Written by | Wen Yehao Edited by | Wang Pan
Since the beginning of this year, AI glasses appear to have hit their first major wall.
According to IDC data, domestic smart glasses shipments in Q2 2026 totaled approximately 611,000 units, a year-on-year decline of 8.0%. This marks the first negative growth since the sector began its rapid ascent. Meanwhile, the global market continued to surge at a rate of 35.3% during the same period.
While overseas markets were booming, the domestic market shivered.
Prior to this, no other category believed more fervently that it would become the "next-generation smartphone" than AI glasses; nor had any category been tripped up so severely by such belief.
Vendors went all-in, capital piled in, and product launches occurred one after another, as if the replacement of smartphones by AI glasses was merely a matter of time. However, the excitement was entirely on the supply side, while demand remained lukewarm. The people wearing these glasses out daily seemed to be the same small group of tech enthusiasts and early adopters.
The number of brands increased year by year, and new models launched one after another, yet the user base did not experience an explosion—there were too many AI glasses, and geeks were seemingly running out.
Consequently, players who once touted "glasses for everyone" began a collective retreat: some adopted a minimalist approach, attempting to revert products to standard eyewear; others dove into niche verticals such as automotive, translation, meetings, and accessibility.
Minimization and niches allowed the industry to survive but failed to satisfy their appetite for becoming the primary entry point. Against this backdrop, these players retreated while quietly placing their bets on Personal Agents.
From hitting walls and retreating to the current counterattack, the first two acts are gradually closing, while the third act seems to have just begun.
Hitting the Wall: Too Many AI Glasses, Not Enough Geeks
Although quarterly data may fluctuate due to new product releases and launch schedules, judging the entire industry based on a single quarter's decline is unfair.
However, for a track with high growth expectations, the fact that the Chinese market—a fundamental base—declined rather than grew amidst dense vendor launches and intensified marketing suggests a certain hollowness in the hype.
From a product perspective, expansion has been evident. Over two years, AI glasses weight dropped from 50 grams to the low 30s, chip counts increased from one to two or three, and features expanded from a dozen to over two hundred.
A major catalyst for this frenzy was capital.
Investors have always been hungry for the concept of an "entry point," willing to bet early on an unrealized future. However, paying consumers remain far more rational, evaluating products against standard consumer electronics criteria.
In the early stages of a category, securing funds and positioning oneself is a viable strategy. The problem lies in the fact that the upstream supply chain for AI glasses is highly mature, with main controllers, optical waveguides, and sensor solutions being largely homogeneous. Apart from a few players with genuine R&D capabilities, there are limited areas where brands can truly differentiate themselves.
The so-called "War of a Hundred Glasses," when viewed by consumers, boils down to permutations of accessories and features, along with price testing among competitors.
From a demand perspective, those willing to wear AI glasses out daily are consistently the same group: tech geeks, digital bloggers, and hardware enthusiasts.
Data from iiMedia Research indicates that the primary consumers of AI glasses are AI enthusiasts and technology early adopters, accounting for 32.5% and 32.2% respectively, totaling approximately 65%.
This group is not only willing to spend but also the most eager to pay, itching to buy upon seeing new products. Some hoard multiple pairs simultaneously, while others trade in old ones for new.
But this small segment does not represent the general public.
A principle in product thinking is "fool-proofing," meaning designs assume users are novices, allowing them to use the product effectively without reading manuals. The more successful mass-market products are, the more they dare to assume users are "dumb," making operations as intuitive as possible.
By this standard, many so-called AI glasses serve almost as cautionary tales.
An industry insider told Photon Planet that previous products defaulted to male head shapes and weight distributions. Women, who typically have narrower heads and are more sensitive to weight, were largely excluded, leading to issues like ear discomfort. Only after vendors recognized this problem did they begin adding temple angle adjustments in subsequent products.
XREAL founder Xu Chi also stated in an interview that many domestic vendors lack "professionalism," showing insufficient patience and respect for complex hardware categories. Upon seeing frontier players release products, they quickly follow suit, disassemble, replicate, and "realize" the product using supply chain combinations, yet their understanding of underlying issues remains shallow.
Given the currently high return rates for AI glasses, the overall user experience remains insufficiently robust.
While many enthusiasts enjoy participating in improving new technologies—actively finding bugs, envisioning scenarios, and providing feedback—the general public lacks this patience and obligation. They want products that work out of the box without requiring maintenance. If a product demands active adaptation, tolerance for errors, and self-justification for wearing it, their choice is often simple—they don't wear it.
Supply accelerated by capital colliding with unopened demand cannot naturally support the weight of being the "next-generation smartphone."
Once the well of geek adoption runs dry, the first wall stands before all vendors.
Retreat: Big Entry Point, Small Business First
After hitting the wall, the industry took an honest step: quietly changing the target audience from "everyone" to "a specific group." Having previously touted the next-generation smartphone, they now retreated to faithfully build specialized tools for vertical demographics and scenarios.
This clarity was forced by reality.
The retreat paths boil down to two directions: subtracting from the product or specializing in scenarios.
RayNeo, the top shipper, made the boldest cuts with its RayNeo iO, removing the camera and speaker, reverting display to monochrome green, and reducing total weight to 34 grams.
For many users, cameras and speakers are standard components of AI glasses; their absence feels incomplete. When the RayNeo iO launched, many social media users were discouraged, criticizing it as "gutted."
Despite this, the RayNeo iO sold out.
The logic is simple: acknowledge that it must first be a pair of glasses before anything else. Regarding speakers, TWS earbuds perform better; regarding cameras, multimodal AI has not yet reached the stage where first-person perspectives can be fully utilized. These two components occupy weight and cost without playing to the glasses' strengths. What truly belongs to the lenses in front of your eyes is near-eye display.
After cutting, the core value proposition became clearer: display, microphone, and AI.
XREAL pursued a different kind of subtraction. It didn't rush to convince users of a future replacing smartphones but returned XR to an ancient need: humans always require a screen, just not a fixed one.
If a projector doesn't fit at home, glasses provide a giant screen; on business trips or travels, familiar visuals go wherever you do. Its logic is to untether needs previously pinned to living rooms from fixed spaces. The story isn't sexy, but it resembles what consumer electronics should look like.
An XR practitioner noted that near-eye display hits the most fundamental starting point of this category: "Before buying Vision Pro, many imagined working out or doing 3D design inside it. Once purchased, however, most used it as an external monitor for office work or simply for watching videos."
The second path is drilling into gaps where smartphones cannot reach.
General scenarios seem grand but fall squarely within smartphone manufacturers' range. Startups engaging in direct confrontation have little chance of winning. Viable footholds exist where hands are occupied, sensory gaps exist, or smartphones struggle to adapt.
Thus, players found their own niches. Rokid entered the automotive space, partnering with GAC to project navigation and vehicle status into the driver's field of view, integrating Tencent WorkBuddy and WeChat Pay to capture commuting and office scenarios. Nanobox focused on classrooms, emphasizing lecture notes and anti-forgetting vocabulary memorization. Lingliang Vision deepened its focus on meeting interpretation and hearing-impaired assistance, with one system replacing a room full of interpretation equipment.
Rokid Founder and CEO Zhu Mingming once said: "Many people buying glasses are no longer buying glasses, but buying capabilities, much like many people buying a Switch solely to play Zelda."
This implies that players temporarily set aside the obsession with being "all-capable," pursuing instead being "indispensable." If a pair of glasses becomes irreplaceable in scenarios like real-time translation, teleprompting, or accessibility subtitles, it can capture the first batch of mainstream users.
There is an even more thorough form of retreat: abandoning user acquisition entirely to pivot toward data collection, acting as "shovels" for another boom.
Currently, embodied intelligence lacks first-person perspective data showing how humans work. Since glasses sit exactly where human eyes are, some players are transforming terminals worn by humans into "glasses" that feed data to robots.
For example, PICO integrated with NVIDIA's Isaac Teleop architecture at GTC, enabling PICO 4 Ultra to directly record first-person operational data. PiMax co-built an embodied data platform with universities and secured strategic investment from Meifeng Technology, a subsidiary of AgiBot.
This means facing the chasm between geeks and the masses, players chose to bypass it—not confronting the mainstream market head-on, but jumping directly to niche markets across the divide to survive first.
Commercially, this is pragmatic. Product repurchase rates, stickiness, and reputation are far more solid than generic "geek toys." However, choosing retreat comes with a cost. Once AI glasses position themselves as "tools for vertical scenarios," they actively abandon the narrative of being the "next-generation computing entry point."
A better hearing aid, teleprompter, or ordering device allows the industry to survive but does not ignite it. Thus, everyone unwilling to settle for being just a tool must find a new banner.
Counterattack: New Stories, Old Problems
Retreated to the corner, the industry placed its remaining chips on Personal Agents.
In September this year, shortly after Meta released Muse, it announced integrating Muse into its own AI glasses. In October, following Meta, Alibaba's Qwen also declared at a hardware launch event that AI wearables, including AI glasses, would serve as extensions of Personal Agents in the physical world.
This is one of the few anchors capable of elevating glasses from a "collection of functions" back to a "smart entry point," offering a rare opportunity for AI glass players unwilling to remain mere tools to reshape their narrative.
A key advantage of AI glasses in the era of Personal Agents is "presence."
For an Agent to understand you and act on your behalf, it must continuously perceive and comprehend you. Smartphones reside in pockets, requiring users to actively pull them out to record things; what they see is only what users allow them to see. Only glasses, persistently attached to the first-person perspective, synchronously seeing, hearing, and recording, can feed the Agent this scarcest resource.
Beyond Personal Agents, the application ecosystem that AI glasses struggled to build over three years collided with the latest AI narrative hotspot: AI Coding.
Previously, developers were unwilling to create apps for such a small market. An XR app entrepreneur told Photon Planet that financing for the XR application side has remained sluggish compared to hardware in recent years.
However, as AI Coding matures and barriers lower, personalized applications will inevitably engage in a "close encounter" with AI glasses.
According to insights from Photon Planet, top-tier vendors are already attempting to incorporate AI Coding capabilities into next-generation products. Rokid's AIUI Studio, launched in September, addresses the same issue.
Yet, narratives aside, how AI glasses should truly evolve remains unresolved.
RayNeo Innovation CEO Li Hongwei once provided a figure: the long-term usage rate of so-called AI features is less than 6%. Meta, which should have been the benchmark case for this logic, ended up serving as a cautionary tale.
As disclosed earlier by Alex Himel, VP of Wearables at Meta, only about 10% of the time users wear glasses is spent utilizing smart features; the remaining 90% is driven by aesthetics and weight, treating them as ordinary eyewear. Long-term wearers also found that pure AI Q&A is the aspect where novelty fades fastest—users actively ask questions in the first month but prefer pulling out their phones six months later.
On the other hand, vision validity is one thing; whether glasses can withstand it is another.
The "all-day intelligence" sought by Personal Agents and the "all-day wearability" essential for glasses are both difficult to achieve under current AI glasses hardware conditions.
To be worn on the face, glasses compress weight, volume, heat dissipation, and battery to their limits. The chips included are essentially downgraded versions of smartphone or watch processors. The limited edge-side computing power suffices only for lightweight tasks like wake-up and recognition, unable to run an agent that requires real-time understanding of visuals and audio alongside inference.
If edge computing fails, move to the cloud. But uploading audio/video and returning results incurs latency, network dependency, and token costs—someone must pay the bill.
Battery life is similarly critical. The premise of Personal Agents is Always-On, yet glasses are the terminal least tolerant of Always-On. If AI, recording, and photography operate at full capacity, battery drains within one or two hours. While swappable batteries or portable charging cases offer partial solutions, they still cannot support high-frequency usage.
In engineering, there are no dead ends, only ways around them. A coprocessor consuming a few milliwatts might keep glasses half-awake, allowing the Agent to maintain "presence."
However, the polishing of weight, optics, design, and applications from the previous round is not yet complete, while a new round of exploration centered on Agents has already begun. This means vendors are destined to chase an "immature" Agent on a pair of "not yet good enough" glasses—the road to the endgame remains long and arduous.