AI will be everywhere in chemistry

Artificial Intelligence is the hottest ticket in pretty much every industry, and that includes chemistry. While some chemists are using the tool to discover new drugs, others are deploying it to generate ecofriendly materials or to make manufacturing processes more efficient and sustainable. C&EN reached out to the leaders of five companies working at the intersection of AI and chemistry to gather their views on the field’s possible defining moments in 2025.

Alex Zhavoronkov

Founder and CEO, Insilico Medicine

Insilico Medicine is a leader in AI-based drug discovery, but the company also uses generative AI for environmental sustainability. Zhavoronkov says this could be a big year for using AI to discover and manufacture materials needed for carbon capture.

“We are very early in utilizing generative AI for CO₂ capture in direct air capture systems,” Zhavoronkov says in an email. “Hopefully, next year, we will see some breakthroughs. I think the cost per ton of captured CO₂ is about $500–600. If we get to $50 per ton, we do not need more conservation efforts in the oil and gas industry, and it can be close to being CO₂ neutral. At some point in time in the next 5–10 years, I am confident we will get there, but 2025 will be a big year for proof of concept.”

Henrik Hahn

Chief digital transformation officer, Evonik Industries

AI technologies such as large language models (LLMs)—AI systems that can understand and generate human language—will see advancements in 2025, according to Hahn.

“While LLMs have indisputably great potential to accelerate materials understanding and discovery, they still fall short as practical research and development tools, primarily because LLMs often ‘hallucinate,’ meaning they generate false or misleading information,” Hahn tells C&EN in an email. “However, elaborated techniques like retrieval augmented generation and the combination of genAI with knowledge graph technologies appear to be effective countermeasures by giving AI models credible sources to draw from. In this sense, LLMs could be utilized in hypothesis generation and testing when applied with human creativity—making AI an ideal companion in the lab.”

Arman Zaribafiyan

Head of strategic alliances, AI simulation, SandboxAQ

Zaribafiyan’s company, Good Chemistry, which he founded in 2021, was acquired last year by the Alphabet spinoff SandboxAQ, a major player in AI and quantum computing services. He thinks 2025 will be the year of improved efficiency in chemistry. Zaribafiyan expects a rise in chemistry-specific generative AI and upgraded models that could facilitate the design of new compounds, identify better reaction pathways, and predict the properties of materials.

“These models would be predominantly trained on large amounts of high-accuracy simulated data as opposed to LLMs [large language models] trained on a large corpus of text. Therefore, the ability to perform high-accuracy physics-based simulations at massive scales will be the critical requirement for accumulating sufficient high-quality training data,” Zaribafiyan says in an email.

Matthias Troyer

Corporate vice president of quantum, Microsoft

Last year brought breakthroughs in quantum computing, and 2025 will see a continuation of that trend, Troyer says in an email. “In April 2024, we entered the era of reliable quantum computing, with an increased pace in innovation leading up to the ability to detect errors, correct the errors, and perform computation on 28 logical qubits. This is a very exciting time for our field as the convergence of AI, high-performance computing [HPC], and quantum computing accelerates scientific discovery.”

This year, organizations will invest in skills and infrastructure to prepare for quantum computing’s coming scientific advantage over classical computing, Troyer says in the email. “In 2025, with the integration of HPC, AI, and resilient quantum computing capabilities, we will see new ways to efficiently model molecules with improved levels of accuracy, equipping researchers with more reliable computational insights and driving design of new materials and chemicals.”

Ashish Kulkarni

CEO, Kebotix

Kulkarni is energized and excited about the convergence of language models, quantitative models, and robotic tools. It could change several aspects of the chemical and materials value chain, Kulkarni tells C&EN in an email. “This convergence is attracting new talent and funds into the space. Combining all of this with the industry’s current state of affairs could lead to unprecedented innovation and productivity for the industry.”