regulatory, and urgent. The U.S. EPA finalized the first-ever national drinking water standards for PFAS in April 2024, setting maximum contaminant levels at 4 parts per trillion for PFOA and PFOS, 10 ppt for several others. Compliance monitoring begins in 2027. Treatment systems required by 2029, though the EPA is considering extensions to 2031 for some utilities.
Across the Atlantic, the European Chemicals Agency is evaluating an EU-wide PFAS restriction proposal through late 2026, while France has enacted consumer product bans phased from 2026 to 2030. A January 2026 study estimated that PFAS pollution in the EU could cost between €330 billion and €1.7 trillion by 2050 depending on mitigation efforts.
For Kemira, partnering with an AI platform to accelerate sorbent discovery isn't altruistic—it's economics. Water utilities face compliance deadlines and capital constraints; faster paths to PFAS-removal materials translate directly into market advantage.
In November 2025, Hyundai Motor Group formalized a strategic partnership with CuspAI for next-generation mobility materials—likely targeting lightweight composites, battery components, and thermal management materials balancing durability, manufacturability, and lifecycle performance. Automakers are under pressure to electrify fleets while managing supply chain constraints and sustainability mandates. AI-driven materials discovery offers a differentiation lever they desperately need.
Perhaps the highest-profile collaboration links CuspAI with Meta's FAIR lab around direct air capture. FAIR co-released the ODAC25 dataset with Carnegie Mellon and other partners; CuspAI participated in that ecosystem. The company's year-end 2025 wrap noted it moved a DAC material "from generative design, through synthesis and experimental validation in six months."
Direct air capture is capital-intensive and material-constrained. The U.S. Department of Energy awarded up to $1.2 billion in August 2023 for the first two DAC hubs (Occidental's 1PointFive and Climeworks among awardees), with an additional $1.8 billion opened in December 2024. The 45Q tax credit offers up to $180 per ton for DAC coupled with geologic storage, adjusted for inflation. If CuspAI's sorbents demonstrate superior CO₂ selectivity or regeneration efficiency, they could influence the economics of large-scale DAC deployment. Could. Not will.
Market Forces Creating Pull
McKinsey estimated in November 2024 that generative AI could unlock $80 billion to $140 billion in annual value across energy and materials functions. The chemicals industry, however, remains an early adopter—just 14% current generative AI exposure versus 23% cross-industry average. Translation: significant headroom, or massive inertia, depending on your optimism.
Market research firms project AI-driven materials discovery platforms will grow from approximately $1 billion to $1.3 billion in 2024–2026 to between $2.77 billion and $12.5 billion by 2034–2035. Compound annual growth rates in the 25% to 30% range. The wide variance reflects differing methodologies, but the directional trend is consistent.
Several catalysts converge. Regulatory timelines for PFAS remediation create near-term, policy-anchored demand for new adsorbents and membranes. DOE DAC hubs and 45Q incentives provide revenue streams for carbon capture materials. The CHIPS and Science Act allocates $11 billion for semiconductor R&D, with notices issued in October 2024 for AI-powered autonomous experimentation targeting sustainable semiconductor materials—up to $100 million per award.
Beyond compliance and subsidies, the underlying economics of faster R&D cycles matter. Microsoft and Pacific Northwest National Laboratory published a 2024 case study in which AI screened 32 million battery electrolyte candidates in 80 hours, narrowing the field to 18 promising compounds. One candidate, NaxLi3–xYCl6, reduces lithium content by roughly 70% and reached prototype battery demonstration.
Aionics, a battery materials AI startup, partnered with Showa Denko to accelerate electrolyte R&D and reported discovering a record-breaking lithium thioborate solid electrolyte using machine learning-guided workflows combining small initial datasets with DFT narrowing and lab validation. Schrödinger documented an active-learning workflow for OLED materials that screened 9,000 molecules with an 18-fold speedup versus quantum-mechanics-only approaches.
Academic benchmarking studies of self-driving labs—autonomous platforms combining AI, robotics, and in-line characterization—report median acceleration factors around 6× across published literature, with effectiveness peaking in higher-dimensional chemical search spaces. Canada's Acceleration Consortium received $200 million from the Canada First Research Excellence Fund to build national self-driving lab capacity, partnering with BASF on formulations for agriculture, coatings, and drug delivery.
A Crowded, Fragmented Field
CuspAI enters a competitive landscape that's both crowded and oddly diffuse. Citrine Informatics offers a materials informatics platform with publicized case studies at AGC Glass. Kebotix combines AI with autonomous labs for novel materials. Intellegens markets Alchemite for formulation and process optimization, with references in polymers, inks, and additive manufacturing. Aionics focuses on battery materials. Orbital Materials, backed by NVentures, develops foundation models (LINUS) and simulation tools it calls "Orb."
Established software providers like Schrödinger, Dassault Systèmes BIOVIA, and Ansys integrate machine learning into existing physics simulation stacks. Google DeepMind and Meta FAIR publish foundational research that enables industrial teams to fine-tune domain-specific models. National labs including Lawrence Berkeley's Materials Project and the University of Toronto's Acceleration Consortium operate as data infrastructure and training hubs.
What distinguishes CuspAI, at least in its early narrative, is a sharp climate and environmental focus—PFAS removal, direct air capture, an expanding push into semiconductors—paired with Fortune 500 enterprise partnerships inked within the first 18 months. The involvement of Hinton and LeCun as advisors lends AI credibility. Van den Brink and Browne signal industrial execution experience.
Whether the platform's six-month claim holds across diverse material classes and customer specifications? That's the central question investors are betting on.
What Could Go Wrong
Materials discovery timelines will likely continue compressing as open datasets proliferate and foundation models mature. The OMat24, OC25, and ODAC25 releases make industrial fine-tuning standard practice rather than competitive advantage. Real differentiation may shift to synthesis scale-up, regulatory navigation, and integration with existing manufacturing workflows.
Risks include synthesis bottlenecks—designing a molecule is easier than making it at scale—data governance under the EU AI Act, and lab automation throughput constraints. The Toyota Research Institute's 2024 survey of 102 autonomous labs researchers documented practical challenges around interoperability, error handling, and skill gaps that slow adoption beyond proof-of-concept. Flow chemistry and modular fluidic robots are emerging as enabling architectures for closed-loop experimentation, but capital costs remain high.
CuspAI's December 2025 update mentioned expansion into semiconductors, a domain where material properties intersect with nanoscale fabrication and supply chain geopolitics. The timing is deliberate: CHIPS Act funding creates addressable market pull, but execution complexity scales exponentially.
The Bet Underneath the Bet
For climate tech investors, CuspAI represents a wager that regulatory deadlines—PFAS compliance by 2029, DAC project timelines, semiconductor supply chain resilience—create market pull strong enough to overcome the traditionally glacial adoption curves in chemicals and materials.
Corporate innovation leaders will watch whether Kemira's six-month target materializes in commercial PFAS adsorbents, and whether Hyundai's partnership yields production-ready mobility materials. Sustainability executives evaluating AI tools should note that McKinsey's $80 billion to $140 billion annual value estimate hinges on enterprise adoption that has barely begun. Chemicals remain at 14% generative AI exposure versus 23% cross-industry.
NEA's Lila Tretikov, who joined CuspAI's board with the Series A, framed the opportunity in September 2025: AI as a "catalyst for discovery" enabling breakthroughs "in months rather than decades."
If that proves out, the implications extend beyond any single startup. Into how industries slow to innovate might suddenly accelerate. Into whether the next generation of environmental solutions arrives in time to matter.
That's the bet underneath the bet. Not just whether CuspAI succeeds, but whether the materials discovery bottleneck—one of the oldest problems in chemistry—finally cracks open. Edwards and Welling have $100 million and 15 months of partnerships to prove it. The chemistry industry, for once, seems willing to watch.
