Since its 2010 founding, DeepMind has used games to drive AI breakthroughs. Founder Demis Hassabis, a former game developer, instilled gaming in the company's DNA. Early work on AlphaGo mastering Go and AlphaStar reaching Grandmaster in StarCraft II demonstrated deep reinforcement learning, later applied to AlphaFold's Nobel Prize-winning protein prediction.

Now DeepMind focuses on generalist agents like SIMA (Scalable Instructable Multiworld Agent), which sees screens, understands natural language, and acts via keyboard/mouse — no APIs needed. Powered by Gemini, SIMA 2 achieves human-like play across games like No Man's Sky and Valheim, enabling AI companions and robust QA testing without code modifications.

A major new partnership with Fenris Creations, creators of EVE Online, explores persistent worlds. EVE's 20-year player-driven economy and diplomacy offer challenges in continual learning, long-horizon planning, and multi-agent dynamics. The partnership extends to EVE Vanguard and EVE Frontier, studying agents across tactical to strategic levels in evolving environments.

DeepMind emphasizes "show, don't tell" prototyping with studios like Hello Games and Coffee Stain Studios to discover fun, AI-enabled gameplay impossible without frontier models.