One day at the start of 2024, Li Chuangqi, who was in charge of Xiaomi’s AI glasses, initiated an unusually consequential conversation.
He dragged a document into his internal Xiaomi office app chat window with Lei Jun and typed: “…I’ve made some supplementary thinking on the intelligence direction; I believe glasses are an important strategic product position and complement for the wearable business.”
Then he hit Enter.
The document was titled “Feasibility Considerations for Xiaomi’s Wearables Business in the Glasses Category,” and it also served as the project background and premise for Xiaomi’s first AI glasses that year. At the time the project still faced internal doubts; if Li Chuangqi wanted to push it forward, he needed higher-level approval and greater support.
Lei Jun’s reply came quickly. He first “pinned” the document — a conspicuous flag that signals the sender’s importance and lets one find it easily later — and then wrote, “Go over this carefully with Jiutang (Xiaomi Group Chief of Staff) and Zhang Lei (Vice President of Xiaomi’s Phone Division and Head of Wearables).”
For the entire Xiaomi AI glasses project team, that message was crystal clear. Although the company had previously experimented with “camera glasses” and “audio glasses” forms, this was the first time the concept of “AI glasses” was formally established at Xiaomi, and the project was soon officially launched.
When a startup lane gets lively, the outside world likes to call the fight a “hundred-X battle” to indicate how hot it has become. After rounds like the hundred-teams or hundred-model battles, the internet’s new phrase of the moment was the “hundred-glasses battle.”
The spark came from Zuckerberg’s company. In September 2023, Meta washed away the failure of the first Ray-Ban Stories with the second-generation Ray-Ban–Meta glasses.
This product, which integrated the LLaMA model and offered real-time translation, voice Q&A, and other AI functions while also supporting audio and photography, impressed the market with a $299 price, a 48 g weight, and a plain sunglasses appearance — selling one million pairs globally in under a year and becoming the category’s first and so far only runaway hit.
That was a huge stimulus to China’s very sensitive tech companies. From startups to AR/VR firms, from phone makers to internet giants, everyone adopted a posture of rushing in.
They were all betting on one question: could AI glasses become the next consumer-level entry point after smartphones?
Landing the beachhead, waiting for the inflection point
At the end of 2022, after ChatGPT’s sudden rise, Xiaomi’s wearables division ran a study on product applications of large models, and one conclusion was that future smart hardware products must be able to see, hear, and sense.
Working backwards from the latter two requirements, manufacturers can design many sensors for various body parts (for example, earphones, wristbands), but the first requirement leaves only one option: glasses, because the visual entry is unique.
Alibaba has reached a similar judgment. At WAIC2025 the Quark AI glasses showcased their technical progress; Song Gang, head of terminal business at Alibaba’s Intelligent Information Group, said that AI glasses will be the most important product form in smart wearables, becoming another pair of “eyes” and “ears” for humans with powerful scene-penetration capabilities.
“Glasses will certainly become the ’sensory hub’ of the next generation of human-computer interaction, triggering a dramatic surge in personal data,” Song said. “They could become the most important personal mobile entry point after smartphones.”
But AI glasses are still waiting for their inflection point.
That turning point will come once the supply chain truly matures and sales reflect it. Xiaomi once estimated “three years, five million units.” Only after crossing that point will the strategic status of AI glasses become clearer; if it never arrives, the category could fall back to being an “electronic toy.”
Although the “hundred-glasses battle” hasn’t fully erupted, the entrants and the ecosystem are already very large and diverse. Among internet giants there are Alibaba and Baidu; phone makers include Huawei and Xiaomi; AR/VR players such as Thunderbird and Rokid are present, and a slew of startups like Shanji, In Air, Beehive Technology, Li Weike, and Yingmu Technology are all in the mix.
Once the direction is set, “seizing the moment” became the common mindset among competitors.
After Xiaomi’s AI glasses were greenlit under the codename “O95,” besides core figures like Zhang Lei and Li Chuangqi, many team members were drawn from other departments as needs arose.
Zhang Lei and Li Chuangqi would go to the heads of other departments and say, “I want these people — who do you think is suitable? Send them here.”
O95 pulled in project managers, product managers, software and hardware engineers from inside the company, and seconded staff from the camera team, Xiao Ai, and core partners. After the project launched, the headcount grew exponentially; at its peak, over a thousand people were involved.
They also set up a temporary office. On the fifth floor of a building in Xiaomi’s technology park, the team converted an old spare-parts warehouse used for phone testing into what they called the “little black room.” Dozens of people camped there while others came and went daily to work on requirements.
Startups felt even greater urgency. Shanji — a company that had moved from power banks into AI glasses — had historically targeted the top 10% of the power-bank market. “But our ceiling was low; if we only make power banks, annual sales are around just over a billion yuan,” Shanji COO Chen Pan told Jiemian News.
In 2021, after replenishing their war chest via investors, CEO Zhang Bo, COO Chen Pan and others began looking for their next core product line. Ray-Ban Meta entered Shanji’s sight during this period.
Running roughly the same rhythm as Xiaomi, by late Q1 2024 Shanji’s team finally sorted out consumer scenarios, user portraits, costs and pricing and decided to enter. Unlike Xiaomi’s quiet work, Shanji moved fast and held a launch conference in late May — a launch without a finished product, driven by marketing. As one of the earliest companies that could have replicated Ray-Ban Meta, this move generated a lot of market attention, but the company’s real goals were to send two strong signals to the industry: we’re raising funds, we’re hiring.
“AI glasses aren’t something you can do with just tens of millions [of RMB]; we want to raise another round,” the company said. That year Shanji secured A/A+ round funding from Guangyuan Investments, Future Lightcone, Yuntian Lifei, Oasis Capital and others, closing the deal within the year.
“One condition was to complete the deal within four months,” Shanji told investors. The team wanted to spend more energy on product development and less on fundraising.
Aside from money, the team’s biggest shortage was a person who both understood AI and knew how to apply it. Zhang Bo and Chen Pan flew between Shenzhen and Beijing several times to recruit such a person; in November 2024 they announced that Pan Xin — who had core AI experience at Google and ByteDance and was then cofounder at Ling Yi Wanwu — had joined as CTO.
Nevertheless, Shanji’s first-generation “AI PaiPai Mirror” A1 was not successful; in some respects it was a failure. Chen Pan admitted the team had to accept reality while still seeking opportunities to make up. What they could not regain was timing; Chen Pan pointed to a root cause: for a small company known for power banks, touching a technically advanced, potentially large-category product like AI glasses required an early bet.
“It’s hard to squeeze in later,” Chen said.
The battle between big and small
As early as 2023, Li Chuangqi had advised many venture contacts not to invest in AI glasses.
In his view this opportunity was essentially reserved for phone makers with wearable ecosystems and internet giants that control large models and national-level apps — and only these two groups. “Users don’t need just a piece of hardware; they need a whole suite of services,” he said.
By that logic, ecosystems of Apple, Samsung, Huawei, Xiaomi and the like would become ever more advantageous; unless smaller vendors can achieve full permission compatibility, the unique product value of an independent maker becomes unclear.
An AI glasses product manager at Goer (Goertek) told Jiemian that after serving multiple clients he personally favors phone makers for AI glasses — they have strong hardware genes, complete and mature systems from R&D and supply chain to quality management, and they can penetrate interaction down to the lowest layers with phones.
Startups don’t buy that logic wholesale. “Ten friends might all have the same phone, but they won’t all wear the same pair of glasses,” Chen said. Glasses have such a private, personality-expressing quality; unlike phones that can be hidden in a pocket, glasses are visible. Small companies believe the market is big enough for them to survive.
Shanji thinks that within three years supply-chain gaps in hardware will be closed. “If you can make the hardware, I can too — there are no engineering hurdles that can’t be overcome,” Chen said. That mirrors the view of In Air founder Huang Hai: China’s hardware supply chain is “open source.”
Of course, some detailed advantages accrue to big players because of their resources. In tests, Xiaomi’s AI glasses’ photo capture time was 0.87 seconds, while Shanji’s A1 was close to 2 seconds. That one-second difference came from joint tuning by Xiaomi’s camera team and chip suppliers — an ability small companies don’t usually have.
Chen said Shanji could work to get it down to around a second, but that’s not a single-step improvement; it requires end-to-end optimization across camera, algorithm and compute hardware. The hope is that such optimization might eventually be mastered by contract manufacturers, so absolute barriers may not persist.
On software positioning, many interviewees pointed out that today AI glasses are essentially accessories that grow out of phones: interactions ultimately return to the phone and app. That’s true, but Huang believes startups must escape the “accessory” mindset and treat AI glasses as independent entities — building a complete application ecosystem and an operating system for the device, offering rich content and interaction so the glasses can become a high-frequency endpoint rather than a relay.
That approach is more complex, but also offers greater potential to build moats. If the bar is too low, big players will just flatten the competition.
Therefore, Shanji plans to emphasize AI agents and software experiences; the team has built Loomo OS with nearly forty people working on it. In Air has created its own spatial operating system, with roughly a 2:1 software-to-hardware team ratio. Huang believes that even if a big company wanted to replicate such a spatial OS, it would take at least a year.
Everyone agrees that AI applications will be the product core in the future, but currently differentiation is hard to establish. Core capabilities cluster around recognition, organization, memory, retrieval and translation.
Xiaomi once tried hard to innovate an AI application, but kept cycling through searching and rejecting ideas. Today’s AI apps are mostly efficiency enhancements and haven’t produced the kind of killer app — like mobile payments or short-video platforms — that the industry has long awaited.
On the model-invocation level, there’s little room to pull away. Nearly every manufacturer’s approach is the same: build a “switchboard” that integrates different model interfaces and routes different questions to different models based on user needs and model performance.
Li Chuangqi believes the immediate need for AI assistants is to be more human-aware rather than staying at the level of tool invocation and function control. As chips and foundational models evolve, more sensors could be introduced to deepen human perception, letting the brain’s System 1 (fast thinking) operate on the glasses and phone, while System 2 (slow thinking) runs in the cloud.
Clearly there is still distance between AI glasses today and the ultimate vision. “If our target is a full score, right now we’re maybe at 60 out of 100,” Li said.
Maturing the supply chain: losing money to make noise
Entrants broadly agreed on one point: the supply chain is “immature.” In March, at a closed-door industry meeting attended by some regulatory officials, participants said the biggest contradiction for the AI glasses industry was that consumer expectations far exceed the industry’s current capabilities.
That made Xiaomi’s supplier selection logic appear as “choose the most expensive.” “This wasn’t a preset rule; we reviewed almost every market solution and evaluated them all — and found that the most expensive was indeed the best,” Li said.
Startups can’t always follow that rule; sometimes they aren’t allowed to. Shanji once tried to use Qualcomm chips, but a license fee was $1 million, which they were willing to pay. However, Qualcomm, with its many large customers, didn’t have the time to service them.
So Shanji chose a domestic alternative, using Unisoc’s W517 as the foundation. That was a compromise. In photo and video — high-frequency functions — the difference between the two chips became immediately apparent in user experience.
“You have no choice,” Chen said. “Either you wait a year for Qualcomm, or the market waits for you?”
Most companies’ first-generation AI glasses were launched with a “take a loss to make noise” mindset. Behind that is the real work of aligning capacity and yield between brands and contract manufacturers before a market can erupt.
From late last year, Xiaomi began ordering materials. Some supply lead times stretched as long as five months.
But how many units to stock was not something any domestic company could reliably estimate; Xiaomi had to build a sales model by working backward from target markets.
Li estimated that China has over 200 million pairs of glasses demanded annually, with 90% priced below 1,000 yuan; if 3–5% are above 2,000 yuan, that’s roughly 10 million pairs, and if Xiaomi captured 5% of that segment, it would be 500,000 pairs.
“Five percent is a relatively safe number,” Li said. “From our ecosystem experience, Xiaomi can take 20% in a given price segment, so 5% seems conservative.”
When they returned from the New Year and entered the profit-and-loss assessment phase, Li realized they would lose money no matter what.
As product lead, he had hoped to set the price at 2,199 yuan. The final price remained undecided until the day before launch; Lei Jun insisted on 1,999 yuan, the same as Xiaomi’s first phone.
Li recalled that the first batch of Xiaomi AI glasses sold out in days and some SKUs went out of stock. Jiemian’s sources indicate that around 20 days after release, the product had sold roughly 100,000 units.
Capacity scarcity can create a frenzy of stockouts; if the product experience is good, it calms consumer anxiety, but it can also create future problems the company will need to fix.
With funding coming in, Shanji expanded its team in the second half of 2024 to accelerate product development. Before tooling in October, the team spent nearly five months on detailed product design and hardware selection, and in December moved into AAC Technologies’ production line.
Because it had an earlier market position, Shanji’s first product A1 had significant buzz; after the December 19 launch, it sold out 50,000 units within 24 hours.
Considering manufacturing limits, the company initially set a sales page and for several hours debated whether to reopen the link. A1’s initial plan was 400 units per day, with an intended yield of 91% before scaling to 1,000 per day; they did not reach that yield on time.
After sales pressure, they reopened ordering and ultimately drew about 100,000 orders. But due to product and capacity issues, delivery cycles stretched long and social platforms accumulated negative user feedback.
This mix of startup urgency to land in the market and associated compromises, plus gaps between company understanding and user expectations, meant first-generation products had to pay the price and try to recoup with later iterations.
That’s the break-in process for new categories. Even Xiaomi’s strong imaging team produced results that many users criticized on social media; such gaps can only be closed over successive product generations.
As the new trend took shape, core component costs fell and supply-chain players paid more attention to the AI glasses category. Jiemian learned that after Xiaomi’s entry, the controller chip price dropped by about 40%. Contract manufacturers such as Goertek, Luxshare, Lens Technology and Longi became more invested at the BU level.
With these conditions, manufacturers could aim for higher targets. For many, the first generation had to be restrained and prudent.
Looking back at Xiaomi’s whole process with AI glasses, Li prefers to describe it as a “measured balance,” because many aggressive or risky designs require large investment and are dangerous in an unproven category.
This balance inevitably shapes big companies’ strategies and tactics for AI glasses. Strategically, a company must position itself for this possible next-generation entry point; tactically, the first battle must look good and cannot be an all-out, cost-blind wager.
暂无评论内容