CuspAI was reported in June 2026 to be raising approximately $400M at a $2.6B valuation, a roughly 5x step-up from the $520M post-money valuation established at its Series A in September 2025.

The Series A was a $100M round co-led by NEA and Temasek, with participation from NVentures, Samsung Ventures, and Hyundai Motor Group. Earlier backers included Hoxton Ventures, Basis Set Ventures, Lightspeed Venture Partners, LocalGlobe, Northzone, Touring Capital, Giant Ventures, FJ Labs, Tiferes Ventures, and Zero Prime Ventures.

Before the Series A, CuspAI raised a $30M seed round in June 2024 at its public launch. Total disclosed funding through the Series A stands at $130M.

CuspAI is a materials discovery platform built around an inverse industrial R&D workflow. Instead of starting with a known material and asking what properties it has, a customer starts with the properties it needs, a sorbent that removes PFAS from drinking water at trace concentrations, is water-stable, and can be manufactured economically, and the platform generates candidate materials that could satisfy those requirements.

The product loop starts with the partner and CuspAI defining target properties and practical constraints together. Generative models then propose large numbers of novel candidate structures across a design space that can span hundreds of trillions of possibilities. Those candidates are scored and filtered computationally, then evaluated more rigorously using physics-based atomistic simulation. The highest-priority candidates are translated into an experimental plan for the partner's lab team to test, and lab results feed back into the next search cycle.

CuspAI's generative models are synthesis-aware, meaning they are constrained to produce structures that can actually be made rather than only theoretically interesting ones. For industrial customers, that matters because a candidate that cannot be synthesized at scale has limited commercial value regardless of its computed properties.

Simulation is a core part of the product. CuspAI has built and open-sourced kUPS, a GPU-native molecular simulation engine supporting Monte Carlo, molecular dynamics, geometry optimization, and differentiable simulation primitives, with benchmarked speedups of up to 49x over legacy tools on certain workloads like MOF adsorption screening. The platform integrates with NVIDIA's ALCHEMI stack and is designed for large-scale active-learning loops in which simulation, data generation, and model training iterate continuously.

The Kemira PFAS collaboration is the clearest public example of the product in use. Starting from a target specification, CuspAI explored roughly 300 trillion possible structures, produced over 5,000 novel material designs with computed property data for three target PFAS molecules, and narrowed those to priority experimental candidates, all within six months.

CuspAI sells to large industrial R&D organizations on a B2B basis, using strategic enterprise partnerships rather than self-serve software adoption. Its commercial structure is a hybrid of platform access fees and custom discovery programs: a partner pays for CuspAI's generative and simulation capabilities applied to a defined materials problem, with scientific delivery staff embedded in the engagement.

The cost structure is heavier than classic SaaS. CuspAI bears compute costs for large-scale simulation and model training, employs interdisciplinary researchers across ML, computational chemistry, and materials science, and uses application scientists as the operational bridge between the AI platform and partner lab teams. That makes early gross margins lower than pure software, but the company is asset-light relative to traditional materials development businesses because it does not own or operate wet labs or manufacturing infrastructure.

Each partner program can generate new simulation data, experimental feedback, and domain-specific validation that improve model quality and candidate ranking for future programs. The OpenDAC collaboration with Meta and Georgia Tech on direct-air-capture sorbent datasets is one example of CuspAI's role in large-scale data creation that feeds back into platform capability.

Over time, monetization may extend beyond discovery fees toward participation in the downstream value of the materials themselves, through IP sharing, licensing, milestone payments, or commercialization arrangements. If the company can repeatedly find commercially deployable materials, that could support economics beyond a pure software subscription model.

CuspAI competes across four layers: frontier AI model quality, physics-based simulation credibility, enterprise workflow integration, and the ability to convert digital candidates into manufacturable materials. Few competitors are strong across all four.

Orbital Materials is the closest direct strategic analogue, combining pretrained chemical models, generative design, and simulation for advanced materials. It is also moving further down the value chain by commercializing its own AI-developed carbon-removal products rather than remaining a pure platform. That vertical integration matters competitively: if the market rewards companies that own the path from model to product, CuspAI faces pressure to do more than generate candidates and hand them off.

Kebotix approaches the market from the autonomous-lab direction, combining AI, physical modeling, and a self-driving lab in a predict-produce-prove cycle with partners including Bayer, bp, and Mitsubishi Chemical. For buyers focused on time-to-proof rather than model novelty, Kebotix's closed-loop execution can be more compelling than a platform that still relies on partner labs for validation.

Periodic Labs is the most ambitious emergent threat, with a founding team that includes contributors to GNoME and MatterGen. It is not yet a feature-level competitor, but it competes for the same frontier talent, capital, and buyer attention as CuspAI.

Citrine Informatics is already embedded in the daily R&D workflows of large industrial customers across chemicals, batteries, metals, and aerospace. Its advantage is not frontier model novelty but procurement maturity: enterprise SaaS packaging, security certifications, and a track record as the system of record for proprietary materials data. A customer that already trusts Citrine to organize its R&D knowledge can absorb more of the generative design workflow without adopting a separate platform.

MaterialsZone competes from a similar angle, leading with data unification across ERP, LIMS, ELN, and PLM systems and layering in AI-guided experimentation through its Maven agentic interface. It can land lower in the org chart, with formulation teams and process engineers, and then move upward into discovery tasks that overlap with CuspAI's value proposition.

Schrödinger is the strongest physics-first incumbent, with a materials science platform that combines molecular modeling, machine learning, and enterprise informatics, plus a strategic integration with Ansys. Its advantage is not generative novelty but prediction reliability, explainability, and integration into existing engineering workflows, attributes that conservative industrial buyers often prioritize.

Microsoft's MatterGen and Azure Quantum Elements combine property-guided inorganic materials generation, accelerated DFT, and HPC in a cloud stack that many enterprise customers already use. Google DeepMind's GNoME expanded the visible search space of candidate crystals at large scale and changed buyer expectations around baseline discovery throughput. Neither is a startup-style direct competitor, but both commoditize core model capabilities and raise the question of why a buyer would purchase a specialized platform when a cloud vendor already offers generative design and simulation acceleration bundled into existing infrastructure.

CuspAI's platform is materials-agnostic by design, so the same generative and simulation infrastructure can be applied to carbon capture, water treatment, automotive materials, semiconductor process chemistry, and advanced batteries without rebuilding the company for each vertical. The expansion logic is to direct a common engine at industrial problems with larger budgets and higher urgency.

CuspAI's earliest commercial proof points span water treatment (Kemira, PFAS remediation) and mobility materials (Hyundai Motor Group), but the platform is positioned across semiconductors, carbon capture, catalysis, separations, and energy materials. Each new vertical adds another class of industrial R&D budget to the addressable market.

Semiconductors and AI compute are a particularly large adjacent market. As AI demand increases pressure on chip performance, packaging, thermal management, and process chemistry, demand for new materials in those areas also rises. CuspAI's investor base, NVentures and Samsung Ventures, provides an entry point into that customer cluster, and the kUPS simulation engine and NVIDIA collaboration are relevant to compute-intensive materials problems in that space.

Regulation is also expanding the water treatment opportunity. The EPA finalized drinking-water rules for PFAS in 2024 and has continued to propose additional rules, creating durable demand for better separation and remediation materials. That makes the Kemira relationship an entry point into a growing regulated market rather than a one-off project.

After the Series A, CuspAI announced plans to open new offices with a focus on Asia, and 2026 hiring shows a Singapore buildout for scientific applications and partner management. Asia concentrates much of the global manufacturing base for semiconductors, batteries, chemicals, and automotive supply chains, the industries where materials bottlenecks are often most costly.

The investor and partner base in the region, Temasek, Samsung Ventures, Hyundai Motor Group, extends beyond financial backing. Those relationships provide access to regional industrial ecosystems where co-development programs, pilot projects, and manufacturing-oriented validation can be structured more quickly than through cold outreach.

This mirrors a pattern seen in other simulation-heavy scientific AI companies like Iambic Therapeutics, where proximity to key research and manufacturing clusters improved deal flow and experimental feedback loops.

CuspAI's current model is relatively high-touch, with application scientists embedded in each partner engagement. The longer-term expansion opportunity is to standardize more of the discovery workflow into reusable platform products, reducing bespoke labor per program and improving gross margins as the customer base grows.

The open-sourcing of kUPS points in that direction. A reusable, modular simulation engine can help recruiting and reduce internal tooling duplication, while the highest-value models, datasets, and workflows remain proprietary. If that productization continues, the model can shift from services-heavy discovery work toward higher-leverage software and workflow infrastructure.

A larger long-term opportunity is to capture downstream value from the materials themselves. If CuspAI repeatedly finds commercially deployable materials, it could negotiate IP-sharing, licensing, or commercialization participation structures that extend its economics beyond a platform access fee alone would support.

Validation timelines: Even if CuspAI compresses candidate generation from years to months, industrial customers still face testing, qualification, scale-up, and procurement cycles that can stretch years, which delays revenue realization from any given discovery program relative to the platform's technical output and concentrates the business around a small number of long-duration programs where a single setback can materially affect near-term financials.

Services drag: Because each enterprise program currently requires embedded application scientists and bespoke scientific delivery, the cost structure scales with headcount rather than software usage, which caps gross margins below what a software model would imply and creates a risk that the business remains a high-touch discovery shop rather than converging toward the scalable software economics implied by its valuation.

Baseline commoditization: Microsoft's MatterGen and Azure Quantum Elements, Google DeepMind's GNoME, and the broader open-source materials AI ecosystem are progressively commoditizing the core model layer, which means CuspAI's defensibility must rest on synthesis-aware generation, proprietary simulation infrastructure, and industrial validation depth rather than model access alone, a harder position to sustain as well-capitalized labs continue to release capable open models.