Google's DeepMind to train AI on player actions in quarter-million-player MMORPG Eve Online

Google's DeepMind plans to leverage data from 'EVE Online,' a complex sci-fi MMORPG with 250,000 monthly active users, to train its artificial intelligence. Known for its deep economic, political, and... Read original →

DeepMind’s move into EVE Online isn’t about game AI—it’s a strategic bet on training general-purpose agents in high-dimensional, adversarial socio-economic simulations. This will spike demand for edge inference chips optimized for sparse, long-horizon reasoning, favoring RISC-V and in-memory computing architectures. Under the EU AI Act, Fenris faces significant GDPR exposure if player behavior data lacks explicit consent, risking operational cost surges. Microsoft and NVIDIA, already entrenched via Minecraft and Omniverse, will likely accelerate acquisitions of sandbox IPs to secure competitive training environments. Within 18 months, game engines will morph into AI training infrastructures; semiconductor firms failing to hardwire dynamic sparsity support will fall behind in next-gen AI accelerator races.