SES AI, NVIDIA, Crusoe, and Supermicro are using AI and supercomputers to discover new battery materials

SES AI has announced a project to accelerate material discovery for EV batteries using artificial intelligence and supercomputers.

Optimizing electrolytes is critical to unlocking the full potential of lithium-metal batteries. Over the past 30 years, the battery industry has studied roughly 1,000 (10^3) organic molecules and achieved a coulombic efficiency of over 99.6%. However, there are an estimated 10^12 organic molecules under 20 atoms and 10^60 organic molecules under 30 atoms, indicating vast untapped potential for discovering new molecules that could revolutionize battery technology and have significant implications for all of science.

Harnessing the Power of AI:

The 'AI for Science' initiative begins with NVIDIA HGX H100 GPU clusters to generate a comprehensive database, predicting the electronic structures and properties of each molecule. This database will be the foundation for the construction, pre-training, and fine-tuning of physics-informed machine learning (ML) models leveraging Crusoe Cloud H100 virtualized instances built with NVIDIA HGX technology. Expanding upon the capabilities of the foundation model, SES AI will next construct, train, and fine-tune an advanced generative model. These efforts are uniquely supported by Crusoe’s sustainable computing infrastructure, helping ensure that the intensive computational tasks are carried out efficiently and with minimal environmental impact.

The final phase will involve the creation of a multimodal large language model (LLM) and AI agent, training a 70-billion-parameter Llama 3 model on approximately 200 billion tokens from various battery-related texts and literature. This is enabled by the NVIDIA accelerated computing platform and Supermicro’s high-performance hardware, significantly broadening the scope and accuracy of SES AI’s predictive capabilities. The most promising candidates from this AI-driven analysis will undergo rigorous laboratory testing in SES AI’s Electrolyte Foundry, where the candidates will be synthesized, evaluated in battery cells, and assessed for real-world performance.

Powering AI for Science with Crusoe:

Crusoe® Cloud is a purpose-built AI-cloud platform that offers the latest NVIDIA GPUs, high-performance networking, and storage solutions. Crusoe powers its cloud platform by developing, building, and utilizing a differentiated portfolio of clean energy solutions to minimize the environmental impacts of both energy production and computing.

This project will leverage Crusoe’s climate-aligned cloud platform, powered by NVIDIA and Supermicro.

SES AI’s AI-driven initiatives include:

+ AI for Manufacturing: Analysing manufacturing data to ensure quality and safety.

+ AI for Safety: Predicting battery incidents to ensure nearly 100% battery safety in the field.

+ AI for Science: Mapping the molecular universe to discover next-generation materials.