CuspAI

CuspAI (registered as Cusp AI Limited) is a British artificial intelligence company, headquartered in Cambridge, England, that builds generative models and molecular simulation software for discovering and designing new materials. It was founded in 2024 by Chad Edwards, a chemist who previously worked in commercial roles in quantum computing and is the company's chief executive, and Max Welling, a machine learning professor at the University of Amsterdam and co-author of the variational autoencoder, who is chief technology officer. ^[1][2][3][4] The company describes its product as a "search engine" for materials: a customer specifies the properties it needs, such as a sorbent that binds carbon dioxide or a semiconductor with a target bandgap, and CuspAI's models generate, simulate and rank candidate structures intended to be synthesizable. ^[3][5][6]

CuspAI came out of stealth in June 2024 with a $30 million seed round led by Hoxton Ventures. It raised a Series A of more than $100 million co-led by New Enterprise Associates (NEA) and Temasek in September 2025, and in July 2026 announced a $450 million Series B co-led by Kleiner Perkins and NEA that the company said valued it at $2.6 billion. ^[3][7][8] Alongside the Series B it launched the AI Materials Foundry, a network of more than 45 founding members, including NVIDIA, the Fundamental AI Research (FAIR) team at Meta, Samsung, Hyundai Motor Group and Applied Materials, that connects CuspAI's software with partners' data, computing and laboratories. ^[8][6] Its disclosed industrial work includes carbon capture materials with Meta and the Georgia Institute of Technology, materials for removing PFAS ("forever chemicals") from water with the Finnish chemicals group Kemira, and a materials partnership with Hyundai Motor Group. Its advisers include Geoffrey Hinton and Yann LeCun. ^{[1][9][10][11]}

Overview

The founders frame materials discovery as an inverse problem. Instead of making a material and then measuring what it does, the user starts from the target properties and asks a model for structures that meet them. Edwards put it this way to TechCrunch in 2024: "if you can put materials or molecules in and get properties out, then why can't you put properties in and get materials and molecules?" ^[5] CuspAI says its platform combines generative AI, deep learning and molecular simulation, and claims that it produces new, synthesizable candidates up to ten times faster than traditional discovery methods. ^[3][7]

A recurring theme in the company's pitch is manufacturability. Edwards has described CuspAI's models as "synthesis-aware generative AI models" that are more likely to propose materials chemical companies can actually make. ^[12] Fortune noted in 2025 that some chemists had criticized earlier AI materials projects, such as Google DeepMind's GNoME and work by Microsoft's quantum team, for producing candidates that would be too difficult or expensive to mass-produce. ^[13]

On its business model, Welling told Fortune in March 2025 that CuspAI would find commercial partners to produce and test the compounds its models suggest, while retaining a portion of the intellectual property so that it could earn licensing and royalty revenue from eventual sales. ^[13] In September 2025 Edwards described a dual strategy of designing materials to order for large partners and developing materials for the company's own projects, with the emphasis so far on partner work. ^[12] Tech.eu and the Financial Times (as relayed by SiliconANGLE) have reported that its customers include ASML and Meta. ^[14][15] The company has not disclosed revenue; in December 2025 its founders wrote that the size of the team and "the value of our contracts" had both grown twentyfold in 21 months. ^[9]

History

Founding and seed round (2024)

Cusp AI Limited was incorporated in England and Wales on 9 March 2024 and has its registered office in Cambridge. ^[2] A 2026 profile of Edwards published by the investor Northzone says that he grew up in rural Wales, was the first in his family to go to university, where he studied chemistry, and saw industrial customers' unmet demand for new materials while working at Cambridge Quantum Computing. According to the profile, he approached Welling as Welling was leaving his vice-president role at Microsoft Research, and the company's name refers to being on the "cusp" of a new era of materials demand. ^[16]

On 18 June 2024 CuspAI announced a $30 million seed round led by Hoxton Ventures, with significant participation from Basis Set Ventures and Lightspeed Venture Partners. Other investors were LocalGlobe, Northzone, Touring Capital, Giant Ventures, FJ Labs, Tiferes Ventures and Zero Prime Ventures, and the angel investors included Mehdi Ghissassi and Dorothy Chou of Google DeepMind. ^[3][5][17] The company announced Hinton as an adviser at the same time, along with a partnership with Meta. LeCun, then Meta's chief AI scientist, said FAIR was looking forward to collaborating with CuspAI on its use of AI, "including our OpenDAC work, to accelerate the discovery of novel DAC sorbent materials." ^[3] Welling's title at launch was chief AI officer. ^[3] Carbon capture was the first target: Edwards described "molecular sponges" that selectively absorb carbon dioxide from the air and release it when heated. ^[5]

Advisory board, hires and first partnerships (2025)

In March 2025 CuspAI announced that LeCun, Kristin Persson of the University of California, Berkeley, who directs the Materials Project, and the AI policy specialist Verity Harding had joined Hinton on its advisory board. ^[13][18] Fortune reported at the time that it had hired several people from Google DeepMind and related companies, including Alessandro De Maria, the first employee of Isomorphic Labs, as vice president of platform engineering, and Markus Hoffmann, formerly head of global partnerships for Google's Quantum AI group, as chief strategy officer. ^[13] The company then had offices in Cambridge, Edinburgh, Amsterdam and Berlin and said it had a team of 27 across four jurisdictions. ^[13][18] Welling told Fortune that it was using an ensemble of models that optimize different aspects of a material and was building a central AI agent to coordinate them. ^[13]

On 9 July 2025 Kemira announced a strategic partnership with CuspAI, describing it as a water-industry first; the first project would target the removal of PFAS from water, with an initial focus on in silico design. ^[19] In August 2025 researchers from Meta, Georgia Tech, CuspAI and other institutions released the Open DAC 2025 dataset for carbon capture sorbents. ^[20]

Series A (September 2025)

On 10 September 2025 CuspAI announced a Series A of more than $100 million, co-led by NEA and Temasek, with participation from NVentures (NVIDIA's venture arm), Samsung Ventures, Hyundai Motor Group and returning investors. Other participating firms included Basis Set Ventures, FJ Labs, Giant Ventures, LocalGlobe, Northzone, Prosus Ventures, Tiferes Ventures and Touring Capital. ^[7][21] The company listed angel investors including OpenAI co-founder Durk Kingma, Welling's co-author on the variational autoencoder paper, ^[4] as well as Google research vice president Zoubin Ghahramani, Dropbox founder Arash Ferdowsi, Hugging Face founder Thomas Wolf and Synthesia chief executive Victor Riparbelli. The company said NEA partner Lila Tretikov, a former deputy chief technology officer of Microsoft, would join its board. ^[7] Fortune reported, citing one source familiar with the deal, that the round valued CuspAI at $520 million; the company declined to comment on its valuation, and its law firm Mishcon de Reya later described the valuation as over $500 million. ^[12][22]

With the round, Martin van den Brink, the former president and chief technology officer of ASML, and Lord (John) Browne, the former chief executive of BP, joined the advisory board. ^[12][7] Fortune put the headcount at about 30 and reported that the money would pay for hiring and new offices, with Asia a particular focus. ^[12] On 6 November 2025 Hyundai Motor Group announced a strategic partnership, signed in Cambridge, to apply AI to materials for its future mobility products. ^[11]

In a year-end review the founders listed a catalysis collaboration with the Danish company Topsoe and the Technical University of Denmark, announced the appointment of Imperial College London professor Aron Walsh as chief scientific officer, and said a London office in Kings Cross would open in January 2026. ^[9] Imperial College later said Walsh had joined in January 2026 while keeping his post as Chair in Materials Design at Imperial. ^[23]

2026: Giannandrea, kUPS and the Kemira results

On 23 April 2026 CuspAI open-sourced kUPS, a molecular simulation toolkit (see below). ^[24] On 27 April Upstarts Media reported that former Apple AI chief John Giannandrea would work part-time with CuspAI to open a Bay Area office and recruit AI staff for its US operations. The same report said CuspAI had already raised more than $200 million in total through a previously unreported extension of its Series A, and that it was reported to be raising $200 million or more at a valuation above $1 billion. ^[25] CuspAI later listed Giannandrea as an adviser and said he was helping to set up its US foundry operations. ^[1][8]

On 21 May 2026 Kemira and CuspAI published the results of their first PFAS project (see below). ^[10] On 17 June SiliconANGLE, citing the Financial Times, reported that a term sheet had been signed for a $400 million round that would value CuspAI at $2.6 billion, with Bezos Expeditions and Kleiner Perkins among the participants. ^[15]

Series B and the AI Materials Foundry (July 2026)

On 20 July 2026 CuspAI announced a $450 million Series B led by Kleiner Perkins and NEA, "with significant participation" from Bezos Expeditions, the family office of Jeff Bezos. The company said the round valued it at $2.6 billion. ^[8][14] New investors included Glade Brook Capital Partners, Lux Capital, AMD Ventures, Tru Arrow Partners, StepStone, Britain's government-backed Sovereign AI fund, the Netherlands' Invest-NL and John Doerr. Temasek, Basis Set Ventures, Giant Ventures, Touring Capital, Prosus, Phoenix Court and Northzone returned. ^[8][26] Dealroom reported that the round confirmed and enlarged the deal first reported in June at a $400 million target. ^[27] Tech.eu called the valuation a fivefold jump from the September 2025 valuation and put total funding at more than $670 million; the company's own description says it has raised over $650 million. ^[14][1]

The company launched the AI Materials Foundry the same day, together with a five-year partnership with Singapore's Agency for Science, Technology and Research (A*STAR) and a new office in Singapore. ^[6][28] It also named Abhi Talwalkar, a board member of AMD and chairman of Lam Research, as an adviser. ^[8] By then the founders said the team was spread across Cambridge (where Edwards is based), Amsterdam (where Welling is based), Berlin, Tokyo, Singapore and the United States, and the company's website also lists London. ^[8][1]

Technology

Discovery platform and MIRA

CuspAI describes its workflow in five steps: define objectives and specifications, generate novel candidate materials, simulate high-fidelity properties, synthesize with laboratory validation, and scale to a production-ready result. ^[1] In their December 2025 review the founders wrote that the system uses physics-based simulation to rule out implausible candidates early, chooses experiments "based on their expected contribution to reducing uncertainty," and feeds the results back to recalibrate its models. They said a carbon capture project called SkyVault had taken materials from generative design through synthesis and experimental validation in six months. ^[9]

At the July 2026 launch the company named its proprietary platform MIRA, which it describes as an agentic platform and an autonomous scientific agent. A partner defines a target, for example a semiconductor with a particular bandgap or a catalyst with a defined reaction profile; MIRA generates candidate structures, predicts properties across millions of candidates, designs synthesis routes matched to the laboratories in the network, routes the work to a suitable facility, and feeds the outcomes back into its models. The company said partner data is kept in private Foundry instances. ^[6]

Data

CuspAI says it holds exclusive AI training rights to the Cambridge Structural Database, through the Cambridge Crystallographic Data Centre (CCDC), and to the Inorganic Crystal Structure Database (ICSD), through FIZ Karlsruhe, and that it has licensed materials science content from Wiley and other journal publishers. ^[6] Executives of the CCDC and ICSD are quoted as founding partners of the Foundry on the company's website. ^[1] The company said the Foundry's simulation layer would also use Meta's UMA (Universal Model for Atoms) models for simulating atomic interactions. ^[6]

Open DAC 2025

The Open DAC 2025 (ODAC25) dataset, released in August 2025 by authors from FAIR at Meta, Georgia Tech, Carnegie Mellon University, Oak Ridge National Laboratory and CuspAI, expands the earlier ODAC23 dataset for direct air capture of carbon dioxide. It contains nearly 60 million density functional theory single-point calculations of CO2, H2O, N2 and O2 adsorption in 15,000 metal-organic frameworks (MOFs), and was released with machine-learned interatomic potentials trained on it. ^[20] Five of the fifteen authors, including Welling, list CuspAI as their affiliation. The dataset's 460 synthetically generated MOFs were selected from 53,000 candidate structures produced by CuspAI's in-house generative model of MOFs, an autoregressive transformer that samples a framework's topology, metal clusters and ligands. The paper's Widom insertion calculations used a package released on CuspAI's GitHub account. ^[20]

kUPS and other open-source software

kUPS, released under the Apache 2.0 licence in April 2026, is a toolkit for molecular simulation written in JAX. It provides composable, differentiable "samplers, potentials, and propagators" that run on CPUs, GPUs and TPUs, covering molecular dynamics ensembles, Monte Carlo methods including grand canonical sampling, and geometry optimizers, with classical force fields and machine-learned potentials such as MACE and UMA. ^[24][29] CuspAI also released Tojax, a library that converts PyTorch functions into JAX functions so that existing PyTorch potentials can be used inside kUPS. ^[24]

The company said it optimized kUPS for GPUs in collaboration with NVIDIA. In benchmarks on a single NVIDIA L4 GPU it reported 1.8 times the throughput of ASE for molecular dynamics of a 114-atom MOF with the MACE potential, and 49 times the throughput of the single-core CPU code RASPA for grand canonical Monte Carlo simulation of CO2 adsorption in a larger MOF supercell. These are company-reported figures for specific workloads. ^[24] The AI Materials Foundry's simulation layer runs on kUPS, which CuspAI said was being integrated with NVIDIA's ALCHEMI (AI Lab for Chemistry and Materials Innovation) platform. ^[6]

Partnerships and projects

Kemira PFAS project

PFAS are persistent synthetic chemicals found in drinking water and subject to tightening limits in the United States and the European Union; the standard treatment is granular activated carbon. Kemira's brief asked for materials that remove specific PFAS molecules at sub-parts-per-billion concentrations, using chemistry that is water-stable, environmentally compatible, synthesizable and cost-effective. CuspAI's platform designed new MOFs from scratch, exploring a design space of about 300 trillion possible structures and producing more than 5,000 designs with property data for three target molecules (GenX, PFBS and PFOS). These were narrowed to about 20 priority candidates that, as of May 2026, were moving into further development and testing, and the discovery phase took six months. ^[10] The companies described it as the first commercial partnership to apply generative AI end-to-end to designing PFAS remediation materials. Kemira chief executive Antti Salminen said the work gave the company "a credible path toward a next-generation PFAS remediation product." ^[10]

AI Materials Foundry

The AI Materials Foundry, launched on 20 July 2026, is described by CuspAI as a global network of data, labs, compute and scientific expertise for designing new materials, "orchestrated by a single agentic platform," with regional hubs in the United States, Europe and the Asia-Pacific region. ^[6] It had more than 45 founding members at launch; A*STAR said it joined 47 other partners. ^[8][28] NVIDIA provides the compute infrastructure, and the Foundry draws on the UMA models developed by Meta's FAIR team. ^[6][26] Edwards said: "If we don't make progress fast, the next 50 years of industrial progress will be constrained by a single challenge: the world needs materials that don't yet exist." ^[6]

Founding members quoted on CuspAI's website include NVIDIA, Meta, Applied Materials, Lam Research, Tokyo Electron, imec, Kioxia, Fujifilm, Mitsui Chemicals, Resonac, 3M, Universal Display Corporation, Qnity Electronics, Oxford PV, SoftBank Corp., Kemira, Hyundai Motor Group, A*STAR, the Tyndall National Institute, Wiley, the CCDC and ICSD. ^[1] The company also named Samsung and Henkel among the founding partners, and the UK's Henry Royce Institute for advanced materials said it had joined as a founding member. ^[8][30] Rob Fergus, head of FAIR at Meta, described Meta as a "longstanding" partner of CuspAI. ^[6]

Funding

The sources do not say whether the $2.6 billion figure is a pre-money or post-money valuation. EU-Startups converted the Series B to EUR 393.2 million at a EUR 2.2 billion valuation. ^[26]

People

Founders

Chad Edwards (chief executive) holds a PhD in nuclear fission and computational quantum chemistry and an MBA, according to a 2026 London Tech Week speaker profile. ^[31] CuspAI's launch announcement said he had worked in deep-tech commercialization at Google and BASF and most recently at the quantum computing company Quantinuum. ^[3] The London Tech Week profile describes him as a commercial leader at Cambridge Quantum who helped the firm through its merger with Honeywell Quantum Solutions to form Quantinuum, and says he earlier worked at Google Quantum AI and BASF; Northzone's profile calls him Cambridge Quantum's first commercial hire. ^[31][16] Fortune reported that he was working on strategic partnerships at Quantinuum before he decided that AI could have a more immediate impact on materials discovery. ^[13]

Max Welling (chief technology officer) is a professor of machine learning at the University of Amsterdam. ^[32][17] He co-founded the university spin-off Scyfer, which Qualcomm acquired in August 2017; at the time he co-directed QUVA, a joint Qualcomm and University of Amsterdam research lab. ^[33][39] He later served as vice president of technologies at Qualcomm and as a vice president and distinguished scientist at Microsoft, and Fortune describes him as a physicist by training. ^[13] He told Fortune that he found too much bureaucracy and "viscosity" at large companies and wanted to build a startup devoted to designing materials with AI. ^[13]

Welling's research includes several methods now used in machine learning for molecules and materials. With Kingma he wrote "Auto-Encoding Variational Bayes" (2013), which introduced the variational autoencoder; with Thomas Kipf he wrote the 2016 paper on semi-supervised classification with graph convolutional networks, a graph neural network architecture. ^[4][34] He co-authored the E(n) equivariant graph neural network (2021) and an equivariant diffusion model for generating molecules in 3D (2022), part of the field of geometric deep learning. ^[35][36] His publication list also includes the 2025 Nature paper "A foundation model for the Earth system," which describes Microsoft's Aurora model. ^[32] CuspAI describes him as a co-inventor of the variational autoencoder and of the equivariant architectures that underpin generative molecular design. ^[6]

Leadership and advisers

Sources for the table: company website, Fortune, Imperial College London, Intelligent CIO and CuspAI announcements. ^{[1][13][12][23][6][8]} Walsh remains at Imperial, where he leads the Materials Design Group, and Intelligent CIO's account of the Foundry launch identifies him as a Fellow of the Royal Society. ^[23][6]

Reception and context

Fortune has described CuspAI as one of several AI startups trying to discover new physical materials. ^[12] When it announced its seed round, TechCrunch named Schrodinger and Dassault Systemes as incumbents in computational chemistry software and Orbital Materials as a newer competitor. ^[5] In September 2025 Edwards said CuspAI's $30 million seed had been the largest for a materials science AI startup at the time, and Fortune noted that Lila Sciences had since raised a $200 million seed round and that Periodic Labs had also raised a much larger one. ^[12]

Investors' public statements have emphasized the team and manufacturability. NEA's Tretikov told Fortune she was impressed by the team's pedigree, its datasets and how far along its foundation models were. ^[12] Kleiner Perkins partner Josh Coyne said that what set the founders apart was that "they design for materials that can actually be built, not just ones a model can dream up." ^[8]

Coverage of the Series B also noted the distance between computational results and products. An eWeek analysis said the company's public results "still stop well short of commercial deployment," noting that whether the Kemira candidates can be made economically and work at industrial scale was unproven, and that the Hyundai announcement did not identify a finished material or publish performance results. ^[37] Kemira and CuspAI themselves described the PFAS candidates as advancing to further development and testing rather than as finished products. ^[10]

Debarghya Das's "The Ultimate Neolab List," an aggregator of new AI research companies, includes CuspAI in its "Science & math" group, under materials, and rates it "Mostly" a neolab. Its note on the company reads: "Scientist-led model company with no disclosed revenue and a big raise; tied to lab partners rather than pure software." ^[38]

See also

• List of neolabs
• Neolab
• AI for Science
• UMA (Universal Model for Atoms)
• Open Catalyst Project
• Periodic Labs
• Lila Sciences
• MatterGen
• GNoME

References

1. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10CuspAI, "CuspAI: AI-powered materials discovery," company website (leadership, advisers, partners and company information), accessed October 2026.
2. ^1 ^2Companies House, "CUSP AI LIMITED, company number 15549537," company overview, accessed October 2026.
3. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10CuspAI, "CuspAI Secures $30M to Combat Climate Change with AI-Designed Materials," press release dated June 18, 2024, republished on the company blog.
4. ^1 ^2 ^3Diederik P. Kingma and Max Welling, "Auto-Encoding Variational Bayes," arXiv:1312.6114, December 20, 2013.
5. ^1 ^2 ^3 ^4 ^5 ^6Mike Butcher, TechCrunch, "CuspAI raises $30M to create a GenAI-driven search engine for new materials," June 18, 2024.
6. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15Intelligent CIO Europe, "CuspAI launches global AI Materials Foundry with NVIDIA, Meta and 45 founding partners," July 20, 2026.
7. ^1 ^2 ^3 ^4 ^5 ^6Chad Edwards and Max Welling, CuspAI, "Securing our $100M+ Series A to Revolutionise Materials Discovery with AI," September 10, 2025.
8. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12Chad Edwards and Max Welling, CuspAI, "Launching our 'AI Materials Foundry' and $450 million Series B," July 20, 2026.
9. ^1 ^2 ^3 ^4 ^5Chad Edwards and Max Welling, CuspAI, "A Year at the Cusp: 2025," December 18, 2025.
10. ^1 ^2 ^3 ^4 ^5 ^6Kemira, "New AI-Designed Materials Show Promising Potential to Remove 'Forever Chemicals' from Drinking Water in Industry-First Breakthrough," press release, May 21, 2026.
11. ^1 ^2 ^3Hyundai Motor Group, "Hyundai Motor Group and CuspAI Partner to Accelerate Material Innovation Using AI," press release, November 6, 2025.
12. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11Jeremy Kahn, Fortune, "AI for chemistry startup CuspAI raises $100 million in new venture capital funding," September 10, 2025.
13. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10Jeremy Kahn, Fortune, "Startup aiming to build AI models for chemistry adds two AI 'godfathers' to advisory panel as it grabs top research talent from Google," March 5, 2025.
14. ^1 ^2 ^3 ^4 ^5John Reynolds, Tech.eu, "Jeff Bezos and Sovereign AI back CuspAI in $450M raise," July 20, 2026.
15. ^1 ^2Maria Deutscher, SiliconANGLE, "AI material discovery startup CuspAI reportedly raising $400M round," June 17, 2026.
16. ^1 ^2Northzone, "Material Revolution: A Portrait of CuspAI's Chad Edwards," June 10, 2026.
17. ^1 ^2Northzone, "Our Investment in CuspAI: Pioneering AI-Designed Materials to Combat Climate Change," June 18, 2024.
18. ^1 ^2Chad Edwards and Max Welling, CuspAI, "Announcing the CuspAI Advisory Board," March 5, 2025.
19. ^1 ^2Kemira, "Kemira and CuspAI Forge Strategic Partnership to Pioneer AI-Driven Materials Innovation," press release, July 9, 2025.
20. ^1 ^2 ^3 ^4Anuroop Sriram, Logan M. Brabson, Xiaohan Yu, Sihoon Choi, Kareem Abdelmaqsoud, Elias Moubarak, Pim de Haan, Sindy Löwe, Johann Brehmer, John R. Kitchin, Max Welling, C. Lawrence Zitnick, Zachary Ulissi, Andrew J. Medford and David S. Sholl, "The Open DAC 2025 Dataset for Sorbent Discovery in Direct Air Capture," arXiv:2508.03162, submitted August 5, 2025 (revised September 23, 2025).
21. ^Kyt Dotson, SiliconANGLE, "CuspAI raises $100M to build AI 'search engine' to transform materials science," September 10, 2025.
22. ^1 ^2Mishcon de Reya, "Mishcon de Reya advises CuspAI on $100 million Series A funding round," September 23, 2025.
23. ^1 ^2 ^3Imperial College London, "Imperial expert leads science at £2 billion AI materials company CuspAI," July 21, 2026.
24. ^1 ^2 ^3 ^4 ^5Nicholas Gao, Jonas Köhler, Felix Hanke and Anita Ramanan, CuspAI, "kUPS: a molecular simulation engine for the AI era," April 23, 2026.
25. ^1 ^2Alex Konrad, Upstarts Media, "Scoop: Apple's Former AI Chief John Giannandrea To Join Science Startup CuspAI," April 27, 2026.
26. ^1 ^2 ^3 ^4Rahul Raj, EU-Startups, "CuspAI raises €393.2 million at €2.2 billion valuation; launches AI Materials Foundry to accelerate materials discovery," July 20, 2026.
27. ^Dealroom, "CuspAI closes $450M Series B at $2.6B valuation, backed by Bezos and UK Sovereign AI Fund," July 20, 2026.
28. ^1 ^2 ^3Agency for Science, Technology and Research (A\*STAR), "CuspAI and A\*STAR Announce Five-year Partnership To Accelerate AI-driven Materials Discovery In Singapore," press release, July 20, 2026.
29. ^CuspAI, "kUPS: A high-performance toolkit for atomistic simulations in JAX," GitHub repository (Apache 2.0 licence), accessed October 2026.
30. ^Luke Richards, Henry Royce Institute, "Henry Royce Institute Joins CuspAI AI Materials Foundry as Founding Member," July 20, 2026.
31. ^1 ^2London Tech Week, "Chad Edwards," speaker profile, 2026.
32. ^1 ^2University of Amsterdam, "Prof. dr. M. (Max) Welling," staff profile and publication list, accessed October 2026.
33. ^Richard Chirgwin, The Register, "Qualcomm slurps Uni of Amsterdam AI spinoff Scyfer," August 21, 2017.
34. ^Thomas N. Kipf and Max Welling, "Semi-Supervised Classification with Graph Convolutional Networks," arXiv:1609.02907, September 9, 2016.
35. ^Victor Garcia Satorras, Emiel Hoogeboom and Max Welling, "E(n) Equivariant Graph Neural Networks," arXiv:2102.09844, February 19, 2021.
36. ^Emiel Hoogeboom, Victor Garcia Satorras, Clément Vignac and Max Welling, "Equivariant Diffusion for Molecule Generation in 3D," arXiv:2203.17003, March 31, 2022.
37. ^eWEEK Staff, eWeek, "CuspAI Raises $450M as AI Materials Discovery Enters Its Validation Phase," July 2026.
38. ^Debarghya Das, "The Ultimate Neolab List," accessed October 27, 2026.
39. ^Global Venturing, "Qualcomm decodes Amsterdam's Scyfer," August 17, 2017.