CuspAI Raises $450 Million: Using Generative AI to Transform New Materials Discovery and Industrial R&D Systems

The first wave of generative AI primarily took place in the fields of text, images, code, and video. While it has transformed knowledge production and digital content creation, it has not yet fully altered the underlying constraints of physical industries such as semiconductors, energy, chemicals, and automotive.

CuspAI aims to tackle precisely these most difficult, slowest, and most fundamental challenges facing these industries:Is it possible for artificial intelligence to directly design materials that have not yet been discovered in nature but meet specific industrial performance requirements?New Materials?

In July 2026, CuspAI, a U.K.-based materials AI company that had been in operation for just two years, announced the completion of a $450 million Series B funding round.financingThe company is valued at $2.6 billion. Investors include an investment firm affiliated with Amazon founder Jeff Bezos, Silicon Valley venture capital firm Kleiner Perkins, and a sovereign AI fund established by the British government. A month earlier, the *Financial Times* had reported that CuspAI was in the process of raising approximately $400 million; at that time, the terms of the deal had been signed but the transaction had not yet been finalized. The final publicly disclosed funding amount was subsequently increased to $450 million.

Unlike many AI companies focused on office software, marketing, customer service, and programming, CuspAI’s goal is not to create yet another chatbot, but to build a “discovery engine” for materials science:

A company specifies the properties a material must possess; artificial intelligence then searches for possible molecular or crystal structures, predicts their physicochemical properties, simulates their behavior in real-world environments, and narrows down trillions of candidate options to a small number of materials worthy of laboratory testing.

This is also the key reason why the capital markets were willing to value the company at $2.6 billion just two years after its founding. Investors are betting not just on a software product, but on a new type of infrastructure that has the potential to transform the way research and development is conducted in the semiconductor, energy, chemical, and advanced manufacturing industries.

I. Background: Materials science is becoming a key gateway for AI to enter the physical world

CuspAI was founded in 2024 and is headquartered in Cambridge, UK. It was co-founded by Chad Edwards and Max Welling.

The combination of the two founders is quite representative: one comes from the fields of chemistry, materials science, and deep-tech startups, while the other is a world-class machine learning researcher. From the very beginning, CuspAI has not been merely an algorithm company, nor has it been an internal digital division of a traditional chemical company; rather, it has sought to bring together cutting-edge artificial intelligence, computational materials science, and industrial R&D processes under one roof.

Co-founder and CEO Chad Edwards has a background in chemical research; he has personally worked in laboratories synthesizing materials and has also co-founded a quantum computing company. His understanding of the inefficiencies in materials R&D stems from his firsthand experience in scientific research: traditional materials research typically requires researchers to read a vast number of papers, formulate hypotheses, synthesize samples, engage in repeated trial and error, and then adjust their approach through experimentation. A seemingly simple performance goal may actually require years—or even more than a decade—of exploration.

Another co-founder, Max Welling, is a leading scholar in the fields of generative AI and machine learning. He is a professor of machine learning at the University of Amsterdam and one of the key contributors to the development of variational autoencoders; his research spans Bayesian inference, generative models, graph neural networks, computer vision, and AI for Science. Previously, Scyfer, an AI company he co-founded, was acquired by Qualcomm, and he has worked on related projects at Qualcomm and Microsoft Research AI4Science.

This founding team structure has shaped CuspAI’s technical direction: rather than simply repackaging a general-purpose large model as a “materials assistant,” the company is attempting to redesign the materials R&D process across multiple levels, including generative models, scientific data, physical simulations, and closed-loop experiments.

The founding of CuspAI also coincided with a major technological turning point.

In the past, while computational materials science was able to predict material properties using methods such as density functional theory, the computational costs were prohibitively high and the search space was far too vast. The sheer number of molecules and material structures that are theoretically possible far exceeds the scope of what can be exhaustively explored through any manual experimentation or traditional high-throughput computing.

With advancements in graph neural networks, generative models, machine-learning-based interatomic potentials, and large-scale materials databases, AI is now capable of rapidly predicting the stability, energy, and target properties of a vast number of candidate materials at a cost far lower than that of first-principles calculations. The OMat24 dataset released by Meta contains over 110 million density functional theory calculations, and the associated models have already achieved a high level of accuracy in predicting material stability and formation energies. This indicates that AI-based material models are evolving from “narrow-domain models trained on small datasets” toward general-purpose foundational material models.

CuspAI sees an opportunity: if these models are combined with industrial companies“ proprietary data, scientific literature, simulation tools, and experimental feedback, materials R&D could shift from ”experience-based trial and error“ to ”computational search driven by target performance.”

II. Financing and Development: Three Stages in Two Years

CuspAI is growing at an exceptionally rapid pace.

In June 2024, the company secured approximately $30 million in seed funding shortly after its founding. At the time, it was described as “a company building a search engine for new materials using generative AI.” This funding round enabled CuspAI to attract talent in machine learning, materials science, and computational engineering, and to begin collaborating with technology firms such as Meta.

In 2025, the company secured an additional $100 million in Series A funding. By September 2025, its valuation had reached approximately $520 million. At the same time, the company moved beyond proof-of-concept algorithms and began collaborating with companies in the semiconductor equipment, automotive, chemical, and internet sectors on actual materials projects. Recent reports indicate that prior to this funding round, CuspAI had raised a cumulative total of over $220 million, with investors including Singapore’s Temasek and the U.S. venture capital firm NEA.

The year 2026 marked CuspAI’s transition from a technology startup to a global materials R&D platform.

In June, the *Financial Times* reported that Kleiner Perkins and Bezos Expeditions participated in a new funding round of approximately $400 million, which would raise CuspAI’s valuation from $520 million to $2.6 billion. A follow-up report on July 20 indicated that the company ultimately completed a $450 million Series B funding round and secured backing from the U.K. sovereign AI fund.

This means that in less than a year, CuspAI's valuation has increased approximately fivefold.

Such rapid growth in valuations cannot simply be attributed to an “AI-related premium.” A more significant factor is that investors are beginning to recognize the economic value of materials innovation.

Virtually every industrial technological leap has depended on breakthroughs in materials: more advanced lithography processes require new photoresists, thin films, and dielectric materials; higher-density batteries require new electrodes and electrolytes; low-cost carbon capture relies on new adsorbents and porous materials; and advanced packaging, aerospace, and quantum computing are similarly constrained by material performance.

Therefore, CuspAI’s potential market is not limited to a single software market, but rather encompasses the R&D spending and materials supply chains behind multiple trillion-dollar industrial sectors.

III. Core Product: Not a Materials Database, but a “Goal-Driven Discovery System”

CuspAI is often described as a “search engine for the materials field,” but this description still does not fully capture the essence of its product.

Traditional search engines address two questions: whether a particular answer already exists, and how to find it among existing information.

CuspAI aims to address:When the required materials do not yet exist, how can we create a structure that might meet the requirements?

The product process can be broadly divided into five stages.

1. Translate industrial requirements into material specifications

The customer first specifies the target performance, for example:

• A lower dielectric constant;
• Higher heat resistance;
• Better electrical conductivity or insulation properties;
• Greater adsorption capacity for specific pollutants;
• Lower rare metal content;
• Higher strength-to-weight ratio;
• Lower costs and easier to scale up production.

This step may seem simple, but it is actually crucial. Industrial materials typically need to meet multiple mutually restrictive criteria simultaneously. A material may have excellent performance but be impossible to synthesize; it may be synthesizable but unstable; it may be stable but too costly; or it may be cost-effective but incompatible with existing production lines.

Therefore, true materials design is not about finding the maximum value for a single metric, but rather about conducting multi-objective optimization that balances performance, stability, manufacturability, cost, and environmental impact.

2. Aggregate literature, patents, and experimental data

CuspAI stated that the company has integrated a vast amount of literature and data on chemistry and materials science into a unified database and uses AI agents to help researchers quickly understand existing findings and technological gaps in a given field.

The first year of traditional doctoral research is often devoted to a literature review, but CuspAI aims to compress this process to a matter of seconds or minutes: the system automatically analyzes relevant papers, material structures, known properties, synthesis methods, and failure criteria, then generates a knowledge graph and research hypotheses that R&D teams can use.

The importance of this aspect is often underestimated.

The barriers to materials AI extend beyond the models themselves. Only those who have access to high-quality experimental data, failure data, process conditions, and proprietary customer data can train systems that are truly suitable for industrial R&D. Published papers typically focus on successful results, while a large number of failed experiments go unrecorded—this can lead to significant data bias in the models. CuspAI also acknowledges that scientific literature primarily tells the model “what has succeeded,” but rarely tells it “what has been attempted but failed.”

Therefore, the deeper significance of CuspAI’s collaboration with industry clients lies in continuously obtaining experimental feedback beyond what is available in public databases.

3. Reverse Design Using Generative Models

Research and development of traditional materials typically begins with known molecules or materials, followed by testing their properties; this is a “structure-to-property” forward approach.

CuspAI employs a reverse-design approach: it first defines the target performance and then works backward to identify the molecular, crystalline, or atomic structures that could achieve that performance.

For example, a semiconductor company needs a new material that meets specific requirements for dielectric properties, thermal stability, and manufacturing conditions. Rather than sifting through an existing list of materials, CuspAI’s generative model explores chemical space to generate a large number of new structures that may satisfy these constraints.

Reverse design is considered one of the most promising areas in AI-driven materials science, but it also presents significant technical challenges. Models must not only generate mathematically sound structures but also ensure chemical rationality, physical stability, synthesizability, and compliance with real-world manufacturing constraints.

4. Multi-stage Filtration and Digital Simulation

Generative models can propose 100,000 or more candidate structures, but these results cannot be directly applied in the laboratory.

CuspAI uses property prediction models, machine-learned atomic potentials, quantum chemical calculations, and other simulation tools to perform multi-stage screening. Low-cost models are used to first eliminate structures that are clearly unsuitable, while simulations that offer higher accuracy but are more costly are used to evaluate the top-ranked candidate materials.

CuspAI also uses AI agents to orchestrate simulation workflows. In the past, researchers had to select computational methods, prepare input parameters, submit tasks to computing clusters, review results, and make iterative adjustments; now, AI agents can automatically orchestrate this process and run a large number of simulation tasks in parallel. Chad Edwards noted that computational data that might have taken years to accumulate in the past can now be obtained in a matter of minutes or hours through large-scale parallel computing.

5. Experimental Validation and Closed-Loop Feedback

AI-generated material does not equate to something that has already been invented.

What truly determines commercial value is whether these candidate structures can be synthesized in the laboratory, whether they can demonstrate the predicted performance, and whether they can be produced at scale in an industrial setting.

Therefore, CuspAI must ultimately establish a closed-loop system with client laboratories, universities, research institutions, and manufacturing facilities. Experimental results are fed back into the model to correct prediction errors and further narrow the search range.

This is also the biggest difference between CuspAI and pure-software AI companies: its product development cycle is longer, and its solutions must undergo validation in the physical world; however, once validated, its competitive barriers may be far higher than those of ordinary enterprise software.

IV. Landmark Case: Finding PFAS Solutions Among 300 Trillion Possible Structures

The best example currently illustrating CuspAI’s technological approach is its PFAS materials R&D project with the Finnish chemical company Kemira.

PFAS are also known as “forever chemicals.” Due to their stable chemical bonds and resistance to natural degradation, PFAS have become a major challenge in drinking water and environmental management worldwide. Traditional treatment methods are often costly, have limited selectivity, and tend to generate secondary waste.

Kemira hopes to develop new materials that can capture PFAS more effectively.

Over the course of approximately six months of collaboration, CuspAI’s system explored roughly 300 trillion potential material structures, ultimately narrowing the field down to 20 candidate designs. These candidate materials then proceeded to the laboratory testing and further development phases.

The importance of this project lies not only in the scale of the search.

CuspAI stated that during the process, the system identified a class of molecular structures with high affinity for PFAS. These structures resemble molecular “cages” that allow PFAS molecules to enter and bind to internal sites. In other words, rather than simply identifying a known adsorbent from a database, the system helped the research team identify a family of materials that may operate through a new mechanism.

However, this case also highlights the practical limitations of AI in materials science.

The fact that 300 trillion possible structures have been narrowed down to 20 candidates does not mean the problem has been solved. These 20 materials still need to undergo synthesis, performance testing, toxicity assessment, durability verification, cost analysis, and evaluation for large-scale production.

What CuspAI truly saves time on is the longest and most haphazard phase of experimentation. It doesn’t eliminate experimentation; rather, it makes it more focused.

V. Team and Advisory Network: Bridging the AI Academic Community and the Global Industrial Ecosystem

CuspAI’s team strength stems not only from its two founders but also from the interdisciplinary network of advisors and industry partners it has built.

Among the advisors who have been publicly named are Geoffrey Hinton and Yann LeCun, leading researchers in the field of deep learning, as well as John Browne, former CEO of BP. Following its latest round of funding, the company also brought on John Giannandrea, a former AI executive at Apple and Google, to help establish its U.S. operations; it is also collaborating with Abhi Talwalkar, a director at AMD and chairman of Lam Research.

This lineup reflects the three types of capabilities required for CuspAI's development:

The first category consists of cutting-edge machine learning capabilities, including generative models, graph neural networks, foundational scientific models, and agent systems.

The second category is materials and chemistry competencies, which include molecular design, crystal structure, quantum chemistry, experimental synthesis, and process scale-up.

The third category is industrial implementation capabilities, which include semiconductor manufacturing, energy, automotive, chemical supply chains, and large-scale production.

A common reason for the failure of AI companies in the materials sector is that they possess only one or two of these capabilities. The algorithm team may not understand the real-world experimental constraints; the research team may lack scalable software systems; and industrial companies may have the data but struggle to consistently attract world-class AI talent.

CuspAI aims to bring these three capabilities together through its founding team, advisors, investors, and client relationships.

VI. AI Materials Foundry: What CuspAI Really Wants to Build May Not Be Software, but an Industrial Network

As part of its Series B funding round, CuspAI launched the “AI Materials Foundry.”

According to the latest reports, the alliance has brought together more than 48 technology firms, industrial companies, and research facilities, with members including NVIDIA, Meta, and Hyundai Motor Company. Its goal is to leverage cutting-edge AI, industrial data, simulation capabilities, and experimental facilities to collaboratively discover new materials needed for semiconductors, energy grids, batteries, carbon capture, and advanced manufacturing.

The Foundry model may be the key to understanding CuspAI's future business model.

By simply selling materials design software, a company risks falling into the traditional SaaS valuation model: charging based on the number of users, licenses, or computational capacity. However, if CuspAI can become a platform that connects customer needs, proprietary data, AI models, supercomputing resources, and experimental facilities, it could generate stronger network effects.

With each new industrial client added, the platform may gain new performance targets and experimental data; with each new laboratory added, the platform’s ability to validate materials is enhanced; and with each completed project, the model’s understanding of specific material systems deepens.

Over time, CuspAI may evolve into a “materials R&D operating system”:

Customers submit their requirements on the platform, and the system automatically performs literature searches, generates candidate compounds, conducts simulated screening, and coordinates experiments, while preserving the data assets generated throughout the entire R&D process.

The ultimate goal of such platforms may not be limited to charging customers software fees or R&D service fees; it may also include:

• Multi-year joint research and development contract;
• Charge a fee for locating materials on a per-project basis;
• Receive milestone payments for successful materials;
• Receive royalties from material patents or intellectual property;
• Establish a joint venture with chemical and semiconductor materials companies;
• Establish a pipeline of proprietary materials and authorize their production;
• Provides proprietary models, data, and computing infrastructure.

From this perspective, CuspAI is more like a new type of enterprise that combines a software platform, a research institute, and a materials intellectual property company.

VII. Competitive Landscape: AI in Materials is Entering the “Foundational Model Competition” Phase

CuspAI isn't the only company that has recognized this opportunity.

AI for Science has become a key focus for global technology investment. Competitors in the materials sector include startups using AI to discover crystal structures, catalysts, battery materials, and chemical molecules, as well as major tech companies such as Meta, Microsoft, and Google DeepMind, and traditional chemical and industrial software firms.

Competition can be broadly divided into four categories.

Category 1: Open Models of General-Purpose Technology Companies

Companies such as Meta are lowering the barrier to entry for AI-driven materials research by making material datasets and atomic models openly available. Datasets like OMat24 and general-purpose atomic models enable researchers to predict the energy and stability of materials on a much larger scale.

This presents both an opportunity and a challenge for CuspAI.

The opportunity lies in the fact that open-source models can become part of CuspAI’s underlying technology stack, reducing the cost of basic research; the challenge is that if foundational models gradually become commoditized, CuspAI must demonstrate that its value stems not only from the models themselves, but also from proprietary data, workflows, customer integrations, and experimental capabilities.

Category 2: Vertical-Focused AI Startups

These types of companies typically focus on specific fields, such as batteries, catalysts, medicinal chemistry, metal alloys, or crystalline materials. Their strengths lie in their deeper understanding of a single industry and their more tightly integrated experimental processes.

CuspAI, on the other hand, has adopted a more general-purpose platform approach, covering multiple industries such as semiconductors, automotive, aerospace, chemicals, energy, and environmental management. Its advantages include a large market potential and the ability to transfer data and models across industries; the risks, however, include excessive business diversification, as each materials system has its own scientific principles and validation processes.

Category 3: Internal Platforms of Traditional Materials and Chemical Companies

Large chemical, semiconductor, and automotive companies already have access to vast amounts of experimental data, engineers, and laboratory facilities. These companies are fully capable of building their own AI-powered materials platforms.

CuspAI’s approach is to collaborate with them rather than directly replace them. Customers provide industry challenges and proprietary data, while CuspAI provides models, agents, computing infrastructure, and cross-domain expertise.

However, in the long run, CuspAI must avoid becoming merely a one-time technology service provider. Only by transforming each project into reusable platform capabilities and data assets can the company continue to improve its gross profit margin and strengthen its competitive barriers.

Category 4: Automated Laboratories and Robotics Scientists

AI-generated material candidates are just the first step. In the future, the focus of competition will gradually shift to automated synthesis, robotic experimentation, automated characterization, and closed-loop optimization.

Only those who can fully integrate the “generation—simulation—synthesis—testing—feedback” process will be able to truly shorten the material R&D cycle.

Currently, CuspAI emphasizes collaboration with research facilities and industry laboratories rather than building out its own experimental capabilities through capital-intensive investments. This strategy facilitates rapid expansion but may also make the company reliant on partners for its experimental work.

VIII. Why Is Capital Willing to Offer a $2.6 Billion Valuation?

From the perspective of traditional financial metrics, a materials AI company that has been in operation for two years achieving a $2.6 billion valuation clearly reflects very high expectations for its future.

The capital markets are primarily betting on four things.

1. Materials are the fundamental bottleneck in industrial upgrading

Artificial intelligence computing power, electrification, the energy transition, and advanced manufacturing are all accelerating, but many technological pathways are constrained by material performance.

For example, more advanced chips require materials with lower dielectric constants and higher heat resistance; next-generation batteries need to simultaneously improve energy density, safety, and cycle life; and carbon capture requires adsorption materials that are lower in cost and more selective.

If CuspAI can solve just a few of these key problems, the economic value it creates could far exceed that of ordinary software tools.

2. AI could significantly reduce R&D costs

Traditional materials research and development involves long development cycles and high failure rates, and a significant portion of the costs is incurred in aimless experimental exploration.

CuspAI doesn’t need to get every prediction right. As long as it can narrow down the 100,000 candidates that would otherwise need to be tested to just a few dozen and significantly improve the success rate of experiments, it has the potential to save customers a significant amount of time and money.

3. Closed-loop data systems may create long-term barriers

General-purpose language models can be trained using text from the internet, but material models rely more heavily on high-quality physical and experimental data.

As CuspAI completes more projects, it has the opportunity to accumulate highly scarce data on which structures are synthesizable, which predictions will fail, and how process conditions affect final performance.

This data is more valuable than publicly available research papers and is also more difficult for competitors to replicate.

4. The platform may share in the proceeds from intellectual property rights related to the materials

Software companies' revenue typically depends on subscription fees, whereas new materials, once incorporated into the semiconductor, automotive, or chemical supply chains, can generate licensing and sales revenue for many years.

If CuspAI is able to secure intellectual property rights, patent licenses, or a share of commercialization proceeds under the contract, its revenue ceiling could be significantly higher than that of traditional R&D software.

IX. Major Risks: There Remains a Huge Gap Between “Candidate Discovery” and “Commercially Available Materials”

Despite its promising prospects, CuspAI still faces several challenges that cannot be ignored.

First, prediction accuracy does not equate to experimental success.

Material properties are influenced by multiple factors, including structure, purity, defects, temperature, pressure, and manufacturing processes. Materials that perform well in computational models may not be synthesized reliably or may fail in real-world environments.

Therefore, CuspAI cannot simply showcase its generation speed and search scale; it must also consistently report its experiment success rates, the extent to which it has shortened development cycles, and actual customer adoption rates.

Second, data quality may limit the model's performance.

There is a clear bias toward publishing successful results in scientific literature, and failed experiments are rarely made public. Data standards, measurement conditions, and material nomenclature also vary across laboratories.

If data cleaning, labeling, and feature engineering are inadequate, the model may draw incorrect conclusions based on superficial correlations.

Third, business cycles may be much longer than those in the software industry.

Semiconductor, automotive, and chemical materials typically require years of validation, and switching materials involves production line adjustments, certification, safety, and supply chain risks for customers.

Even if CuspAI has identified high-quality candidate materials, it may still take a long time to translate them into significant revenue.

Fourth, intellectual property rights are complex

The materials are developed based on customer requirements, generated by the CuspAI model, and validated by partner laboratories; the question of who ultimately owns the patent will be a key issue in the commercial partnership.

If CuspAI is unable to retain sufficient intellectual property rights, it may end up becoming a high-end R&D outsourcing firm; if it demands too many rights, however, it may deter large industrial clients from partnering with it.

Fifth, valuation places extremely high demands on commercialization.

A $2.6 billion valuation suggests that the market has already factored in the possibility of CuspAI becoming a global materials R&D platform.

Going forward, the company must not only demonstrate the effectiveness of its model, but also prove that projects can be delivered consistently, that customers are willing to pay on an ongoing basis, that the technology can be adapted to different material systems, and ultimately establish a scalable revenue model.

X. Future Outlook: CuspAI May Face Three Different Outcomes

Over the next five to ten years, CuspAI is likely to follow one of three development paths.

Path 1: Become a specialized software platform in the field of materials research and development

In this context, CuspAI sells models, agents, and computing platforms to companies in the chemical, automotive, and semiconductor industries, with revenue primarily derived from software subscriptions and R&D contracts.

This is the model with the lowest risk and the greatest potential for scaling, but its valuation ceiling may also be close to that of industrial software companies.

Pathway 2: Becoming a Global Infrastructure for Materials Discovery

Through AI Materials Foundry, CuspAI connects technology companies, laboratories, supercomputing centers, and manufacturers, serving as a cross-industry platform for materials innovation.

Under this model, the company benefits from data network effects and influence over industry standards, much like a cloud computing platform or an R&D operating system in the materials sector.

This is the direction CuspAI is most eager to pursue at this time.

Path 3: Become an “AI-driven chemical company” with its own material assets”

A more aggressive approach would be for CuspAI to go beyond simply helping clients discover materials; instead, it would independently identify major challenges, establish a materials R&D pipeline, file for patents, and generate long-term revenue through licensing, joint ventures, or manufacturing.

If successful, the company could position itself as a new chemical group or materials intellectual property platform; however, it would also face higher risks related to experimentation, regulation, manufacturing, and capital.

Judging by the scale of its latest funding round and its foundry strategy, CuspAI is likely transitioning from the first path to the second, while keeping the third path open as a possibility.

Conclusion: The greatest industrial value of AI may not lie in generating content, but in generating physical objects.

The rapid rise of CuspAI reflects a significant shift in the narrative surrounding the artificial intelligence industry.

Over the past few years, the market has focused primarily on whether AI can generate articles, images, videos, and code. In the next phase, what will truly determine AI’s long-term economic value may be its ability to solve real-world scientific and engineering problems.

Materials serve as the interface between the digital and physical worlds.

Chip performance depends on materials; battery efficiency depends on materials; the cost of clean energy depends on materials; and industrial emissions reduction and water pollution control likewise depend on materials. Algorithms can be continuously upgraded, but if the chips, energy, and manufacturing technologies that support their operation are constrained by material limitations, overall technological progress will slow down.

CuspAI is attempting to build a machine that can navigate the vast chemical space: companies describe a material that does not yet exist, and the AI is responsible for proposing structures, organizing calculations, simulating performance, and then passing the most promising options on to the laboratory.

This is not an easy path. There are still significant hurdles between AI-driven predictions and industrialization—including experimentation, manufacturing processes, costs, regulations, and the supply chain. But if CuspAI can continue to narrow down “trillions of possibilities” to “dozens of viable candidates” and ultimately transform some of those into new materials that actually enter the industrial system, it will revolutionize more than just materials science.

It could change the way humans create materials.