Opening: What Might Xiaomi’s Real Problem Be?

Over the past decade, Xiaomi’s “hardware + new retail + internet” ironman-triathlon strategy carved out a real place for it in global consumer electronics. But zoom out to the present, and the picture may not be so rosy—the stock price has been cut in half from its peak, its PE hovers around 17x, and it can’t even keep up with the valuations of some traditional hardware makers, let alone pure-play AI companies.

My thinking is that the core problem probably isn’t that Xiaomi’s execution has slipped. It’s that the market has largely stopped buying the “ecosystem” story. Apple and Huawei draw their strength from operating-system-level control, and capital markets are willing to pay a premium for that. Xiaomi’s HyperOS still rests on an Android+ARM foundation, which in investors’ eyes makes the moat “not deep enough.”

Building on that observation, here’s a wild idea: Xiaomi could simply reposition itself as an “AI company.” Rather than slugging it out with Huawei and Apple over the definition of “ecosystem,” it could flip the script and lean into its unique “human-vehicle-home” full-scenario hardware advantage—using AI to rebuild every product line, from phones and chips to OS, IoT, cars, robots, and laptops. This isn’t a rejection of the existing business; it’s more like finding a sexier new story for the assets it already has.

Below, I’ll unpack this brainstorm across six dimensions.

Part One: The Current State and the Dilemma

I. What Is the Phone Business Going Through?

1.1 “Volume, Price, and Profit” All Under Pressure at Once

In Q1 2026, Xiaomi shipped roughly 8.7 million phones domestically, down 35% year-over-year—the first time since 2017 that IDC placed it in the “Others” bucket. Huawei led with 19.8%, Apple sat at 18.9%, and OPPO, vivo, and Honor took third through fifth, with Xiaomi at around 12% in sixth. Omdia’s data also showed this was the steepest decline among the top five brands.

Globally, Xiaomi still ranked third with 33.8 million units (12% share), but its 19.1% year-over-year drop was also the largest in the top five.

(2) Memory Chip Price Hikes Strike the Low- and Mid-End Where It Hurts

The core driver of this decline was DRAM and NAND Flash prices rising 55%–95% quarter-over-quarter in Q1. About 60% of Xiaomi’s shipments are sub-RMB 1,500 models, with per-unit gross profit of only around RMB 90 and memory costs accounting for over 30% of the device. Memory costs roughly quadrupled year-over-year—raise prices and users flee; don’t raise them and you bleed. In the end, Xiaomi had to voluntarily scale back shipments of volume models like the Redmi Turbo and Note.

The financials tell the story plainly: Q1 phone revenue was RMB 44.3 billion, down 12.5% year-over-year; gross margin fell from 12.4% to 10.1%; and shipments dropped from 41.8 million to 33.8 million units. IDC predicts the memory shortage could persist through the end of 2027, meaning this cost pressure is likely here for the long haul.

(3) Three Cycles Piling On at Once

On the Q1 earnings call, Lu Weibing also noted that cost, demand, and competition cycles are all bearing down simultaneously:

Pressure Dimension	Specific Manifestation
Cost cycle	Memory price hikes expected to run through all of 2026; entry-level segments with weak pricing power face elimination risk
Demand cycle	Global smartphone shipments in 2026 projected to fall 13.9% year-over-year, the lowest since 2013
Competition cycle	Huawei’s domestic share at 19.8%, up 7% YoY; Honor growing fastest at +19% YoY; OPPO and vivo hold firm offline channels

(4) Resources Shifting to Cars and Appliances—Phones Are “Bleeding”

Internally, the 2026 phone sales forecast has already been cut by 13%, and resources are clearly shifting toward cars and major appliances. In Q4 2025, phone gross margin dropped to 8.3%, while IoT and major appliances sat at 23.1%. The offline channel has stopped expanding stores and shifted to “operations-oriented” assessments, with the sales focus moving from phones to higher-margin ecosystem products. The side effect is that new-product promotion has weakened—the Xiaomi 17 series may sell only 7 million units over its entire lifecycle.

In 2025, the share of domestic high-end phones priced above RMB 3,000 rose 3.8 percentage points year-over-year to 27.1%. The RMB 4,000–6,000 price band held a 17.3% market share, and the RMB 6,000–10,000 band sat at 4.5%—both up year-over-year. Yet overall annual ASP actually fell from RMB 1,138 to RMB 1,129, with revenue at RMB 186.4 billion (–2.8% YoY), shipments of 165.2 million units (–2.0% YoY), and gross margin down from 12.6% to 10.9%.

In other words: high-end phones did sell better, but low-end phones in overseas emerging markets dragged the average price down. Premiumization is breaking through in pockets, but the overall structure hasn’t stabilized yet.

(2) Compared with Huawei and Apple, the Brand-Perception Gap Remains Large

Above the RMB 6,000 threshold, Huawei’s and Apple’s brand moats are simply too deep. Xiaomi’s share there is only 4.5%—up 2.3 percentage points year-over-year, but off a tiny base. In most consumers’ minds, the “Xiaomi = value for money” label is firmly cemented. One industry analyst put it bluntly: “Premiumizing a brand is absolutely not something that can be achieved with a single product or in a short time—it’s tied to changing consumer perceptions.”

(3) The Technology Brand Hasn’t Been Established

Xiaomi’s premiumization relies more on “stacking specs”—Leica imaging, Qualcomm flagship chips—than on building brand through OS, chip, and AI capabilities. Users pay Huawei a premium because the “HarmonyOS + Kirin” combo tells a story of technological sovereignty; they pay Apple a premium for the experience lock-in of the iOS ecosystem and A-series chips. The brand pull of the Surge chips, honestly, still lags far behind. The Xuanjie chip direction is right, and a RMB 50 billion investment over 10 years is itself both R&D and marketing—but the effect on “brand image elevation” still needs time to prove out.

(4) “Mi Fan” Sentiment Is, in a Sense, a Double-Edged Sword

Mi Fan culture is Xiaomi’s ace in the hole, but on the premiumization front it can sometimes become a drag. In the KOL controversy early in 2026, the official spokesperson used the word “severe” three times in a row, and Lei Jun himself said the PR team was “swept along by the hype,” leading to “distorted actions.” The early feeling of equal dialogue—“Because of Mi Fans, there is Xiaomi”—has in some ways morphed into emotional hostage-taking. If fans feel “betrayed,” what used to be a moat can backfire. Premiumization requires a certain “distance” and professional tone, but Xiaomi got its start by “mingling with users”—and those two temperaments are inherently in tension.

(5) Summary: The Essential Dilemma of Premiumization

At bottom, Xiaomi’s real premiumization problem isn’t that “high-end phones won’t sell.” It’s that “high-end phones do sell, but the brand premium and a true moat haven’t followed.” The RMB 4,000–6,000 band is sellable; above RMB 6,000 gets hard. Specs have caught up, but the technology brand hasn’t been established. User affection is still there, but it can’t support a higher level of value identification. If Xiaomi could shift from “making users feel it’s worth the price” to “making users feel they can’t do without this brand”—that might be exactly where an “AI+” strategy can help.

II. Why Lean Toward AI?

2.1 Valuation: The Ecosystem Narrative Has Lost Its Power

Current market cap is roughly HK$733.8 billion, down nearly half from its 2025 peak. Phones are taking hits from the memory cycle, cars are still in the investment phase, and AI R&D spending is enormous but can’t be valued separately. With all three segments tangled together, investors have no idea how to price it—target prices from different institutions diverge by as much as 60%.
The deeper reason: the market only grants an “ecosystem premium” to operating-system-level control (Huawei’s HarmonyOS commands roughly a 3–5x PS premium). The fact that HyperOS rests on Android+ARM makes the moat “not deep enough” in investors’ eyes. Keep telling the “full ecosystem” story, and you’ll just keep getting measured against Huawei.

2.2 Competition: Wedged Between Apple and Huawei Is Uncomfortable

Apple: 2.5 billion high-value devices + a closed system—a rock-solid “AI distributor,” with a services gross margin of 74%.
Huawei: HarmonyOS + Ascend builds a “sovereign technology system” narrative that plays to both the state and consumers.
Xiaomi: Domestic phone share has fallen to fifth, car growth is slowing, and the gap in original AI capability versus top-tier large-model companies is conspicuous. Staying on the “ecosystem” track will only ever prove “I’m a little worse than they are.”

2.3 Vision: The Market Hasn’t Understood What Lei Jun Is Saying

Lei Jun has spoken several times about “the grand convergence of chips, OS, and AI on a single device” and “the AI gateway to the physical world.” It’s a grand vision, but the market has read it as “yet another company making cars and phones.” The likely reason: the external story is still centered on “products,” not on “AI capability.” If investors can’t see “how AI rebuilds every product and creates new revenue,” they’ll keep valuing it as a hardware company.

My personal take: Xiaomi doesn’t lack assets, doesn’t lack scenarios, doesn’t lack data. What it lacks is a new “AI-first” story that both capital markets and users can understand and believe in.

2.4 The Latest External Signal: Huawei’s HarmonyOS 7 Has Already Pushed the Battlefield to Agent OS

If in previous years HarmonyOS’s core narrative was “a domestic operating system to replace Android,” then after HDC 2026 in June 2026, that narrative has clearly leveled up. Public reports indicate that at the conference Huawei launched the HarmonyOS 7 developer beta and announced that “HarmonyOS Intelligence is fully evolving toward an Agent architecture.” Strictly speaking, “AI Agent OS” may not be Huawei’s official full name, but it’s an apt way to characterize this directional shift: HarmonyOS isn’t just bolting an AI assistant onto the system—it’s transforming the operating system itself into the substrate for agents to run, understand, invoke, and distribute.

Several signals from this HarmonyOS 7 release are key:

At the system-architecture level: An agent-affinity system architecture was introduced, building a cloud-device-collaborative Agent technology substrate that supports three paths: application intelligence, application Agent-ification, and application Skill-ification.
At the developer-framework level: The HarmonyOS intelligent agent framework was upgraded to HMAF 2.0 and opened to industry developers; reports mentioned a complex-task success rate exceeding 90%, and over 20 AI capabilities were opened up, including GUI control.
At the entry-point level: Xiaoyi continued its evolution from a traditional voice assistant into a system agent, integrating on-device and cloud large-model capabilities to handle context understanding, task planning, and service scheduling, with the goal of becoming a personal intelligent computing hub.
At the ecosystem level: HarmonyOS 6 devices have surpassed 66 million, registered developers exceed 11 million, and available apps and services top 400,000. In other words, Huawei isn’t talking about Agents on a paper system with no users or apps—it’s migrating toward an Agent OS on an already-scaled, independent ecosystem.

The implication for Xiaomi is direct: the next phase of competition in phones and full-scenario ecosystems won’t stop at “whose system is smoother, who has more apps, whose interconnection works better.” The next round of differentiation may become: whose OS is better suited for Agents to reside persistently, who can turn app capabilities into Skills, who can safely let AI orchestrate across devices and services.

This also explains why Xiaomi can’t just treat AI as an upgraded Xiaoai. What Xiaomi truly needs to benchmark against isn’t “a smarter voice assistant,” but “a new generation of operating-system capability centered on Agents.” If HyperOS remains merely a device-interconnect layer on top of Android, it will keep being out-narrated by Huawei. But if HyperOS can become the Agent Runtime across the “human-vehicle-home” world—turning phones, cars, appliances, and robots into physical capabilities that agents can understand and dispatch—it could carve out a path that’s more lifestyle-grounded and higher-frequency than Huawei’s.

Part Two: The Transformation Vision—If They Actually Did This

III. If the Transformation Succeeds, What Would Xiaomi Become?

3.1 The Valuation Logic Gets Rewritten

No longer “the Apple of China,” but more like “the OpenAI of the physical world” + “a Tesla-tier AI hardware company.” Pure AI-application companies can fetch 10–20x PS, and AI-empowered hardware companies (like Tesla) can get 5–8x.
Sum-of-the-parts valuation: Phone × AIoT as an AI-enhanced consumer-tech company (PE 25–30x), cars as a smart-driving service provider (PS 3–5x), and the AI business could even raise capital and be valued separately. Conservatively, simply removing the current “conglomerate discount” would give fair market value 30%–50% upside.

3.2 The Competitive Logic Flips

Xiaomi wouldn’t need to prove “my system is better than HarmonyOS,” only that “my AI understands your daily life better than anyone.”
The one-line differentiator: “Huawei gives you a self-sovereign, controllable system; Xiaomi gives you a proactive, service-minded AI butler.” Stickiness shifts from “switching systems is costly” to “switching AI makes life inconvenient.”
But here I have to layer in the latest shift: HarmonyOS 7 is no longer just “a self-sovereign, controllable system”—it’s moving toward an Agent OS. If Xiaomi still stalls at “a collection of AI features,” Huawei’s new narrative will outclass it again. The real flip should be: Huawei builds a sovereign Agent OS; Xiaomi builds the Agent OS that best understands physical-life scenarios.

3.3 The Revenue Model Widens

On hardware: AI becomes the justification for a premium—AI-enhanced photography, real-time translation, driving assistance can be packaged as paid features or high-end exclusives.
On services: AI subscription bundles built on the full “human-vehicle-home” scenario (advanced smart-home automation, cross-device memory services, etc.). Using Tesla’s FSD logic as a reference, even at just 10% penetration, 746 million monthly active users translate into a revenue imagination on the order of tens of billions of RMB per year.

3.4 Product Thinking Changes from the Root

From “stacking specs and competing on configurations” to “every experience starts from AI.” A product manager’s first question is no longer “how big is the screen, how big is the battery,” but “can AI solve a real pain point in this scenario?”
R&D resources get redeployed: chips prioritize AI acceleration, the OS deeply integrates an AI agent runtime, and cars treat smart driving as the core selling point rather than an assistive feature.

Part Three: How Could It Actually Be Done?

IV. Four Main Lines

Main Line One: Make the “AI Company” Identity Clear

Direction for a new brand slogan: “Xiaomi—rebuilding the physical world with AI,” or “AI for Everything, for Everyone.”
De-emphasize the word “ecosystem.” In earnings reports, launch events, and investor communications, push metrics like “AI penetration rate,” “AI feature activation rate,” and “AI service revenue” to the front.
Hold an annual “Xiaomi AI Day” to benchmark against Tesla AI Day—showcasing models, agents, and robotics progress—slowly planting the impression that “this is an AI tech company.”

Main Line Two: Rebuild Every Product Line with “AI-Native” Logic

Product Line	Possible AI-Rebuild Direction
AI+Phone	Make agents like miclaw the system-level entry point, replacing the leftmost screen and the traditional voice assistant to enable proactive cross-app services
AI+Chips	Publicly commit to a Surge chip roadmap—within three years, all SoCs integrate dedicated AI acceleration units, deeply optimized for large-model inference
AI+OS	The next HyperOS version takes “AI Runtime / Agent Runtime” as its kernel, with a low-power always-on on-device large model, Skill-ified app capabilities, and an open, unified framework for permissions, memory, tool invocation, and cross-device scheduling
AI+IoT	Every Mijia device has at least one AI capability (environmental adaptation, fault prediction, voice interaction), all uniformly orchestrated by Xiaoai
AI+Car	Expand the smart-driving team to over 3,000 people; cover 100 cities with urban NOA by 2027; explore an “in-car AI space” subscription
AI+Robotics	CyberDog and CyberGear go beyond developers toward home services (cleaning, companionship), letting AI deliver genuine usability
AI+Laptop	Built-in on-device large model; productivity features like meeting notes, code assistance, and PPT generation, serving as the AI entry point for office scenarios

Main Line Three: Close the Gap in Original AI Capability—Walk on Two Legs: Open Source + Applications

Short term (1 year): Embrace open source deeply; continue contributing high-quality models based on the MiMo series; build a developer community on GitHub to attract third parties to develop AI applications on Xiaomi hardware.
Medium term (2–3 years): Establish a “Xiaomi AI Fellowship,” funding roughly 30 top AI PhDs per year to specifically tackle multimodal, embodied intelligence, and on-device inference. Recruit more young, bold thinkers like Luo Fuli.
Long term: Build a full-stack “model-chip-scenario” closed loop and achieve global leadership in a few vertical domains like on-device multimodal and smart-home policy.

Main Line Four: Find the Path to Monetize AI

Free + value-added pilots: Basic AI features are free; premium features (long memory, complex cross-device tasks, personalized model fine-tuning) are monthly subscriptions. Start by testing the waters on cars and high-end phones.
B2B opening up: Open Mijia’s AI capabilities (environment perception, device-control APIs) to third-party developers, charging per call volume—essentially a “smart-home AI app store.”
Data-contribution rewards: Make privacy boundaries clear; users who voluntarily share de-identified data to participate in model training get cloud services or subscription discounts, building a positive trust relationship.

V. Some Breakout Ideas for the Immediate Dilemmas

5.1 On Phones: Use the AI Experience to Sidestep the “Memory Chip Trap”

The essential dilemma: Memory price hikes are eating into the margins of low- and mid-end phones, while high-end phones lack differentiated selling points to support their prices.

Possible directions:

First, Use AI Features to Redefine “Worth It”

If phone differentiation relies solely on hardware specs, it’s easily swallowed by supply-chain swings. AI-experience differentiation is soft, continuously evolving, and hard to replicate. Specific brainstorms:

Starting in the second half of 2026, preinstall the miclaw agent as a system-level AI entry point across all mainline models. The selling point shifts from “the latest Snapdragon chip + Leica lens” to “AI-native OS: your phone actually gets you.”
Rather than raising hardware prices to offset memory costs, push an “AI subscription service pack” (premium AI features + cloud storage + cross-device synergy), creating new revenue through a monthly fee and shifting part of the cost from BOM to service payments.
Lu Weibing nailed the direction with one line: “An AI phone isn’t a smartphone with AI features bolted on—it’s a shift from App-centric interaction to OS-Agent-centric interaction.” If this pivot succeeds, the share of software and services revenue rises noticeably, and the impact of hardware-margin volatility shrinks.

Second, Pivot the Operating Metrics Accordingly

Put “monthly active AI-feature users,” “per-capita AI usage time,” and “AI subscription conversion rate” into each region’s core KPIs. Teams optimize for whatever you assess—this shift itself sends a signal.

5.2 On Premiumization: Switch from a “Specs War” to an “AI Identity War”

The essential dilemma: For phones above RMB 6,000, what users buy isn’t just specs—it’s identity and a technology narrative. Huawei has “sovereignty,” Apple has “ecosystem,” and Xiaomi currently has only “extreme value for money,” which can’t support an ultra-premium price.

Possible directions:

First, Build the Premium Narrative with an “AI-Native Experience”

The “five-year pledge” has matured—it’s time to upgrade from “maxed-out specs” to “an experience-defined AI terminal”:

The next foldable/Ultra: AI capability is the number-one selling point. For example, a system-level AI agent handles cross-app complex tasks—“help me plan next week’s Hangzhou trip and book a hotel,” and the AI works the whole flow itself—maps, ticketing, hotel booking. Huawei’s HarmonyOS 7 has already set the tone by leading on Agent OS; Xiaomi can no longer treat this kind of capability as a “bonus feature” but must make it the core purchase reason for flagship phones. Pricing shifts from “how much are the specs worth” to “how much is the capability worth.”
Deepen AI-personalized “indulgent consumption”: Car customization and the custom-personalized phone craze around the Xiaomi 17 Pro have already validated that users will pay for personalization and emotional value. Extend that logic to AI services: personalized model fine-tuning (training your own AI assistant), customized smart-home policies, cross-device memory services (phone–car–IoT full-domain synergy), and AI-generated exclusive content. Let AI become a new way for users to “express themselves.”
“Proactive service” as brand identity: Position differently from Huawei’s “system sovereignty” and Apple’s “secure walled garden”—build the perception of “the AI life butler that understands you best.” Choosing Huawei means “sovereign and controllable”; choosing Apple means “worry-free and safe”; choosing Xiaomi means “thoughtful and proactive.”

Second, Use the “Physical AI Gateway” Positioning to Reshape Brand Associations

Right now, when people think Xiaomi, they think “high-value-for-money phones.” Over the next five years, aim to refresh that to “Xiaomi = the key to the AI world”:

Make AI Day an annual event, benchmarking against Tesla AI Day, systematically signaling “we are an AI company.”
Launch flagship phones with the latest model capabilities and AI-chip acceleration first, gradually forming the perception that “annual AI progress = a reason to upgrade.”
Build a “Xiaomi AI Certified” system—third-party apps and developers only get ecosystem support after passing certification, giving users “something only the Xiaomi ecosystem has.”

Third, Upgrade Fan Community Sentiment—Don’t Cut It Off

Premiumization isn’t about severing Mi Fans; it’s about upgrading that relationship:

A “Xiaomi AI Co-Creation Program”—core users participate in training and optimizing AI agents, continuing the “growing together with users” ethos but elevating the interaction to AI-tech co-creation.
Upgrade the community into an “AI developer community,” where users build personalized Agents on Xiaomi’s AI capabilities—“participation” shifts from flashing ROMs to AI co-creation.
Set up a “pioneer user program” for premium product lines, using a more professional, more influential user base to generate new word-of-mouth momentum, gradually diluting the “fandom-ization” risk.

Fourth, Land the “Three-in-One” of Chip + AI + OS

This is the ace that genuinely anchors ultra-premium phones. What Lei Jun described—“realizing the ‘grand convergence’ of self-developed chips, self-developed OS, and self-developed large models on a single device”—is inseparable from premiumization:

Before 2027, close the loop on ultra-premium models: “Surge AI acceleration + HyperOS deeply integrated with MiMo + a system-level AI agent.”
Make this “three-in-one” the core selling point and the spine of the brand narrative for ultra-premium models.
Borrow the Xuanjie chip logic of “it’s both R&D spend and marketing spend”—breakthroughs in chips, OS, and AI should themselves be run as systematic brand events.

5.3 “On-Device Intelligence” Could Be Xiaomi’s Natural Advantage

Cloud-device synergy: Cloud large models (MiMo) + on-device inference, deeply combined—on high-end phones, “strong AI even offline,” while protecting privacy. When most AI features still depend on the cloud, Xiaomi offering a “personalized, localized” experience becomes a moat.
Cross-device data flywheel: 1.079 billion AIoT connected devices, forming a “full-domain data loop” from phones to cars to homes. The more data agents accumulate through cross-device scheduling, the more accurate the experience gets, and the harder it is to leave—gradually forming a “data moat” that single-device makers can’t replicate.
Accelerated self-developed AI chips: Surge’s AI-acceleration module upgrades from a “symbolic presence” to “substantive experience improvement,” with a public roadmap declaring “within three years, Surge AI performance will be industry-leading,” building technical trust with high-end users.

5.4 Upgrade from “HyperOS Ecosystem” to “HyperOS Agent OS”

HarmonyOS 7’s direction is a reminder to Xiaomi: the core of the next-gen OS is no longer the old paradigm of “desktop, notification bar, device interconnection, app store”—it’s a system substrate that lets Agents run safely, stably, and frequently. If Xiaomi truly wants to tell an AI-company story, it should define HyperOS’s next phase as an Agent OS, not just “HyperOS with AI bolted on.”

The good news is that Xiaomi isn’t starting from zero. An April 2026 announcement from the Xiaomi Surge OS developer platform shows that Xiaomi miclaw is a system-level AI Agent built on the MiMo large model; the Agent ecosystem platform has opened a small-scale public beta for developers, supporting MCP, Skill, and Agent uploads. This shows Xiaomi already has an early prototype of an Agent ecosystem—the problem is that it hasn’t been framed as a sufficiently clear, sufficiently ambitious OS-level strategy.

Concretely, this can be broken into four things:

Agent Runtime: A system-level always-on agent runtime that uniformly manages models, tools, memory, permissions, context, and task queues. The user doesn’t open an app each time; instead, they hand a goal to the Agent, which then invokes apps and devices.
A Skill-ified ecosystem: Abstract phone apps, Mijia devices, car functions, and robot actions into Skills the Agent can call. Third-party developers don’t just list apps—they list “capabilities that can be orchestrated by Xiaomi’s Agent.”
A physical-world permission system: Xiaomi needs a fine-grained permission framework more than a pure phone maker does, because an Agent might turn on the AC, adjust the car window, order consumables, or control a door lock. The deeper you go into the physical world, the more the security boundary becomes a core competitive advantage.
Cross-device memory and a scenario graph: Let the Agent remember a user’s long-term preferences across home, car, office, exercise, and travel—but it must default to local-first, transparent authorization, and viewable-and-deletible data. Xiaomi can make “understands life” a brand asset—but only if it doesn’t make users feel violated.

If Huawei’s Agent OS is strong on “self-developed system + chips + government/enterprise ecosystem + sovereign control,” Xiaomi should position its own Agent OS as “the most complete in consumer-life scenarios”: from the wristband waking you in the morning, to kitchen appliances making breakfast, to the car planning your commute route and the phone handling messages, to the evening synergy of AC, lights, and entertainment. These high-frequency physical touchpoints are Xiaomi’s unique board to play on.

5.5 Rough Staged Evolution Direction

Timeframe	Phone Business Focus	Premiumization Target	AI Capability Landing
End of 2026	Domestic share back in the top four; phone gross margin held above 12%	Market share in the RMB 6,000+ band breaks 8%	miclaw system-level preinstall covers mainline models
End of 2027	Share steady in the top three; ASP breaks RMB 1,300	At least one AI-native model anchors the RMB 7,000 tier	Chip AI acceleration + OS deep large-model integration initially realized
End of 2028	AI service revenue exceeds 5% of phone revenue	Market share in the RMB 6,000+ band breaks 12%	The “chip + OS + AI” closed loop fully lands across high-end phones

The point isn’t to “beat” Huawei and Apple on every metric in the short term—it’s to use the “AI+” transformation to build a brand-new value narrative and competitive dimension, fundamentally changing the competitive landscape and user perception of the premium market.

Part Four: If They Really Push It, How to Manage the Pace and Risks?

VI. The Rough Implementation Pace

6.1 Phase One (0–12 months): Change the Story First, Adjust the Organization

Lei Jun personally delivers an “AI+” strategy manifesto; AI metrics get added to the KPI system, and every business line has to carry them.
Set up an “AI Strategy Office” reporting directly to the CEO, stringing together the AI resources across chips, OS, phones, cars, and other departments.
Launch the first batch of “AI-native” products (e.g., Xiaomi 15 AI Edition, Xiaomi Car OS 2.0 with AI Agent).

6.2 Phase Two (12–36 months): Products Land, the Data Flywheel Starts Turning on Its Own

All mainline products complete their AI rebuild; launch an AI-native device defined by “no feature without AI.”
AI subscription penetration reaches 10%, contributing over 5% of total revenue.
Open-source community activity breaks into the global AI-company top ten; at least one self-developed model becomes a benchmark in some domain.

6.3 Phase Three (36–60 months): The Ecosystem Takes Shape, Value Starts to Be Realized

“AI+” is thoroughly fused into the company’s cultural DNA; externally, the perception of Xiaomi is completely different.
The valuation switches to “AI platform company” logic; market cap returns above HK$1.5 trillion.
Service-revenue share rises above 25%, becoming the primary profit driver.

6.4 Possible Risks and Responses

Risk	Possible Response
Heavy AI investment, ugly short-term profits	Set staged investment caps; prioritize high-return scenarios like phones and cars; the AI business can raise capital independently to ease the group’s cash pressure
Strong organizational inertia, can’t push it through	Lei Jun personally leads; mid-level managers who don’t embrace AI get replaced when they should; set up an “AI Transformation Award” for teams willing to charge ahead
Users unwilling to pay for AI subscriptions	Long-term free trials first, building the habit; borrow Xiaomi’s early hardware playbook—occupy the mind-share first, then figure out how to monetize
Can’t catch up on original technology	Don’t chase across-the-board leadership; focus on Xiaomi’s two unique advantages—“on-device inference” and “cross-scenario orchestration”—and break through locally

Closing: Xiaomi’s Real Ace May Be “the Broadest Set of Physical AI Touchpoints”

Think about it carefully—Xiaomi isn’t an ordinary hardware company, nor an ordinary car company. It’s currently the only tech enterprise in the world simultaneously holding hundreds of millions of phones, millions of smart cars, and billions of IoT devices. These devices aren’t islands—they’re the tendrils AI extends into the physical world.

Apple is still agonizing over whether to build a car, Huawei is constrained by capacity, and OpenAI has no hardware of its own—Xiaomi is already standing on the best starting line for “physical AI.” The key isn’t whether it has the capability; it’s whether it dares to reintroduce itself.

If one day Xiaomi can say, both internally and externally, “We are an AI company”—that’s not abandoning hardware; it’s giving hardware a soul. It’s not discarding the ecosystem; it’s letting the ecosystem grow in AI soil.

Perhaps in the next decade, Xiaomi will have the chance to become the benchmark that “re-understands the physical world with AI.”

Disclaimer once more: The above is purely personal observation and brainstorming, organized into writing with DeepSeek’s assistance, and does not constitute a formal strategy proposal. Data comes from public sources and may be lagging or contain deviations. Discussion, critiques, and joint brainstorms are all welcome.