Aligning realities: Bringing AI into LNMO batteries

 In this series, we examine the potential of AI in material science across three areas of Topsoe’s research and development: batteries, electrochemistry, and heterogeneous catalysis. Drawing on insights from leading AI-for-science startups and Topsoe scientists, we explore AI’s potential to redefine energy – and the very materials it depends on.

Whether replacing combustion engines in cars or offering scalable solutions for energy storage, rechargeable batteries have long been central to answering the question: “How will we build reliable green energy systems?”

The answer, of course, is varied – it depends on energy diversification, infrastructure, different methods of production, and efficient ways of storing and transporting energy – but batteries have remained at the crux of this answer for decades. Because of their importance, how we design and manufacture batteries is crucial to the future of decarbonization – and the same goes for the materials we select to make them.

A battery’s material composition directly affects its voltage, lifecycle, durability, efficiency, and cost – all central factors in determining its effectiveness and potential scalability in future energy systems. Topsoe has been innovating in battery materials for 15 years, developing next-generation, cobalt-free, high-manganese LNMO cathode material that addresses key issues related to battery cost and the raw material supply chain.

In this article, we discuss potential applications of AI in batteries with Topsoe's R&D Director Jonathan Højberg, and the people leading strategy at the AI startup Entalpic, Lucile Ritchie and Indira Fabre. Together, we explore the current bottlenecks within material development and testing, their impact on the reliable scaling of battery technology, and the role artificial intelligence may play in leapfrogging these constraints.

Topsoe’s development of LNMO batteries

As part of its broader commitment to energy diversification and the energy transition, Topsoe has been developing new battery materials for over a decade. Led by R&D Director Jonathan Højberg, Topsoe Battery Materials is a dedicated group within Topsoe focused on the development, refinement, and scaling of Lithium Nickel Manganese Oxide (LNMO) batteries, a multipurpose, electrode-active material optimized for high energy density and low cost.

Topsoe’s battery development has focused both on material optimization and innovation and on operationalizing commercial-ready production. In material innovation, Topsoe has developed cathode active materials with high phase purity – meaning the microscopic framework of the material is ordered to directly match the target chemical formula – and superior morphology – meaning particle size and structure are optimized and uniform – enabling easy processing and high electrode energy density. These material characteristics help maximize energy capacity and voltage stability while extending overall battery lifetime.

In production, a key focus for Topsoe’s battery team has been demonstrating production processes for LNMO battery materials on an industrial scale. In 2024, Topsoe inaugurated a large-scale industrial pilot in Frederikssund with a design capacity of 200 tons/year. This demonstration plant serves two essential purposes on the path to industrializing LNMO battery materials. First, it enables production at a scale required for customer qualification, confirming that the material solves customer challenges in commercial cell formats and helping drive customer commitment. Second, the plant in Frederikssund serves as a testing ground to refine equipment and optimize material production processes for future full-scale production.

Material and testing development in battery technology

Topsoe’s work in battery materials has always been founded on material innovation and optimization. To help build a future that relies on lithium-ion rechargeable batteries, Topsoe has developed LNMO cathode active material that enables water-based electrode preparation, uses more abundant materials such as manganese, has low nickel content, and is entirely cobalt-free. This makes LNMO a long-term and effective way to reduce the use of critical and hazardous materials in battery supply chains – and an essential step in scaling future-proof energy infrastructure.

However, the path to creating and validating these material innovations is anything but straightforward.

“Developing new materials is, put simply, a process of understanding how the changes we apply translate into a difference in performance in the battery cell,” explains Højberg. But the connections between these causes and effects are not always transparent. Depending on the quality of the testing available, drawing these connections can in fact be quite difficult.

Some variables, such as energy density, are relatively straightforward to test. Battery lifetime, however, depends heavily on having commercially relevant electrolytes, anodes, and other cell components built in the right format and tested in the right way. According to Højberg, developing more commercially relevant testing methods at Topsoe has been a 15-year process in the making.

“For the first seven years of Topsoe’s LNMO battery development, testing was quite limited. So, we did the simple things: optimizing energy density, phase purity, and testing variables that are easy to measure. Over time, we developed more exacting tests that expanded the range of variables we can test reliably. Now, with more commercially relevant testing, we can focus more on optimizations that improve the lifetime of the cell.”

The testing conundrum

Creating commercially relevant testing for battery components relies on a variety of factors. One, for example, is the scale of the battery cell. A typical commercial battery cell measures around 100 amp-hours (Ah), and in Topsoe’s battery testing Højberg and his team are testing cells at 1 Ah. While 100 times smaller, these cells are still close enough to replicate variables relevant to a commercial-scale cell. “Previously, we were testing at 1 milliamp-hour (mAh), which is 1,000 times lower capacity. At this size, there are just some things you do not see – gas evolution, temperature variation, etc.”

Another factor in commercial testing is making electrodes that approximate those used in commercial cells. “By doing this, we are testing the right anodes, the separators, the electrolyte – all of these things. Over the past four years, we have really grown the people and competences within these areas to make sure that the choices we make are as relevant as possible and that we are developing the most exacting tests to validate the highest-quality materials.”

Even with these advances in material testing, Højberg believes artificial intelligence may provide effective tools to further this progress.

“Commercially relevant tests are all about measuring improvement across a multitude of extremely detailed parameters. This is an immensely time-consuming process. It takes, on average, one week to synthesize at scale, one week to make the cell, and then around three months to run the test before getting the results. When you want to do a lot of iterations and optimizations, this creates an incredibly slow feedback loop. It is possible to speed up the optimization loop if you can use your understanding of reaction mechanisms to simplify the problem to a few parameters that can be measured with a reliable and, ideally, easy test. The same logic applies to the use of AI, and in our discussions with Entalpic, it has been possible to reduce the problem in a way that allows both AI modelling and reliable experimental validation.”

AI’s potential in material development for batteries

To optimize this feedback loop with simpler, faster tests, Højberg and his team have started exploring solutions with Entalpic, a startup based at Paris’s Station F that works with AI, atomistic simulations, and experimental feedback loops to design new materials and processes that drive the energy transition.

“From a technical and scientific perspective, our focus is on surface engineering at the atomic scale,” explains Lucile Ritchie, Director of Strategy and Operations at Entalpic. “Two of our co-founders did their PhDs on AI for materials and had experience with some of the first models applied to crystals, such as GFlowNets. Entalpic is built on this scientific knowledge and on our collective desire to make a positive impact on industry and the environment.”

Entalpic’s AI-based platform uses computational chemistry, reactor-scale physics, and multi-fidelity, multimodal data to solve materials problems. Through a mixture of generative AI models, molecular generative models, and Machine Learning Interatomic Potentials (MLIPs), Entalpic’s system explores millions of potential molecular structures, predicts key functional properties, and tests potential molecular candidates through digital twin models.

“The main bottleneck we are trying to solve is how to go beyond an incremental trial-and-error process, which is the classical way of making experimental discoveries,” explains Indira Fabre, Entalpic’s Strategic Program Lead. “This is where AI can be really interesting: it allows you to do high-throughput screening and testing. This way, you can screen millions of candidates and then only focus on the most promising ones for testing. It also creates opportunities to be more creative – synthesizing new materials and combinations that we have not even imagined yet.”

The Magic of MLIPs

A particularly fascinating part of Entalpic’s process is its ability to test surface reactivity through Machine Learning Interatomic Potentials (MLIPs).

Using this type of machine learning, Entalpic digitally tests the surface reactions of molecular candidates at an atomic scale. By running these tests, Entalpic’s system not only develops stronger candidates with a higher likelihood of success in real-world reactors but also accompanies these candidates with data documenting properties such as reaction kinetics and adsorption energies that are essential in large-scale screening.

For example, one case that Topsoe and Entalpic are currently working on involves applying these models to test the reactivity of different types of LNMO surfaces with electrolytes and to investigate the impact of different dopants in these reactions. By slightly altering the surface reactivity of an LNMO with different dopants, the types of reactions that consume electrolytes, impede lithium-ion movement or create gas formations that accelerate battery ageing can be mitigated and solved-for. By mass-testing different combinations of LNMO surfaces, electrolytes and dopants, Topsoe scientists can develop a more refined understanding of how to achieve optimal surface reactions in LNMO batteries.

Making AI testing industrially relevant

While AI systems open up enormous potential, there are still challenges in turning that potential into real, reproducible results. “Right now, the hardest part is bridging the simulation and reality,” Fabre explains. “We can simulate materials that look really good on paper, but the real difficulty is having materials that we can actually make and synthesize.”

She continues: “Also during generation, we simplify the models around these materials so we can have a model that can be applied at high throughput. This means that there are a lot of things that are not modelled during the screening, and it implies a difference between the actual conditions these materials will be tested under and the models in which we simulate them.”

Closing the gap between digitally modelled scenarios and the reality of in-lab testing is essential if AI is to deliver on its promise in material discovery. “There’s a lot of in silico calculation where you get some nice graphs that look amazing and you can argue that you have calculated millions of different compounds,” says Højberg, “but in the end, it is worth nothing if it cannot be translated into reality.”

According to Entalpic, aligning these realities comes down to two main factors: investing in sufficient computing power and integrating industrial constraints into the models. First, the computational power required to run these programs and simulate atomic-scale chemistry is intensive and costly. Second, even where AI models can effectively identify promising material candidates, it is not enough to produce materials that succeed only in lab testing – they must also succeed under full-scale industrial conditions.

“Even if you have simulated and predicted a good material and have been able to validate it experimentally in a small lab, the last step is to go to an industrial scale,” explains Fabre. “This brings in a lot of other elements to consider, and today there is a challenge in integrating those industrial constraints into the discovery loop. This is, of course, a very important consideration for a company like Topsoe.”

Topsoe and Entalpic share priorities in generating industrially validated data, accelerating high-throughput testing, and discovering new materials—creating a strong foundation for collaboration. Entalpic can build on Topsoe’s data and experience, while Topsoe can explore how AI could support its materials innovation

The priorities of sourcing industrially validated data, accelerating high-throughput testing and material discovery are highly compatible and form an excellent foundation for companies like Topsoe and Entalpic to work together. Entalpic’s systems can benefit from Topsoe’s data and learnings, and Topsoe can benefit from exploring the potential of AI in its own material innovation.

Final thoughts

The path to scalable, sustainable battery technology depends not only on discovering better materials, but also on developing faster, more commercially relevant ways to test and industrialize them. The introduction of systems like Entalpic’s into the bottlenecks found in material testing and experimentation opens up an entirely new field of possibilities and could drastically improve the way battery materials are discovered and tested. “Our system is for those who want to explore broader chemical spaces and go faster,” says Fabre, and in the field of rechargeable batteries, Topsoe wants to do both at scale.

Through test cases like the one Entalpic and Topsoe are currently exploring together, the potential for deploying AI in meaningful, precise, and impactful ways continues to expand. When grounded in manufacturability, lab validation, and industrial reality, AI systems are poised to make a measurable difference in the future of battery materials and testing.