Xiaomi has restructured its AI around the MiMo base model, splitting the technology behind its decade-old voice assistant XiaoAI into three pillars: the MiMo base-model team led by Luo Fuli, the cloud engineering team led by Luan Jian, and the on-device OS capability pushed down into the phone and car teams. The move signals that Xiaomi now treats a self-developed base model as core infrastructure, not a line item.
The logic is the agent shift. For years phone makers monetised through app stores, browsers, pre-installs and ads, positions in the path between user intent and service. AI agents threaten that path: a user who simply tells an agent what they want skips the store, the browser and the feed. Whoever owns the agent owns the entry point, and the distribution of value that flows through it.
Why build instead of rent
There is a real case against it. Base models are brutally expensive, training, inference, talent, compute, and a hardware company’s cash flow rarely supports an internet-scale spending war. Plenty inside the industry argue phone makers should just do on-device polish and plug in external models. Xiaomi is betting the opposite, that without its own agent it becomes a hardware shell under someone else’s intelligence, with external models deciding what service to show, which platform to call, how to rank and what to charge.
MiMo will not be the only model inside Xiaomi, product teams can still pick by cost and latency, but it must reach the core scenarios: phone, car, home, wearables and robots. The org change exists to give every terminal a shared understanding base while each business schedules it for its own context.
The closed-loop tax
The cost is real coordination. A base model only becomes true infrastructure when business teams feed back scenario data, user signals and real call chains. That is the loop that justifies the burn. If it forms, Xiaomi keeps its “human-car-home” ecosystem intact at the intelligence layer; if it does not, MiMo becomes an expensive project drifting from the business. Either way, the wager is strategic: pay now or risk losing the entry point later.
Read the original report (LeiPhone)
Translated and adapted from LeiPhone (leiphone.com).