AI Glasses: Navigating the Dark

In Shanghai, the July air was thick and viscous, like syrup that refused to dissolve.

Standing at the entrance of the West Bund International Convention and Exhibition Center, the queue of people waiting to enter snaked back for dozens of meters; sweat trickled down temples as the line inched forward at a snail's pace. Every few seconds, the security gate emitted a short, sharp beep, admitting a drenched visitor.

Inside the venue, the crowd was dense in a different sense.

Over twenty AI eyewear manufacturers packed their booths into the better part of the exhibition hall; the air was thick with the mingled scents of disinfectant wipes and metallic frames. Crowds swarmed every booth; staff members repeatedly wiped fingerprints off the lenses, only for the glasses to be snatched up and tried on by the next person the moment they were put back.

A reporter from Zhixiedao (Smart Machine Island) squeezed into the crowd and took a pair of glasses from someone’s hand. The moment they put them on, real-time Chinese subtitles appeared in their field of vision, translating the words of a German exhibitor standing opposite: "This product is very interesting."

After taking them off, another pair was handed over; this time, the reporter saw a heart-rate graph hovering in the top-right corner of their view. Switching to yet another pair, a to-do list popped up on the lens, reminding them: "14:00 – Interview with the founder."

Standing in the aisle with six or seven promotional flyers in their pocket—each bearing nearly identical claims of being "smart," "lightweight," "hands-free," and featuring an "AI assistant"—the reporter was left with just one question after trying seven or eight models: Would I actually wear this when I go out tomorrow?

Yet, at the booths scattered across the WAIC venue, no one could provide a convincing answer to that question.

I. The Pearl River Delta Snaps Its Fingers

Four years ago, if someone had told you that eyewear would become the next major computing platform, you might have thought they were talking science fiction. Today, however, at a one-meter-wide counter in Shenzhen’s Huaqiangbei district, you can pick up a pair of AI glasses equipped with a camera for just 300 yuan.

This is no miracle; it is simply the result of China’s supply chain capacity overflowing.

From camera modules in Dongguan, optical lenses in Jiangxi, and micro-batteries in the Jiangsu-Zhejiang region to contract manufacturing in Shenzhen, a complete industrial chain has grown to an astonishing scale.

Inside the WAIC exhibition hall, Zhixiedao spoke with an ODM factory based in Shenzhen’s Nanshan District. Boss Chen told us: "If a client suggests a modification in the morning, we can have a sample ready by the afternoon; the entire process from project initiation to mass production can take as little as a month and a half."

He casually picked up a temple module and pried it open to show us the internal structure: a main control chip, Bluetooth module, power management chip, microphone, and speaker—all integrated onto a circuit board narrower than a finger.

Why is it so fast? Since most components rely on readily available standard designs and molds, there is no need for ground-up development; instead, manufacturers simply make trade-offs based on existing solutions—deciding which features to include, how to reduce weight, or which chip supplier to choose.

Over 80% of global smart glass manufacturing is concentrated in China. To find a contract manufacturer, Meta turns to Goertek; realizing optical solutions requires working with SeeYA Technology and Sunny Optical; and display chips come from JBD. These suppliers are located within a two-hour drive of one another; if a problem arises with a component, a boss can simply hop on an e-scooter to discuss it in person.

Behind that 80% figure lies a dense, interconnected industrial network.

In a pair of AI glasses equipped with a display, optical display components typically account for 40% to 50% of the cost, while chips make up another 20% to 30%—together representing over 70% of the total. Whoever controls these two elements holds the lifeblood of the entire industry.

The optical module serves as the "eyes" of AI glasses, determining their display quality, weight, and power consumption.

The core technology in this field is optical waveguide. In Huzhou, Zhejiang, a startup named Zhige Technology—spun out of Tsinghua University’s Department of Precision Instruments—has overcome a challenge that long plagued the global AI eyewear industry: the "rainbow effect." They also managed to reduce the weight of their diffractive optical waveguide lenses to under 4 grams—or, as they put it, "less than the weight of a spoonful of salt."

The chip serves as the brain of AI glasses. It determines the speed at which AI models run, the device's battery life, and the complexity of the scene understanding it can handle.

Chinese companies are heavily invested in this sector as well. The BES2800 chip from Zhuhai-based Bestechnic (Hengxuan) has become a key component for Meta’s Ray-Ban glasses, and Rokid’s new AR glasses are powered by Bestechnic’s 6nm chip. The RayNeo V4 utilizes a dual-chip setup—combining Qualcomm’s AR1 with Bestechnic’s BES2800—and ByteDance’s second-generation Doubao AI glasses have adopted this same configuration.

Notably, chip architecture is evolving. AI glasses are shifting from single-chip to dual-chip designs: a low-power chip handles standby modes and basic interactions, while a high-performance chip manages AI inference and complex computations.

Rokid’s new product features Bestechnic’s 6nm chip as the main controller, paired with an external Qualcomm Snapdragon "Ultimate" spatial computing coprocessor. This "dual-chip, dual-system" architecture is becoming the industry standard.

As of July 20, over 60 A-share companies within the AI glasses supply chain had released their semi-annual earnings forecasts; notably, upstream companies were the first to reap the rewards of this boom.

Profits in the upstream chip sector have surged: BIWIN Storage projects a net profit attributable to the parent company of 7.0–7.5 billion yuan for the first half of the year—a year-on-year increase of over 32-fold. Shannon Microelectronics expects a net profit of 3.5–4.0 billion yuan, up more than 21-fold year-on-year. Other SoC and storage controller chipmakers—such as Fudan Microelectronics, Montage Technology, Rockchip, and Allwinner Technology—have also generally forecast significant profit growth or a turnaround from losses to profits.

From optics and chips to contract manufacturing, every link in the supply chain is accelerating. This has led to an awkward dilemma: the more brands rely on the supply chain for rapid product launches, the more homogenized their products become; the more homogenized the products are, the harder it is to answer why a user should choose one brand over another.

II. No One Has Fully Tapped the Potential of AI Glasses

The explosion of AI glasses has largely benefited from the dividends accumulated during the smartphone era.

Companies like Goertek, Luxshare Precision, and Lens Technology—giants in contract manufacturing during the smartphone era—have successfully transferred their lean manufacturing capabilities to the AI glasses sector. For instance, Lens Technology managed to bring the weight of AI glasses down to 49 grams and kept assembly gaps so minimal they are virtually imperceptible to the naked eye; meanwhile, the packaging for Meta’s Ray-Ban AI glasses bears the "GTK" mark—Goertek’s contract manufacturing code.

The role of these ODM (Original Design Manufacturer) firms is expanding upstream; an increasing number of stages—ranging from ergonomic design and chip platform adaptation to the integration of near-eye display optics, system development, and mass production delivery—are being consolidated into comprehensive, turnkey solutions.

For internet companies, traditional eyewear brands, and AI startups, adopting a mature reference design and customizing it around aesthetics, features, and AI models is often far faster than building a hardware team from scratch.

This raises a key question: when asked by Zhixie Dao (Smart Machine Island) when the industry would truly take off—given his years of experience in contract manufacturing—Boss Chen paused before asking in return, "How do you think smartphones took off back in the day?"

Nokia wasn't the first to make a mobile phone; it was the iPhone that redefined what a phone actually was. The hardware had always existed, but no one knew what it should look like or what it should be used for. The situation with AI glasses is much the same; what they lack is someone to clearly define "what they are" to the public.

When all core components are available on the open market, the hardware itself ceases to be a competitive moat. Any achievement regarding weight, battery life, or packed-in features can be matched by competitors within six months.

At the World Artificial Intelligence Conference (WAIC), the question "Why should I buy your AI glasses?" was repeatedly raised. Manufacturers offered a variety of answers, but they all converged on the same themes: "What functions can I provide?" and "In what scenarios can I solve your problems?"—features such as real-time translation, teleprompting, navigation, meeting minutes, memory storage, and emotion sensing.

Everyone is digging a well, searching for that one "killer" use case that makes wearing the device indispensable for users. Yet, so far, no one has struck water.

After spending two hours browsing the exhibition hall, "Zhinxie Dao" (AI-Mech Island) discovered that the core specifications of AI glasses were strikingly similar: Qualcomm AR1 or Bestechnic chips, a weight of around 40 grams, multi-microphone arrays, and AI voice assistants. The only real differences lay in exterior design and software tuning. As "Boss Chen" put it, if you were to take apart 80% of the AI glasses currently on the market, the internals would look virtually identical.

Of course, even though the supply chain has all the necessary components ready, simply stuffing them into the temples of the glasses isn't enough.

The supply chain can provide the best hardware achievable under current technological constraints. Yet, there is a ceiling to this "best": if battery energy density doesn't improve, weight cannot be reduced; without a breakthrough in chip manufacturing processes, one cannot have both high computing power and long battery life; and until optical solutions mature, display quality and a slim, lightweight form factor remain fundamentally at odds.

Supply chain optimization can only push us closer to these limits; it cannot break through them. A breakthrough requires definition—telling the supply chain exactly which direction to take.

III. The Three Walls Hemming In AI Glasses

Manufacturers are answering the question "What should AI glasses be?" in their own ways, but more fundamental than the debate over product strategy are the three "walls" facing AI glasses—obstacles that no one can bypass.

The first wall is the "impossible triangle" of hardware.

To achieve a weight of just 26 grams, Liweike sacrificed the display screen and high-performance local chips; Moonix managed 14.9 grams by voluntarily dropping the camera and translation functions; and Qianwen’s S1 glasses—which do feature a display—had to adopt a hot-swappable battery design to ensure adequate runtime, meaning users have to carry spare batteries in their pockets when they go out.

Weight, battery life, and performance are inherently mutually exclusive. Want to add a display? Both weight and power consumption rise. Want a larger battery? Weight skyrockets. Want to keep them lightweight? You have to offload computing power, making even local AI inference impossible.

Currently, AI glasses with displays generally offer only 2 to 5 hours of battery life per charge, whereas screenless models can last over 8 hours. Consumers have voted with their wallets: in the first half of 2025, sales of screenless AI glasses surged by 463%. When faced with a choice between mere functionality and wearability, users unhesitatingly chose the latter. This "trilemma" exists because each of the three sides faces its own physical ceiling.

Battery energy density improves by only a few percentage points annually, whereas the computing power required for AI models doubles every year; chip manufacturing processes are approaching physical limits, with each generation beyond the 3nm node becoming more expensive and difficult to produce, while yielding diminishing returns; and optical solutions must simultaneously achieve thinness, high brightness, full-color display, and low cost—a combination of requirements that no single solution has yet managed to satisfy.

Even after overcoming one bottleneck, another challenge inevitably lies ahead; these issues are rooted in fundamental science, and the pace of scientific breakthroughs is not dictated by the product launch schedules of any single company.

The second barrier is trickier than the first.

Technical hurdles can be overcome, but once social trust collapses, it is difficult to rebuild. At the Rokid booth, as staff demonstrated the "look-to-pay" feature, a bystander whispered, "So, can it also take a picture of me?" The unease behind that remark reflects the industry's deepest anxiety.

In late 2024, two Harvard students conducted an experiment using a pair of Ray-Ban Meta glasses: they snapped a photo of a stranger, used facial recognition to retrieve the person's name, and cross-referenced databases via a large language model. In less than a minute, they had the subject's home address, phone number, occupation, and relatives' details. Although they did not release the code, they issued a warning: embedding facial recognition into smart glasses crosses an inviolable red line.

Meta’s response was to add a white LED light to the frame—flashing once when taking a photo and continuously while recording. Yet, this safeguard is woefully fragile; a simple 9.9-yuan opaque sticker placed over the light renders it invisible.

The core value of AI glasses lies in that camera; without it, real-time translation cannot "see" a menu, navigation cannot "read" road signs, and AI cannot identify objects before you. Yet, with it, you become a walking surveillance camera wherever you go.

This contradiction is not merely a product issue; it is a structural dilemma for the entire product category. It requires not just technological upgrades, but a restructuring of societal rules regarding the right to film in public spaces, data ownership, and the boundaries of informed consent.

An industry cannot wait for regulations to be fully established before developing, yet its progress can be stalled by rules at any moment—and the pace of regulatory reform lags far behind the speed of hardware iteration.

The third wall is simultaneously the softest and the hardest; it resides within the consumer's mind.

AI glasses have launched with over 200 new AI features, yet RayNeo CEO Li Hongwei revealed at the launch event that the long-term usage rate is less than 6%. Return rates on platforms like JD.com and Tmall hover between 30% and 40%, while on Douyin, the figure reaches as high as 50%.

Users might feel excited for two or three days after purchase, only to realize that the translation function is useful only when traveling abroad, the novelty of the camera wears off after a week, and checking the weather via voice assistant is less convenient than simply glancing at their phones. Consequently, the glasses end up in a drawer, never to be taken out again.

It is not a matter of price or the number of features; the fundamental issue is that users simply cannot find a reason to wear them in the long run. The industry is desperate to explain "what I can do," while users are wondering "why do I need this"—and between these two perspectives lies a chasm that no one has yet managed to bridge.

The first wall compromises the product experience—making it too heavy, lacking sufficient battery life, or not smart enough. The second wall limits social acceptance—users fear being mistaken for voyeurs when wearing them in public or worry about being barred from certain venues. The third wall undermines user retention—the initial novelty fades after three days, leaving the device to gather dust.

These three walls reinforce one another, creating a vicious cycle: hardware flaws exacerbate privacy concerns; privacy anxieties limit usage scenarios; a lack of scenarios prevents users from finding value; and this lack of value, in turn, leaves brands with no incentive to invest resources in overcoming hardware bottlenecks.

None of these walls can be dismantled in isolation; they are inextricably linked, locking one another in place. The buzz at the WAIC event serves, to some extent, merely to mask the silence surrounding this intractable dilemma with noise. IV. Conclusion

After the first day of WAIC concluded, a reporter from "Zhixie Dao" (Smart Machine Island) ran into an industry analyst friend at the venue exit. Clutching a stack of documents and looking weary, the friend sat on the steps with the reporter for a ten-minute chat. One remark made by the friend gave the reporter plenty of food for thought: "Don't you find it strange? Alibaba, Tencent, and ByteDance—even though ByteDance didn't have a booth today—have each offered a different answer regarding AI glasses, yet it seems none of them has truly convinced the market."

At WAIC, Alibaba’s Qwen announced an upgrade of its AI glasses into "AI agent glasses," emphasizing all-day perception and proactive service; Li Weike and Tencent Cloud’s WorkBuddy jointly launched "X-AI Memory Glasses," betting on long-term memory capabilities; meanwhile, although ByteDance’s first-generation Doubao AI glasses were halted internally, industry rumors suggest their second-generation model is pursuing an ultra-lightweight design—weighing just over twenty grams and omitting speakers entirely.

With 50 million yuan invested in R&D and a planned production capacity of 100,000 units all scrapped, why was ByteDance’s first-generation Doubao AI glasses project cancelled? Because the product was too similar to existing market solutions; consumers couldn't distinguish it from offerings by Xiaomi, RayNeo, or the "white-label" products from Huaqiangbei.

This cancelled product serves as a mirror, reflecting the awkward predicament facing the entire industry. AI glasses remain in a phase where individual products exist, but a distinct product category has yet to fully emerge. Consumers recognize the term "AI glasses" but struggle to define exactly what they are: Are they talking sunglasses? A camera worn on the face? Or a hands-free mobile phone?

History has repeatedly demonstrated a single truth: Google wasn't the first search engine, the iPod wasn't the first MP3 player, and Taobao wasn't the first e-commerce platform. They succeeded for one reason: they were the first to define their product category, telling consumers—in the simplest, clearest terms—exactly who they were and what problem they solved.

In June 2026, Mark Zuckerberg launched Meta’s own-brand smart glasses for $299. His definition of the category was simple: AI glasses are, first and foremost, glasses; the intelligence is merely an added value. Zuckerberg believed that within a few years, it would be hard to imagine a world where most glasses weren't AI glasses.

That definition might not be objectively correct, but it is certainly clear; it envisions a future where buying AI glasses is as routine as buying a pair of sunglasses.

Meanwhile, domestic manufacturers are still fighting their own separate battles. A reporter might carry several brochures from the WAIC (World Artificial Intelligence Conference) in their pocket, each telling a different story: Li Weike is betting on long-term memory, aiming to become your "second brain"; Rokid is pushing its YodaOS, building an ecosystem free from the influence of Android or iOS; and Moonix is obsessively focused on ultra-lightweight design (just 14.9 grams), betting that once you put them on, you won't want to take them off.

No single path is clearly right or wrong. The issue isn't about correctness; it’s about what happens when a consumer sees these competing pitches. Ultimately, only one thought will remain in their mind: Which one should I commit to buying?

By 2026, global shipments of AI glasses are projected to exceed 16 million units, with Meta alone capturing 60% to 70% of the market. Apple is expected to launch its first pair of AI glasses in 2027, Google’s Android XR ecosystem is taking shape, and Samsung’s Galaxy Glasses are poised for launch.

The field of major players will grow, and products will mature. Yet, regardless of who wins, the prerequisite for victory remains the same: the company that first answers the question "What exactly are AI glasses?" will hold the key to defining this market—a sector worth hundreds of billions.

Until then, all these various players are merely groping in the dark. The WAIC spotlights may be bright, but the true dawn has yet to break.

Source: Zhixiedao (Smart Machine Island); Author: Huo Rujun