Key takeaway
Xiaomi's current shift in AI investment is essentially the company completing a "second heavy bet" that is similar to its 2021 venture into car-making but at a higher level — upgrading from betting on a new terminal category to betting on nextgeneration system-level capabilities. During the car-making phase, the market awarded Xiaomi a higher valuation primarily because it added a major track and a second growth curve. In the AI phase, if the market truly buys in, what changes is not just the valuation of a certain segment but the perception of the entire group's attributes. If the market recognizes that Xiaomi is not a hardware patchwork but a company capable of reorganizing the value of its terminal ecosystem through AI, then the traditional SOTP becomes incomplete and should shift to an "overall platform revaluation." AI is harder to price in the short term than car-making, but its long-term imagination space is larger. Xiaomi is a rare full-stack AI player that simultaneously possesses self-developed models, system permissions, terminal entry points, and manufacturing scenarios.
Thesis
Xiaomi's current shift in AI investment is essentially the company completing a "second heavy bet" that is similar to its 2021 venture into car-making but at a higher level — upgrading from betting on a new terminal category to betting on nextgeneration system-level capabilities. During the car-making period, Xiaomi used an initial RMB10bn and a ten-year commitment of USD10bn to secure the "car" heavy-asset entry point within the "Human x Car x Home" ecosystem. In this round of AI deployment, Xiaomi has clearly committed to investing RMB60bn over the next three years, RMB16bn in 2026 alone, and over RMB200bn in total R&D from 2026 to 2030. The goal is no longer just to add a business line but to compete for the unified intelligent hub, interaction entry point, and system definition rights for all "Human x Car x Home" terminals in the future.
Lei Jun, as the founder, remains on the front line with extremely high decision-making authority. The core significance of this is not just "boosting morale" but significantly reducing the three types of risks Xiaomi is most prone to during strategic transitions: directional wavering, organizational internal friction, and resource misallocation. If building cars validated the idea that "when Lei Jun personally takes charge, Xiaomi can succeed in a new heavy-asset business widely considered extremely difficult," then the AI phase will validate whether "Xiaomi can be pushed from a hardware ecosystem company to an AI platform company." The biggest difference between the two is: car manufacturing primarily changes the revenue structure, while AI could rewrite the valuation framework.
Why should we view Xiaomi's AI push alongside its earlier car-making venture? The significance goes beyond making a historical analogy; it is about answering three more fundamental questions. First, why does Xiaomi always pursue a heavy-investment strategic pivot when its core business appears not entirely worry-free; second, is Lei Jun's role in such pivots that of a brand founder or a substantive decision-making hub; third, compared with the car-making venture back then, is this round of AI merely a repetition of past methodology, or has it advanced further at the strategic level.
Foundation model: the trillion-parameter MiMo model enters the first tier of open-source models. The evolution path of Xiaomi's foundation large model is not centered on simply pursuing parameter expansion, but rather on continuously upgrading along the route of "starting with lightweight on-device deployment— expanding into a cloud-edge-device matrix—achieving breakthroughs in open-source reasoning foundations— building out an Agent and full-ecosystem base." 2023 was the period of technical project establishment and MiLM validation, 2024 was the period of MiLM2 productization, 2025 is the period of the MiMo open-source foundation leap, and 2026 will be the capability release period as MiMo-V2/V2.5 challenges for a global firsttier position. 2026: MiMo-V2 and V2.5 will propel Xiaomi into the global first tier. 2026 will be a year of concentrated capability release for Xiaomi's large model. In March 2026, Xiaomi officially released MiMo-V2- Pro, its flagship foundation model for the Agent era, with total parameters exceeding 1 trillion, 42 bil lion activated parameters, and support for a 1-million-token context window. Concurrently, MiMo-V2-Omni and MiMo-V2-TTS were also released, and Xiaomi's mobile Agent, Xiaomi miclaw, entered small-scale closed testing, beginning to validate the model's upgrade path from "conversational capability" to "system-level execution capability." Equally important is its cost-effectiveness and token usage efficiency: with comparable capabilities, it can reduce token consumption by approximately 40%-60% compared to Claude Opus 4.6, Gemini 3.1 Pro, and GPT-5.4. For enterprise-level call scenarios, it is a core variable that directly impacts the cost structure of commercialization.
The core of Xiaomi's 2025–2026 AI application deployment is to expand Agent from a task executor on mobile phones into a unified scheduling hub spanning four types of physical terminals: phones, cars, homes, and robots. The key shift in this process is that AI is no longer confined to a "Q&A-style entry point," but begins to acquire system permissions, cross-device memory, multimodal perception, device orchestration, and physical execution capabilities, thereby upgrading Xiaomi's "Human x Car x Home" full ecosystem from an interconnected network to an action network that can be driven in a unified manner.
In terms of deployment pace, Xiaomi has already established a relatively clear three-layer advancement path: the first layer is phone Agent integration , represented by Xiaomi miclaw, which validates the system-level execution loop; the second layer is car and home space Agent integration , which, through XLA/OneVL cognitive intelligent driving and Miloco whole-house smart solutions respectively, brings AI from screens into driving spaces and living spaces; the third layer is the embodied execution end , which, with the help of Xiaomi Robotics-0, TacRefineNet, and factory internship robots, ultimately extends AI into real industrial action execution.
From Xiaomi's perspective, the most realistic return path for AI may not necessarily be subscriptions, but first manifesting in product strength and organizational efficiency. In terms of business model, Xiaomi is not in a hurry to monetize AI separately at this stage. Currently, it is more focused on maturing models, applica tions, and ecosystem capabilities. Token plans, subscription models, premium pricing for high-end products, and API services are all under exploration, but none have been set as core KPIs yet.
AI is harder to price in the short term than car manufacturing, but its long-term imaginative potential is greater. The advantage of car manufacturing is higher visibility; sales volume, ASP, and gross margin can all be tracked. The difficulty with AI lies in the fact that a large amount of its value is hard to reflect in financial statements in the short term, making the pricing mismatch of "seeing the strategy but not the performance" more likely to occur. But for this very reason, once AI truly breaks through system-level entry points, subscriptions, and hardware premiums, its overall revaluation magnitude could be higher than that of a standalone automotive business. Xiaomi has become one of the few companies that simultaneously possesses self-developed models, system permissions, terminal entry points, and manufacturing scenarios a fullstack AI player.
Although the traditional core business is under pressure in the current macro cycle and memory prices are still rising rapidly, with the cost cycle inflection point likely after 2027, the smartphone gross margin is expected to be well controlled, thanks to earlier memory stockpiling, the company’s proactive product mix adjustments, and price increases for some products starting in April. In addition, the new automobile product cycle and production ramp-up could serve as catalysts. On the other hand, the company continues to invest heavily in AI, which will be one of its core competitive strengths over the longer term.
Revenue of Xiaomi Group is expected to reach RMB484.6bn/RMB607.7bn in 2026/2027 (YoY +6%/+25%). Adjusted net profit is expected to reach RMB28.6bn/RMB39.3bn (YoY -27%/+37%). We assign 20x PE to the company’s core businesses for 2026 and 2.5x PS to the automobile segment for 2026, with a target price of HKD38.7 and a “buy” rating.
Risks:
New product shipments falling short of expectations; downward risks in the smartphone market, such as a decline in overall shipments and weaker new product promotions; inventory impairment provision risks; policy management risks, including domestic restrictions on internet business development; increased competition in overseas markets; strong cyclical nature of IoT, with weak demand; macroeconomic downturn leading to decreased demand for 3C products; impact of energy shortages in Europe; exchange rate fluctuation risks; geopolitical risks leading to business losses in certain countries; impact of macroeconomic conditions on internet business, with advertising revenue continuing to be under pressure; increasing R&D expenses for new product businesses such as automotive, leading to long-term pressure on profits.