Analysis of Dilemmas in the AI Glasses Industry and Breakthrough Strategies for Domestic Manufacturers

Based on the latest market data and industry information collected, I provide you with a systematic and comprehensive analysis report.

:

• Domestic Market
  : AI glasses shipments in 2025 are approximately at the million level, lower than expected
• Meta Benchmark
  : The Ray-Ban Meta series has accumulated shipments exceeding 2 million units by December 2024, with an estimated annual sales volume of 4 million pairs
• Gap Multiple
  : More than 4x sales gap

According to industry research, the AI glasses industry faces three core problems [1]:

Problem Type	Performance Characteristics	Specific Cases
Pseudo-demand	Function hollowing out, lack of real value	Early audio glasses only focused on listening to songs and making calls, essentially a glasses-shaped variant of Bluetooth headsets
Weak demand	Shallow solutions, pain point resolution is superficial	Prompting cannot be offline, translation recognition is inaccurate in noisy environments, still need to take out the phone to operate
Virtual demand	Lack of experience details, user purchase motivation is mainly for trial	Frequent charging, trouble with myopia glasses fitting, return rate as high as 40%

: This is one of the core bottlenecks for smart glasses to land [2]

• China has a large base of myopia population, and the myopia rate among high school students is extremely high, requiring higher technical requirements for myopia adaptation [1]
• Mainstream buckle/magnetic suction schemes will increase wearing weight and volume
• The fit scheme is technically difficult, with strict requirements on lens flatness, precision, thickness, and weight

:

Technical Scheme	Advantages	Disadvantages	Representative Manufacturers
External Hanging Type	Low cost, simple implementation	Increases weight and volume, affects aesthetics	Early products
Fit Type	Good aesthetics, integration	High technical difficulty, long customization time, temporarily unable to mass produce	In R&D
Magnetic Suction Type	Adaptable to degree changes, supports offline fine-tuning	Requires special lens design	Leading manufacturers like Rokid

:

• The founder is also a myopia user, so he paid attention to this demand from the beginning
• Any store can complete the degree adjustment, greatly facilitating users
• Eliminated the psychological barrier of “trouble with glasses fitting when purchasing a machine” [1]

:

Scheme	Advantages	Disadvantages
Type-C Interface	Strong universality	Prone to wear and tear due to plugging and unplugging, risk of damage when charging while using
Charging Case	Similar to TWS earphone experience	Need to take off the glasses frequently, interruption in use
Magnetic Charging	One-handed 0.5-second adsorption, can charge while using	Requires special interface design

Rokid’s magnetic charging scheme has undergone

, and the magnetic port material has undergone , finally adopting an anti-oxidation platinum-plated silver scheme [1].

:

• 100+ scene optimization algorithms
• 15-dimensional imaging evaluation system
• Solved 229 effect problems such as backlight and motion capture
• Achieved natural imaging of “bright parts not overexposed, dark parts have details”

:

• Knowledge distillation technology lightweight large model
• Voice assistant delay compressed to 249 milliseconds
• 100,000 hours of dialogue data training
• Accurate recognition in noisy environments, avoiding false wake-up

:

• “Scenario-based resource scheduling” scheme
• Dynamically adjust module priority in composite scenarios such as cycling navigation + listening to songs + taking photos
• Achieve seamless switching

[2]:

1. Customer Group Matching Degree
   : Smart glasses are mid-to-high-end priced (1,000-3,000 yuan), with low customer overlap with small optical stores
2. Customized Car Room
   : Smart glasses have complex myopia schemes, and small stores without back-end customized car room capabilities cannot undertake
3. Single Store Model Improvement
   : Taking Boshi Optical as an example, the single store sales have increased significantly after the smart glasses volume

(Cooperation with memories.ai)

According to network search information, the multimodal AI memory engine cooperated by Shanjitech and memories.ai has the following capabilities [3]:

• Full Data Collection
  : Real-time collection of multi-modal data such as video, audio, and images
• Vectorization Conversion
  : Convert raw data into unified embedded vectors for easy retrieval and analysis
• Real-time AI Search and Analysis
  : Supports fast retrieval, editing, and querying
• Automated Processing
  : Automatically generate transcripts, summaries, and to-do items

Shanjitech’s self-developed AI memory system “Lumou OS” includes [4]:

• AI Cloud Disk: Cloud data storage and management
• AI Flash Memory: Fast recording and intelligent review
• AI Agent Store: Application ecology and third-party services
• Cloud AI Center: Centralized AI processing capability
• Data Security System: Multi-level privacy protection

(Cooperation with Appotronics)

Advantages of Appotronics’ ALPD laser display technology [5]:

• Adjustable viewing angle projection system
• Suitable for multiple scenarios such as engineering, education, cinema, and automobile
• High brightness, high contrast, low power consumption

• Meeting Record
  : Automatic transcription, generate summaries and to-do items
• Customer Communication
  : Complete recording of dialogue content, avoid missing important information
• Time Management
  : Arrange meetings according to daily rules rather than just looking at the calendar

According to the analysis of the AI memory system, memory enables AI Agent to have cross-session continuity in real scenarios, able to remember past interactions and optimize future behaviors [6].

• Class Notes
  : Automatically organize class content
• Interview Recording
  : Research data collection and sorting
• Knowledge Management
  : Maintain context continuity across sessions

• Inspiration Capture
  : Record sudden ideas and creativity
• Project Collaboration
  : Full record of team communication
• Data Organization
  : Automatic classification and archiving

According to the application scenarios of AI memory pendant products [7]:

• Forgetful People/Elderly
  : Remind daily tasks, avoid missing important information
• Alzheimer’s Patients
  : Help them record life and assist memory
• Youth Education
  : Follow the child’s perspective for dialogue recording

Comparison Dimension	Traditional AI Glasses	Shanjitech AI Memory Glasses
Core Value	Fixed hardware parameters	Memory data increases in value with time accumulation
Usage Stickiness	Easy to idle after trial	The more you use it, the more valuable it becomes
Competitive Barrier	Easy to imitate	AI memory system + data assets

: The AI memory system has the ability of , maintaining context continuity across different sessions, devices, and time periods [6], which means that the longer the usage time, the richer the memory data, the deeper the AI’s understanding of the user, and the higher the value.

:

• User’s memory data becomes personal digital assets
• Memory is not simply storing chat records, but a persistent internal state that evolves with each interaction [6]
• Once dependent on the AI memory function, the conversion cost is extremely high

:

• Lumou OS system builds a complete ecosystem
• AI Agent Store provides third-party services
• Cloud AI Center provides continuous computing power support

According to the technical analysis of the AI memory system [6], the three pillars of memory are:

• State
  : Understand the current situation
• Persistence
  : Retain knowledge across sessions
• Selectivity
  : Judge which information is worth remembering

These three pillars together give AI

, and this continuity capability has a high technical threshold.

Pain Point Type	Traditional Solution	Shanjitech AI Memory Solution
Pseudo-demand	Function stacking	AI memory is a real just need
Weak demand	Shallow tool	Deep integration of work processes
Virtual demand	Lack of details	450mAh battery life + optimization algorithm + myopia customization scheme

:

• Retail price: 1,499 yuan
• Co-creation version: 999 yuan
• Has price advantage compared to competitors

[4]:
Key components are basically selected from domestic brands:

• Main control chip: Unisoc W517
• Storage chip: Biwin Storage
• Camera: 16MP
• Battery: 450mAh

This localized supply chain strategy helps:

• Control costs
• Ensure supply stability
• Rapid product iteration

Based on the searched Rokid Glasses information [8]:

Function Dimension	Rokid Glasses	Shanjitech AI Memory Glasses
Display Technology	Micro LED + Diffraction Waveguide	Dragonfly Optical Engine (ALPD Laser Display)
Core Function	Navigation, translation, prompting, health reminder	All-weather active AI memory
Interaction Mode	Head movement, gesture, touchpad	Active recording + intelligent retrieval
Data Value	Instant tool use	Lifelong memory accumulation
Camera	12MP	16MP
Weight	49g	Not announced

According to iResearch survey,

[1].

• Traditional AI Glasses
  : Functions such as prompting and translation have certain value, but they are not core just needs
• AI Memory Function
  : Solve the real pain point in the era of information overload—can’t remember, can’t find, can’t use

• Traditional Hardware
  : The highest value when bought, depreciates over time
• AI Memory
  : The longer the usage time, the richer the memory data, the deeper the AI understanding, the higher the value

According to the definition of the AI memory system [6]:

“Memory is not simply storing chat records, nor is it stuffing more data into the prompt box. It is a persistent internal state that evolves with each interaction, and can still provide continuous context for AI even after weeks or months.”

This continuity makes the conversion cost extremely high once users use it.

Multimodal AI memory involves:

• Real-time multi-modal data processing
• Intelligent screening and dynamic forgetting
• Cross-session context maintenance
• Privacy protection technology

These technologies have high thresholds and are not simple function stacking.

:

• Involve others’ privacy, need clear notification mechanism
• Differences in recording laws in different regions (e.g., some states in the US require multi-party consent) [7]
• Social acceptance: Psychological resistance of the photographed object

[7]:

• Consent Mode: Only record the voice of conversationalists who explicitly agree verbally
• Data Encryption: All recordings and transcripts are encrypted and stored
• LED Prompt Light: Lights up when recording to ensure others know

:

• All-weather recording has extremely high requirements for battery life
• Need to optimize algorithms to reduce power consumption

:

• Multi-modal AI processing requires strong computing power support
• Need to find a balance between local and cloud

:

• Cloud storage and transmission costs of massive data
• Need efficient data compression and retrieval technology

:

• Users need time to adapt to the lifestyle of all-weather recording
• Need to establish trust relationships

:

• It may be difficult to immediately perceive the long-term value of AI memory in the early stage
• Need to show value through usage scenarios

:

• Users’ concerns about their data being recorded
• Need to establish a credible privacy protection mechanism

1. Localization of Myopia Adaptation
   • Optimize magnetic suction and fit schemes according to the large base of myopia population in China
   • Cooperate with leading lens manufacturers (such as Conant Optics) to provide customized services

2. Wearing Comfort
   • Lightweight design (target <50g)
   • Long battery life (target >12 hours)
   • Optimize weight distribution, nose pads, and frames for Asian head shapes

3. Charging Convenience
   • Magnetic charging scheme
   • Charging case scheme
   • Ability to charge while using

1. From Pseudo-demand to Real Just Need
   • Focus on real pain points such as AI memory
   • Avoid function stacking
   • Deeply挖掘 user core needs

2. From Weak Demand to Deep Integration
   • Deeply integrate work processes
   • Seamlessly connect with commonly used applications
   • Provide end-to-end solutions

3. From Virtual Demand to Detail Optimization
   • Solve small troubles such as charging, glasses fitting, and privacy
   • Establish user co-creation system
   • Rapid iteration and optimization of products

1. Concentrate on Head Optical Stores
   • Cooperate with head chains such as Boshi Optical and Baoshida Optical
   • Establish customized car room capabilities
   • Optimize single store model

2. Build Application Ecosystem
   • AI Agent Store attracts third-party developers
   • Cooperate with industries such as enterprise services, education, and medical care
   • Open API and SDK

3. Establish Brand Awareness
   • Emphasize differentiated value (all-weather AI memory)
   • Create usage scenario cases
   • Establish user community

1. Technical Landing Stability
   • Whether the AI memory function is really easy to use
   • Whether the battery life meets all-weather use
   • Whether the computing power meets real-time processing needs

2. Privacy Protection Credibility
   • Establish multi-level privacy protection mechanisms
   • Obtain authoritative certification
   • Transparent privacy policy

3. First Batch User Experience
   • Optimize product details
   • Respond quickly to user feedback
   • Establish word-of-mouth communication

1. Effectiveness of User Education
   • Let users understand and accept the value of AI memory
   • Show value through usage scenarios
   • Reduce user usage threshold

2. Ecosystem Construction
   • Richness of AI Agent Store
   • Third-party application integration
   • Enterprise-level service capabilities

3. Scalability
   • Supply chain guarantee
   • Delivery capability (target to ship more than 100,000 units within one year)
   • Cost control

1. Data Assetization
   • Let users realize the value of memory data
   • Provide data export and migration services
   • Establish a data value quantification system

2. Continuous Iteration
   • Continuous optimization of AI algorithms
   • Continuous enrichment of functions
   • Continuous upgrading of hardware

3. Competition Barrier Construction
   • Technical patent layout
   • Data barriers
   • Ecosystem barriers

[2].

In the past decade, the demand side of smart glasses has been weak, not because consumers are not interested in AR or “first-person information acquisition” capabilities, but because the

—the core experience, form, weight, and battery life of the product have not reached the critical point of “good enough to use”.

Since 2024, driven by continuous iteration of large models and improvement of wearing comfort, devices have broken through traditional interaction limitations, with multi-modal perception capabilities, realizing real-time capture and feedback of voice commands and visual information [2].

1. Real Just Need
   : AI memory solves the real pain point in the era of information overload, not function stacking
2. Value Increment
   : The longer the usage time, the richer the data assets, the higher the value, forming a positive cycle
3. High Conversion Cost
   : Memory data becomes personal assets, cross-session continuity makes conversion cost extremely high [6]
4. Technical Barriers
   : Multimodal AI memory engine has high technical thresholds, not simple function stacking
5. Ecosystem Lock-in
   : Lumou OS + AI Agent Store + Cloud AI Center build a complete ecosystem

1. Privacy Risk
   : Privacy and legal risks of all-weather recording need to be properly resolved
2. Technical Implementation
   : Technical challenges such as battery life, computing power, and storage need to be overcome
3. User Acceptance
   : Need time to cultivate user habits and trust
4. Increasing Competition
   : Giants such as Meta and Google are also laying out similar functions