By Allison Proffitt
Anthro Energy is betting that an internally built AI materials discovery platform will let it scale faster than the rest of the battery industry.
The Alameda, California-based battery materials company has announced Anthro Atlas, its internal AI-enabled machine learning platform, designed to compress electrolyte development timeline. Company co-founder and CTO, Joe Papp, explained that batteries aren’t one-size-fits-all; a battery for a car, a watch, and a drone each demand different performance profiles. In a conventional lab workflow, tuning a formulation for a specific application can take six months or more, largely because realistic cycling and performance data takes a long time to collect.
Atlas is built around a four-step loop: material and component discovery, material and formulation property prediction, cell and battery performance prediction, and final validation testing. In the discovery phase, Papp said the platform uses physics-based modeling alongside newer machine learning techniques to sift through millions to billions of candidate components. In the prediction phases, Atlas can estimate properties such as viscosity and conductivity across hundreds or thousands of material combinations and can generate cell and battery performance predictions in a matter of hours, a process Papp said used to take roughly six months.
The performance-prediction models are trained largely on five years of Anthro’s own battery testing data, supplemented by public materials datasets. Papp said literature scanning, once useful, has become a comparatively minor input as internal and partner-generated datasets have grown. In some cases, Anthro is co-developing datasets directly with customers, which Papp said helps both sides — improving Atlas’s predictions while giving customers a better sense of what’s achievable in their own device designs.
On the modeling side, Papp said the team began with a classic compositional model commonly used to analyze mixtures of materials and extended it into a more sophisticated system capable of modeling formulations with dozens of components simultaneously. Anthro runs its fine-tuning largely on on-premises hardware, which Papp said the company has found more cost-effective than comparable cloud resources, and has completed hundreds of thousands of fine-tuning runs across different models for properties like conductivity and capacity.
Scaling vs R&D
Asked how a company pursuing mega-scale manufacturing manages constant formulation changes typical of an R&D-driven business, Papp said the two aren’t as much in conflict as they might appear. Electrolyte suppliers commonly offer customers dozens of formulations even in conventional chemistries, he said, and Anthro’s newest electrolyte manufacturing facility in Louisville, Kentucky, is being designed with multiple parallel production lines at varying scale to accommodate that. Atlas, in his view, is what allows the company to generate and filter through a much larger set of candidate formulations in the lab before narrowing down to the smaller set with real commercial viability worth scaling.
Much of that filtering is driven directly by customers. Anthro has formal collaborations across the industrial and consumer electronics space, according to Papp, with partners providing specific constraints around safety, cycle life, rate performance, and cell swelling. One recurring theme, he said, is AI at the edge. As more power-hungry AI functionality moves into phones, glasses, and other wearables, device makers are increasingly constrained by the physical size of conventional batteries and are looking to Anthro to help fit more capability into smaller spaces.
Keeping the Technology In-House
Some battery companies have shifted their business models toward licensing AI-driven materials-discovery tools to other manufacturers rather than making batteries themselves. Papp said that isn’t the direction Anthro is headed. The company is building Atlas for internal use, he said, and remains focused on scaling its own battery product rather than becoming an AI vendor to the rest of the industry. “We’re being greedy,” he said. “We’re keeping our knowledge in-house.”
That doesn’t mean Papp wants Atlas to stay in the background, however. Asked whether the goal was for Atlas to eventually fade into the background the way batteries themselves ideally would for end users, Papp pushed back. Atlas, he said, is meant to become more capable over time, not less visible, continually feeding new data back into itself to help the company develop and scale faster.
Papp said Atlas has already contributed to two commercial programs over the past six months, though he declined to name the customers, including one where Anthro demonstrated improved rate capability alongside high energy density and low cell swelling. Both programs are expected to lead to commercial products within the next year.
Validation Evolution
For now, Anthro still runs real-world validation cycles before finalizing products for customers. Validation typically takes a few weeks of testing on top of model predictions, even though full validation for a battery’s real lifespan can take years to observe directly. Papp said the company’s goal over the next one-to-two years is to build enough confidence in Atlas’s predictions that experimental validation eventually becomes optional rather than required. Though long-duration cycling data, including batteries the company has now tracked for three to four years, will remain a particularly valuable input for improving the models further.
