Why electrolyte design is so complex

An electrolyte in a battery is not a simple chemical. It is a complex mixture of salts, solvents, and additives that interact with each other, respond to temperature, and must meet often conflicting requirements — high conductivity, chemical stability, low viscosity, and safety. Just the potential molecules for electrolytes alone number approximately 10⁶⁰ — that's a number with sixty zeros, exceeding the number of stars in the observable universe.

When you add the practically infinite number of ways to combine these substances in different ratios and concentrations, it becomes clear that a human researcher has no chance of searching this space in an entire lifetime. This is exactly where generative artificial intelligence comes in.

ElectrolyteGPT: AI that understands the entire mixture

A team from the UChicago Pritzker School of Molecular Engineering led by Professor Chibueze Amanchukwu developed a model called ElectrolyteGPT. Unlike most existing AI approaches that only select suitable individual molecules, ElectrolyteGPT generates complete electrolyte formulations — it determines not only which substances to use, but also their exact concentrations, mutual ratios, and other mixture parameters.

"Next-generation battery electrolytes must meet a number of often conflicting property requirements," explains Jaemin Kim, the study's first author. "Thanks to the model's ability to generate outputs under various conditions, ElectrolyteGPT can propose new candidates that meet all desired properties simultaneously."

The research was published in the prestigious journal JACS Au and builds on an earlier project that received a Google Research Scholar Award of $60,000 in 2024.

fLine: A new chemical language for mixtures

The key technical innovation of the entire project is the creation of a new chemical notation called fLine (derived from the standard SMILES format). While traditional chemical languages describe only the structure of a single molecule, fLine can capture an entire mixture — including solvent ratios, salt concentrations, temperature, and other variables.

To give an idea: common table salt is written in SMILES as [Na+].[Cl-]. FLine goes a step further and adds information like "how much salt in which solvent at what temperature" to the notation. Thanks to this language, AI for the first time "understands" the electrolyte as a whole, not just its individual components.

Professor Amanchukwu emphasizes that fLine isn't only useful for electrolytes: "It's beneficial for mixtures in general. Now you can generate a complete formulation with multiple different salts and solvents in various concentrations and ratios."

What the AI actually achieved

This isn't just a theoretical concept. The team synthesized several formulations recommended by the ElectrolyteGPT model, and the results were encouraging — the new electrolytes achieved performance comparable to top-tier commercial alternatives in lithium metal batteries. This is significant because lithium metal batteries are considered one of the most promising energy storage technologies with significantly higher density than today's lithium-ion cells.

"We had a range of formulations that performed at the current state-of-the-art level, and that was exciting for us," comments Amanchukwu. "At the same time, there's still a lot of work ahead — the goal is to find electrolytes that surpass the current best solutions."

AI in the service of materials science: Broader context

ElectrolyteGPT is not an isolated project. The use of AI for discovering new materials has been booming in recent years. DeepMind (Google) introduced the GNoME tool, which discovered over 2.2 million new crystal structures. Microsoft developed MatterGen for designing materials with desired properties. In March 2026, scientists published a study in Nature Communications where they used deep active learning to triple the lifespan of lithium batteries — after just three iterations of model learning.

What sets ElectrolyteGPT apart from these approaches is precisely its ability to work with entire mixtures, not just individual materials. Most AI models developed for molecular discovery were also originally trained for drug discovery, not battery materials. "When you use what is available in the literature, it generates drug-like molecules. That's not relevant to us," explains Amanchukwu. His team therefore created their own dataset specific to electrolytes.

What this means for the average user and the European market

The practical impact is clear: faster development of better batteries. Everyone who uses a mobile phone, laptop, or electric vehicle benefits from advances in energy storage. AI can shorten the development cycle of new battery technologies from decades to months.

For Europe, this topic is particularly timely. The European Union is massively investing in the battery industry through the European Battery Alliance, and gigafactories are emerging across the continent. According to the new EU Battery Regulation (EU Battery Regulation 2023/1542), batteries must be equipped with a digital passport by 2027 and meet strict environmental standards — AI tools like ElectrolyteGPT can help European manufacturers develop batteries that meet these standards faster.

In the Czech Republic, several research groups focused on electrochemistry and materials research are active — for example at the University of Chemistry and Technology in Prague or at CEITEC in Brno. Although ElectrolyteGPT is not yet publicly available as a commercial tool, similar AI methods are likely to quickly spread to the academic sphere as well.