More on-device AI. Shift from off-prem compute to on-prem. Open source LLMs run locally, no more black boxes.