A team at New Jersey Institute of Technology has used generative AI to design five new porous materials that could transform energy storage and reduce reliance on lithium.
AI-powered discovery: Researchers have used generative AI to identify five new porous materials that could outperform lithium-ion batteries.,Beyond lithium: The new materials could enable magnesium, calcium, aluminium, and zinc batteries, reducing dependence on scarce lithium supplies.,Faster innovation: The AI approach shortens material discovery from years to hours, with implications for clean energy and electronics.
The race to move beyond lithium
The AI discovers new battery materials story begins with an industry problem. Lithium-ion batteries, the workhorses of electric vehicles and consumer electronics, face supply bottlenecks and environmental costs. Mining lithium is water-intensive, geographically concentrated, and politically sensitive.
Alternatives exist. Metals such as magnesium, calcium, aluminium, and zinc are far more abundant and less environmentally damaging to extract. Yet they bring a formidable challenge: their ions carry two or three positive charges, compared with lithium’s single charge. This makes them capable of storing more energy, but also harder to stabilise inside a battery.
It is here that AI enters the laboratory.
How AI accelerates material discovery
At the New Jersey Institute of Technology (NJIT), Professor Dibakar Datta’s team combined two strands of artificial intelligence into a novel discovery engine.
The first, a Crystal Diffusion Variational Autoencoder (CDVAE), generated thousands of new crystal structures with potential for hosting bulky multivalent ions. The second, a finely tuned Large Language Model (LLM), sifted these candidates for stability, eliminating materials unlikely to hold up in real-world synthesis.
As Professor Datta put it:
“One of the biggest hurdles wasn’t a lack of promising battery chemistries — it was the sheer impossibility of testing millions of material combinations. We turned to generative AI as a fast, systematic way to sift through that vast landscape and spot the few structures that could truly make multivalent batteries practical.”
“One of the biggest hurdles wasn’t a lack of promising battery chemistries — it was the sheer impossibility of testing millions of material combinations. We turned to generative AI as a fast, systematic way to sift through that vast landscape and spot the few structures that could truly make multivalent batteries practical.”
The combination of CDVAE and LLM meant the AI could roam through vast design space, then bring the shortlist back to physics.
Five porous structures with promise
The outcome was the discovery of five entirely new porous transition metal oxide structures, each with large, open channels designed for moving multivalent ions efficiently.
The “porous” quality is crucial. Because magnesium, calcium, aluminium, and zinc ions are bigger and carry more charge, they often clog up inside denser materials. Porous frameworks allow them to flow more freely, making energy storage both quicker and safer.
NJIT validated the designs with quantum mechanical simulations, confirming their stability and suggesting that laboratory synthesis is achievable. These discoveries could eventually underpin batteries with higher capacity, longer life cycles, and lower costs.
Implications for Asia and beyond
The breakthrough matters globally, but Asia has a particular stake. China currently dominates the lithium supply chain, from mining to refining to cell manufacturing. Southeast Asian economies like Indonesia and the Philippines are also vying to capture more of the battery industry. If AI-designed alternatives reduce lithium dependency, the geopolitical and economic map of clean energy could shift. For more insights into how AI is affecting the region, read about APAC AI in 2026: 4 Trends You Need To Know.
For Japan and South Korea, both leaders in electronics and battery technology, the research opens new avenues for diversification. India, with its growing EV ambitions, could benefit from cheaper, locally sourced raw materials if magnesium or aluminium batteries take hold. This aligns with themes around the AI Wave Shifts to Global South.
The wider point is that AI now offers a shortened innovation cycle. What used to take years of laboratory trial-and-error can now be compressed into hours of computational exploration. This approach is not limited to batteries; it could extend to semiconductors, catalysts, and medical materials. A recent study by the World Economic Forum highlights the transformative potential of AI in material science.
From concept to commercial reality
Of course, discovery is only step one. The next challenge is scale: taking AI-designed materials from theoretical stability to manufacturable products. NJIT researchers are already planning collaborations with experimental labs to synthesise the new structures.
If successful, multivalent-ion batteries could address two pressing needs at once: energy security and environmental sustainability. And crucially, they would prove the principle that AI is not just analysing the world we know, but helping design the materials we do not yet have.
If AI can accelerate battery research from years to hours, what other industries in Asia might be transformed by the same approach? Pharmaceuticals? Construction? Food systems? The question is not whether AI can help, but where laboratories and businesses are willing to let it. The discussion around What Every Worker Needs to Answer: What Is Your Non-Machine Premium? becomes increasingly relevant in this rapidly evolving landscape.
Latest Comments (6)
interesting to see NJIT applying generative AI this way. we're seeing similar discussions in korea about how AI can accelerate materials science, especially with our national push for self-sufficiency in key industrial components. it's a critical area for policy makers in asia to consider for future competitiveness.
years to hours" for material discovery, that's massive. reminds me of how much time we spend just checking data quality for our compliance models. AI is definitely a shortcut for certain stages, but getting it production-ready always takes longer than we expect.
@arjunm: interesting how they chained the CDVAE and then the LLM for filtering. makes sense for the use case. using a VAE for generating new crystal structures is actually pretty neat, especially when you're looking for porous materials. the stability filtering with the LLM is probably where the real time-saving kicks in, since chemical synthesis is such a long feedback loop. i wonder what kind of embeddings they used for the crystal structures in the LLM.
While the acceleration of discovery is certainly impressive, I'd be keen to understand the datasets used to train the LLM for sifting candidates. The reliability and ethical sourcing of training data, a key focus for organisations like the UK AI Safety Institute, will be crucial if these methods are to be widely adopted in critical infrastructure like energy.
@chenming: This focus on multivalent ions like magnesium and zinc is really interesting. We're seeing similar research here in China, particularly with institutions exploring non-lithium battery chemistries for grid storage. The stability challenge Professor Datta mentions is definitely key.
If we can use AI to find new battery materials this fast, imagine what it could do for optimizing logistics routes or warehouse management in places like Indonesia. That would be huge.
Leave a Comment