AI in the Lab: The Billion-Dollar Race to Accelerate Materials Discovery

Hello readers,

One of the most exciting parts of being a researcher was finding smarter ways to build better materials. And doing it faster! This was something which could save a lot of valuable time in research, where the environment for innovation can feel quite stressful. In science, the right tools change everything. Recently, artificial intelligence (AI) has entered the spotlight as one such tool that is reshaping how we do research. This will be the focus for today’s newsletter.

In materials science specifically, AI can now help predict the properties of new compounds before they are ever made in the lab. Theoretically, this would mean researchers could skip years of trial-and-error and go straight to materials that are more likely to work. From identifying stable semiconductors to designing molecules with specific optical or thermal properties, AI has the potential to make the discovery process faster, cheaper, and far more targeted. I use the word ‘potential’ carefully here because there are also a lot of drawbacks with AI which I focus on later in this newsletter.

However, to highlight just how significant this trend is: the global market for AI in materials science was worth around $1.1 billion in 2024. This is projected to surge to $11.7 billion by 2034.

🔍 Why This Matters Now
There are a few key reasons why this moment is so important:

• Explosion of data: The field of materials science generates enormous datasets! Every property that can be measured has been measured, from crystal structures to conductivity measurements. AI thrives on this as it can be useful for identifying patterns amongst a set of different materials.

• Rising cost of R&D: Companies and universities are under pressure to innovate faster while spending less. AI models can cut experimentation time significantly.

• Competitive edge: Nations and corporations are racing to own the “next-gen” materials, whether it is for batteries, electronics, or medical devices. Whoever gets there first could control the market. Therefore, the research capabilities and tools, such as AI, could be a strong catalyst for these new discoveries.

And in a world where AI tools like AlphaFold and DeepMind’s GNoME are proving their worth, the ecosystem for AI-led materials discovery is only gaining speed.

📈 Market Intelligence: Where This Is Headed
The AI + materials space is attracting serious investment. We're talking deep tech venture funds, corporate R&D budgets, and national research strategies all converging.

What’s fueling the projected growth?

• Tech convergence: AI, quantum computing, and automation are beginning to stack together, creating exponential potential. This increases the relevance and computing power that may be available.

• Sustainability goals: Governments are seeking greener materials — lighter, cheaper, less toxic. AI makes it possible to screen thousands of material candidates against these criteria quickly.

• Real-world traction: Startups like Citrine Informatics and Kebotix are already licensing AI-discovered materials to manufacturers, shaving years off product development timelines.

🧠 My Analysis: A Closer Look

While this growth looks exciting, I am slightly sceptical about the current portrayal of AI in research. it’s worth unpacking some of the assumptions. Let’s have a look together:

🔸 First, AI models are only as good as their training data, and the time required for the model to learn and produce the results. Much of the historical data in materials science is inconsistent, siloed, or lacking in edge-case examples, which can limit the performance of even the best models. Furthermore, AI models that are trained using computational data may not be accurate enough which also distorts the accuracy of any outputs that are produced. Even AI models require multiple iterations to refine and make them more accurate. In addition, these AI models would likely require long periods of time to learn from the data, which increases the timeline of a viable product.

🔸 Second, AI doesn’t replace experimentation. It just reorders it for later. We still need to synthesise, test, and verify the materials AI suggests. In some cases, researchers may spend just as much time validating a promising lead as they would have discovering it manually. This is because the AI models may require long durations in order to produce any outputs, as data sets can be very large to analyse even for AI models!

🔸 Third, regulatory and trust barriers exist. If a new material is found by an algorithm, how do we ensure it’s safe, scalable, or sustainable? These are factors which the algorithm may not consider. What happens if the AI model suggests something that no one knows how to manufacture?!

That said, the long-term potential is hard to ignore. Imagine small labs and startups gaining discovery capabilities that were once exclusive to billion-dollar R&D departments. Imagine finding sustainable alternatives to rare-earth metals or toxic materials without decades of trial-and-error.

💭 Final Thoughts

AI in materials science isn’t just another tech buzzword but a foundational shift in how we invent things. And when you speed up invention, you change who gets there first — and who profits. But, even with all that said, I am still currently sceptical of the practical use of AI because of the amount of time required to train useful AI models and the quality of data needed. As a former researcher who recognises that a lot of published data is inflated, AI models still have a while to go to reach their full potential. The issue of obtaining reliable data first needs to be addressed before widely-used AI models can be developed.

As always, I’ll be keeping an eye on how this evolves, especially the new models, and startups in this space.

Thank you for reading.

Until the next time,
Qasim Ibraheeme