On the eve of AWE2026 in Shanghai, Midea Group unveiled MevoX, a self‑evolving home intelligence agent, and a “one network, one brain, one open platform” strategy intended to knit its appliances into a single cognitive system. The company said it will invest more than RMB 60 billion over the next three years in AI, embodied intelligence and new energy—a clear signal that its core white‑goods and HVAC businesses are shifting from hardware volume to software‑driven services.
Midea casts MevoX as a leap beyond today’s device‑centric smart homes. Rather than merely connecting appliances for remote control, MevoX is designed to build persistent representations of users and spaces, to reason about ambiguous commands and to remember long‑term preferences so the home can act proactively. That is a different product ambition: from an app that toggles devices to a spatial system that understands intent and context.
At the launch, Xu Yi, head of Midea’s AI Research Institute, identified two technical shortfalls that most commercial smart‑home systems still face: inference and memory. He illustrated the point with a common failure mode—when a user asks a system to “turn off all the air conditioners except the study,” many current platforms will misidentify scope and disable the wrong unit. Effective inference requires mapping natural language to a household topology and device state; durable memory requires storing user preferences and patterning behaviour over weeks, months or years.
Midea says MevoX will address those gaps as part of a six‑capability architecture—connect, sense, reason, execute, remember and optimise. Making that work at scale combines mature ingredients (sensors, voice, vision, networks) with still‑hard problems in model composition: small, local models for latency and privacy; cross‑device planners; and lifelong learning systems that accumulate and generalise from household data without degrading performance.
The product strategy has a clear commercial logic. Midea sits on a vast installed base and distribution network that gives it leverage many pure‑software entrants lack. By turning refrigerators, washing machines and HVAC units into nodes of a shared cognitive platform, Midea can seek higher‑margin recurring revenue from services, maintenance and data‑enabled features. It also intensifies competition with global platform incumbents—Amazon and Google—and other Chinese ecosystem players already experimenting with device OSes and cloud‑edge stacks.
Robots are flagged as a likely future vector for full‑house intelligence: mobile manipulators could become the physical embodiment of MevoX’s cognition. Yet Midea’s executives were candid that humanoid or household robots remain constrained by manipulation capabilities. Current demonstrations emphasise mobility; the harder problem for home use is reliable, fine‑grained manipulation and safe autonomous operation in cluttered, human environments.
The path to adoption will run through thorny trade‑offs. Personalisation demands data retention and profiling; consumer trust hinges on transparent controls, strong security and local‑first processing to limit exposure. Interoperability is another obstacle: a proprietary, closed platform can lock users but risks fragmentation in an industry that benefits from common standards. Regulators will scrutinise how household and biometric data are collected, stored and shared.
Midea’s RMB 60 billion commitment elevates the stakes for the smart‑home sector. If MevoX succeeds, it could speed a shift from discrete connected devices to environments that anticipate needs, and it would deepen the business model transition of appliance makers from discrete sales to platform services. Observers should watch how Midea balances openness with ecosystem control, how it deploys edge versus cloud reasoning, and whether robotics delivers the practical utility that justifies the hype.