ElectrolyteGPT: UChicago's AI that writes battery chemistry from scratch
Researchers at the University of Chicago have built an AI model that doesn't just suggest battery ingredients — it writes the entire chemical recipe. Called ElectrolyteGPT, the tool generates complete electrolyte formulas, including concentrations, mixing ratios, and predicted performance, and has already produced results in the lab that rival commercial lithium-metal batteries. The development matters because electrolytes are the single most complex variable holding back the next generation of longer-lasting, faster-charging batteries for phones, EVs, and grid storage.
The problem it solves
Electrolytes — the liquid or gel mixture that carries charge between a battery's electrodes — have been a 30-year bottleneck. The theoretical number of molecules that could form a viable electrolyte is 10 to the 60th power, according to Chemistry of Materials (2025). Traditional materials science attacks that space through manual trial and error, a process that routinely takes years per candidate formulation. Previous AI tools could screen individual components but still left researchers to assemble the full recipe by hand.
ElectrolyteGPT, developed at the Pritzker School of Molecular Engineering under Professor Chibueze Amanchukwu, was trained on 250 electrolyte research papers spanning the history of lithium-ion battery science. It outputs a full formulation — not a shortlist of molecules — and scores candidates using a metric called eScore that weighs ionic conductivity, oxidative stability, and Coulombic efficiency simultaneously.
A bias problem, then a fix
Early versions of the model had a significant flaw: most large language models are trained on data where pharmaceutical chemistry dominates. The system kept proposing molecules suited to drug delivery, not energy storage. The team fixed this by restricting the training data strictly to electrochemistry literature. After that correction, the model began generating chemically stable compounds relevant to real battery applications.
Every formula the AI produces is still reviewed by chemists and physically tested in the lab — the model accelerates the search, it doesn't replace the scientist. Lab-synthesized formulas have already shown performance comparable to the best electrolytes currently on the market, per UChicago PME (2024).
What comes next
Amanchukwu, who received the 2024 Camille Dreyfus Teacher-Scholar Award and a $60,000 Google Research Scholar grant, frames the work as an early demonstration rather than a finished product. No commercial partnerships or licensing deals have been announced. But with lab-validated results now published in a peer-reviewed journal, the path toward faster EV battery development and grid-scale energy storage looks a little shorter than it did before.
