The company’s technology is already used by some firms to train AI models without needing to pool compute resources or data. AI companies currently build their models by combining vast amounts of training data with huge quantities of compute concentrated inside data centers stuffed with advanced GPUs that are networked together using superfast fiber-optic cables. Even open source models, like Meta’s Llama and R1 from DeepSeek, are built by companies with access to large data centers. Or it could allow countries that lack conventional infrastructure to network together several data centers to build a more powerful model. Divide and ConquerDistributed AI training involves rethinking the way calculations used to build powerful AI systems are divided up.