For AI to transform materials, it can’t stop at discovery.

The pitch for AI revolutionizing material discovery and development is both straightforward and tantalizing: 

• The materials that underpin our modern world were discovered through decades of painstaking trial and error, often relying on experienced researchers’ intuition and accumulated knowledge. 

• Using ever-advancing AI techniques and more powerful computers, researchers can compress those decades to months by identifying novel compounds and better predicting their characteristics. 

• These new compounds will lead to a wave of new products that are better, cheaper, and have a lower carbon footprint than their predecessors. 

Unfortunately, the reality is significantly more complicated.

Experts consistently told us that AI tools focusing on discovering potential new compounds only address a small portion of what it takes to bring new materials to market – maybe 10%-20%. The rest of the journey involves identifying viable synthesis pathways, scaling those synthesis pathways, optimizing the resulting processes, developing supply chains, and integrating new materials into devices. The latter parts of this process can still take 10+ years. To truly advance the commercialization of new materials, this remaining 80% must also be addressed.  

So where should startups and investors looking to unleash the next wave of materials put their time and capital? Below, we explore in more detail the barriers that remain for developing and scaling the next generation of materials, as well as the opportunities to address them. 

Key takeaways

• AI focused solely on material discovery isn’t sufficient to bring new materials to market effectively. Instead, there must be more focus downstream on efficient synthesis, scalable manufacturing, and integration into end-use products.

• A major limiting factor for AI for materials discovery is insufficient datasets. Companies who are able to develop proprietary, robust datasets will hold a distinct advantage. Autonomous labs and synthetic data are two solutions today, and we expect other creative approaches in the future.

• Business models that allow companies to capture value commensurate with their contribution will be key. While we believe innovative business models will emerge, we’ve seen some companies opt to become fully integrated material developers in order to ensure they can capture all of the value from their material discovery and formulations.

If you’re an early-stage startup founder building software to unlock the next generation of materials, or if you’ve also been closely looking at the intersection of AI and material development, we’d love to hear from you! Contact us.

What is AI for materials, and why it matters for energy, DAC, chemicals, and manufacturing

At its core, AI for materials is using computational techniques to: a) predict atomic structures that will exhibit a specific set of characteristics, b) assess the stability of the identified structures, and, in some cases, c) suggest potential pathways for materials synthesis (that is, the process by which a set of raw materials can create the target material). 

AI for materials isn’t new, but like many areas of AI and machine learning, it has benefited from substantial advancements in computing power and computational techniques. Google DeepMind’s announcement in November 2023 that it had discovered 2.2 million new crystals, including 380,000 stable materials, exemplifies the steps forward. Google estimated that its discovery “would be equivalent to about 800 years’ worth of knowledge and demonstrates an unprecedented scale and level of accuracy in predictions.” 

For an accessible dive into the different technical approaches used in AI for materials, we recommend the research piece by Hitachi Ventures and KOMPAS VC, AI is Powering the Future of Material Science: From Lab to Real-World Breakthroughs.  

While next-generation materials will be relevant across a number of fields, there are particularly exciting opportunities within the areas of energy and climate tech. Just a few possibilities include:

• Clean energy: Next-generation battery materials, more efficient solar cells, better catalysts for cost-effective low-carbon hydrogen production

• Carbon capture: Novel sorbents and membranes for direct air capture

• Sustainable chemicals: Chemicals with improved performance and lessened environmental impacts

• Advanced manufacturing: Lighter, stronger materials for transportation and construction

  
  

Unlocking the next generation of materials

The opportunities for startups and investors to unlock value lie in addressing pain points throughout the materials development and commercialization process. 

Pain Point #1: Shortcomings of existing datasets

AI can only be as effective as the data used to train it. And across critical parts of the AI for materials landscape, there are significant gaps in the data. This is true for three reasons:

1. Failed experiments, which should make up an important part of a robust dataset, rarely get published.

2. Corporate R&D teams that often focus on synthesis and scale-up (understandably) don’t share their results.

3. Decades of research remain locked away within PDFs and lab notebooks, making them inaccessible for training AI. 

Startups that  develop a proprietary dataset that includes failed experiments will hold a distinct advantage.

1. Opportunity: Data organization, management, and AI accessibility

• Organizing data is step one to using data: Even within organizations with meaningful datasets, the data that could help inform successful innovations using AI are sometimes locked away in inaccessible formats. Data platforms that not only organize data but also make it usable for AI applications can play an important role in allowing companies to maximize the potential of the data they already have. That said, we expect there to be a ceiling to the total value companies solely focused on this pain point can capture, and a few leaders have already emerged.

• Notable startups: Citrine Informatics (Series C), Ontochem (Acquired by Digital Science), Compular (Seed)

  
  

b. Opportunity: Autonomous, robotic labs

• Speeding up experimentation and creating a closed-loop feedback cycle: One way to speed up the process of experimentation is simply to do experiments much more quickly. Robotic labs, which could be instructed to act with increasing levels of autonomy based on the feedback they receive, provide one pathway. Google DeepMind and Lawrence Berkeley National Lab published a paper in Nature specifically showcasing the success of their autonomous “A-Lab” in developing and testing synthesis pathways for the solid-state synthesis of inorganic powders. As another example, Dr. Benji Maruyama of the Air Force Research Lab also led work on what would become ARES OS, an open-source software that allows users to transform automated experiments and carbon nanotube synthesis reactors into autonomous “research robots” capable of directing and conducting their own research using AI and automation. Speeding up experimentation speeds up data creation, helping companies improve their models and make better decisions.

• Notable startups: Radical AI (Seed), Dunia Innovations (Seed), Kebotix (Series A), Lila (Series A), Altrove (Seed)

c. Opportunity: Synthetic data creation

• Creating data where it doesn’t exist: In some situations where the dataset is incomplete but the physics is well understood, there can be opportunities to use synthetic data as a supplement. We believe synthetic data can be one tool applied alongside others to create a compelling training dataset.

Pain Point #2: Missing cost-effective, commercially viable synthesis pathways and tools to produce at scale

It is one thing to create a new material with exciting properties in the lab. It is quite another to find synthesis pathways with materials and steps that can be done cost effectively, then to do so at commercial scale.

1. Opportunity: Synthesis focused datasets, tooling, and analysis

• Focusing on how we make materials: If classic AI tools for materials discovery cover only about the first 10% of a new material’s journey to market, tooling focused on synthesis pathways takes on the next critical step – figuring out the raw materials and processes that can make those materials a reality. The best startups will pursue synthesis with an eye toward available and affordable raw materials, as well as pursuing synthesis techniques that can be scaled commercially. Like companies focused on materials discovery, these companies will need to find creative business models to successfully capture value.

• Notable startups: Newfound Materials (Pre-VC), Matched Materials (Pre-VC)

  
  

b. Opportunity: Improving manufacturing outcomes for advanced materials

• Improving tooling to increase yields: For certain subsets of advanced materials, it can be difficult to successfully execute known viable synthesis pathways with sufficient consistency to drive commercially viable yields. Embedding AI-powered software into tools used for high precision manufacturing can create new, relevant datasets, then use that data to generate and retrain models very quickly. This in turn could unlock materials that haven’t been commercially viable to date. 

• Notable startups: Atomscale (Pre-VC), Gauss Labs (Seed)

Pain Point #3: Challenges integrating new materials into established supply chains and products 

Even materials with compelling properties and viable synthesis pathways can be stymied by the inability to integrate them successfully into established supply chains and end-use products. Existing supply chains for common products are derisked, diversified, and optimized. Legacy materials companies will also likely have a limited appetite to pay for materials only proven at a small scale, knowing that they’d take on the scale-up and commercialization risks. 

1. Opportunity: AI-first integrated materials companies 

• Going from discovery through manufacturing: Given these dynamics, some companies have opted to focus on a narrow set of materials or applications and become fully integrated, including discovery, synthesis optimization, and manufacturing. This more capital-intensive approach gives companies the ability to capture the full value of their discoveries, while also assuring they aren’t blocked by stakeholders with non-aligned incentives. 

• Notable startups: MitraChem (Series B), Orbital Materials (Series A), Immaterial (Series A)

b. Opportunity: New business models for collaboration with legacy manufacturers

• Finding pathways for shared success: Some material companies have been anxious to license their discoveries to major companies, similar to the way a biotech firm might license the discovery of a new drug. However, in practice we have rarely seen this approach work well. Unlike new medicines, which are usually manufactured and distributed through well-known processes and have a well-defined customer set, new materials require taking significant scale-up risk during manufacturing and must penetrate markets with well-defined substitutes. It also often requires large capital outlays. This has made legacy manufacturers wary of signing a licensing agreement that would provide exciting returns to the discoverer. Successful AI for materials companies who choose not to be fully integrated will need to find creative ways to limit those risks for potential partners, and may also need to explore alternative contracting structures. 

Summarizing where we see opportunities in AI for materials at Powerhouse Ventures

At Powerhouse Ventures, we back seed-stage founders building innovative software to advance clean energy, mobility, and industry. Given that each of these sectors can be revolutionized by the commercialization of new, high-performing materials, we are excited by the role AI tools can play in speeding up their timelines. The opportunities lie not only in discovery, but in accelerating all of the steps necessary to bring those discoveries to market. 

If you’re an early-stage startup founder building software to unlock the next generation of materials, or if you’ve also been closely looking at the intersection of AI and material development, we’d love to hear from you! Contact us.