November 18，Chinese AR glasses company Rokid has unveiled its latest AR glasses, the Rokid Glasses.

(Photo Credit: JBD)

The Rokid Glasses boast a stylish design, weighing just 49 grams, and are comfortable for all-day wear.

A standout feature is the integration of Alibaba’s Tongyi Qianwen multimodal AI model, enabling functions such as making calls, AI-powered search, object recognition, AI-assisted photography and problem-solving, multilingual translation, navigation, transcription, and health reminders.

For the near-eye display system, the Rokid Glasses utilize JBD’s Hummingbird Mini II monochrome light engine, which is only 0.15 cubic centimeters in size and weighs a mere 0.3 grams.

Despite its compact size, the Hummingbird Mini II delivers a brightness of 8 lumens at a typical power consumption of 60mW. The Rokid Glasses Priced at RMB 2,499 yuan, are set to hit the market in Q2 2025.

(Photo credit: JBD)

Eight New Micro LED AR Glasses Released in 2024

According to “TrendForce’s 2024 Near-Eye Display Market Trends and Technology Analysis Report”, Micro LED is regarded as a critical micro-display technology for AR devices due to its ability to meet high brightness, compact size, and low power consumption requirements.

TrendForce projects that Micro LED’s market share in AR devices will reach 18% in 2024 and grow to 44% by 2030.

Over recent years, demand for Micro LED technology in AR glasses has been steadily increasing. Based on LEDinside‘s incomplete statistics, 14 AR glasses incorporating Micro LED technology were released in 2023—a record high. As of now, 8 new models featuring Micro LED have been announced for 2024.

Although fewer models are being launched this year compared to last, the involvement of major industry players and the integration of AI technologies have made these products highly discussed. These developments could be pivotal for the evolution of Micro LED-based near-eye display technologies.

Meta and Apple Entering the AR Glasses Arena Could Boost Micro LED’s Popularity

In September, U.S. tech giant Meta unveiled a prototype of its full-color Micro LED AR glasses, Orion, weighing just 98 grams. According to TrendForce, Orion uses a diffraction waveguide made from silicon carbide and JBD’s three-chip full-color LEDoS technology, achieving a 70-degree field of view (FOV).

TrendForce anticipates Meta’s consumer-grade AR glasses to launch after 2027. With major brands driving development, application advancements are expected to accelerate. By 2026–2027, more refined and affordable mass-market devices are likely to emerge.

(Photo credit: Meta)

Alongside Meta’s Orion, rumors suggest that Apple also has AR glasses in development. Reports indicate that Apple is still committed to Micro LED technology, with AR glasses featuring Micro LED expected to enter mass production in 2026.

Micro LED as a Key to AR+AI Integration

Such as Rokid Glasses, Meta’s Orion, StarV Air2 by Xingji Meizu, and G1 by Eyewear Technology, are all equipped with advanced AI functionalities. These include translation, object recognition, voice-to-text transcription, summary generation, and intelligent prompts.

AI technology has become a key selling point for AR glasses. The application of AI in complex, all-day scenarios makes it highly practical and relevant.

With the addition of Micro LED’s high brightness and transparency features, AI-generated information can be displayed clearly and promptly to users. The AI+Micro LED combination significantly enhances the practicality of AR glasses, driving the industry towards smarter and higher-definition products.

TrendForce highlights AR devices as ideal platforms for AI due to their lightweight and compact design. The characteristics of Micro LED technology—high brightness, small size, and low power consumption—perfectly align with AR device requirements, making it a critical technological reserve.

If full-color Micro LED solutions can be achieved through vertical stacking or color conversion, this technology is likely to become the optimal solution for future AR glasses.

(Photo credit: Rokid)

According to Storm Media, citing Financial Times, several Chinese tech companies are reportedly accelerating efforts to establish AI research and development teams in Silicon Valley. They are offering lucrative incentives to attract top American talent, aiming to strengthen their R&D foundation ahead of potential stricter restrictions on Chinese firms under the Trump administration.

The Financial Times report points out that companies like Alibaba, ByteDance, and Meituan have been expanding in California over the past few months, actively headhunting experienced professionals from U.S. competitors. Their goal is to catch up in the competitive race for generative AI development.

The report highlights that the reason these Chinese companies rely on their U.S.-based offices to undertake these efforts is closely tied to existing U.S. regulations. The U.S. government prohibits the export of cutting-edge NVIDIA AI chips to Chinese firms, but these chips are essential for developing AI models, and the inability to access them poses a major obstacle for R&D.

To bypass the restrictions, Chinese tech firms or their subsidiaries have been acquiring high-end AI chips under the name of their U.S.-based data center. The report points out that this loophole has not yet been closed by U.S. regulations.

Alibaba’s Ambitions in Silicon Valley

According to the report from Financial Times, citing sources, Alibaba is currently hiring for its AI team in the San Francisco Bay Area. They have not only posted public job advertisements but have also actively approached engineers, product managers, and AI researchers with experience at companies like OpenAI and other major U.S. tech firms.

The report notes that this R&D team will focus on advancing Alibaba’s AI search engine, “Accio,” and that the company intends to turn the California AI team into an independent startup based in California.

Meituan’s AI Expansion in the U.S.

Meanwhile, Meituan, initially known as a food delivery platform, is also expanding its team in California, as the report notes. CEO Wang Xing has invited co-founder Wang Huiwen to return and lead the effort to build a dedicated AI R&D team, focusing primarily on menu translation functions and AI assistants. The new team is directly overseen by the company’s founders.

ByteDance Leads in Chinese Company’s AI Operations in the U.S. 

As for ByteDance, the parent company of TikTok, its U.S. AI operations are more advanced than those of Alibaba and Meituan. The report indicates that the company already has multiple teams working on various projects, including one focused on integrating AI features and another, a collaboration between employees from China and Singapore, developing a large language model named “Doubao.”

Baidu’s Scaled-Back U.S. Presence

The report also mentions that Baidu, China’s largest search engine operator,  previously operated an AI lab in Silicon Valley, employing top scientists and engineers to work on projects such as speech recognition and autonomous vehicles.

According to the report, at its peak in 2017, Baidu’s U.S. R&D center employed hundreds of people. However, due to internal conflicts within the company and worsening U.S.-China relations, Baidu has significantly scaled back its U.S. operations, as the report notes.

Read more

• [News] Latest Development on AI Models of China’s Top Techs: Alibaba, Baidu, ByteDance, Huawei and Tencent
• [News] ByteDance Reportedly Turns to TSMC on in-house AI Chips to Cut Purchase Cost on NVIDIA

(Photo credit: Alibaba)

According to TrendForce’s latest memory spot price trend report, regarding DRAM, price decline for both DDR4 and DDR5 products has begun to moderate, as sellers realize that further price reductions are unlikely to spur procurement activities. As for NAND flash, the price reduction of eMMC has gradually subsided. Details are as follows:

DRAM Spot Price:

As in the previous weeks, the transaction momentum in the spot market has yet to show any improvement this week. The average spot price of mainstream chips (i.e., DDR4 1Gx8 2666MT/s) has dropped by 0.33% from US$1.840 last week to US$1.834 this week. TrendForce has made the following observations:

(1) Module houses are currently prioritizing inventory reduction, resulting in sufficient supply in the spot market.

(2) The price decline for both DDR4 and DDR5 products has begun to moderate. However, this is not due to an improvement in the demand situation, but because sellers realize that further price reductions are unlikely to spur procurement activities.

(3) DDR4 products have a weaker procurement momentum compared with DDR5 products, indicating that demand has largely shifted to the DDR5 generation.

(4) Looking ahead to 1Q25, the strategy of lowering inventory levels is expected to continue. The overall trend of spot prices will be similar to that in 4Q24, remaining weak.

NAND Flash Spot Price:

The spot market also lingers amidst enervation this week. Observations on market performance are as follows:

(1) Buyers are not rushing in procurement as module houses, channels, and suppliers are all equipped with sufficient stocks.

(2) Sellers understand that an ongoing price drop would not be able to generate demand having seen the recently subsided price reduction of eMMC, after a significant depletion previously.

(3) Sluggish market demand has led to constant piling up of inventory among module houses, which further inhibits their purchase momentum on wafers, and results in continuous accumulation of inventory among suppliers. The end result is the surging inventory for both the upstream and downstream sectors of the supply chain.

Spot prices of 512Gb TLC wafers have dropped by 0.63% this week, arriving at US$2.380.

The flames of war continues to spread, as the HBM battle seems to extend to the next-gen products. According to the Korean Economic Daily, Samsung and SK hynix are reportedly developing HBM4 samples for Tesla, as the EV giant joins other U.S. tech giants in the race to develop its own AI chips.

The report indicates that Tesla has requested the South Korean memory duo to supply HBM4 chips for general use, while it is expected to select one of the two companies as its HBM4 supplier after testing their samples.

The HBM4 products are expected to be incorporated into Tesla’s Dojo, its custom-built supercomputer driven by the company’s D1 AI chip, which is designed to train the “Full Self-Driving” neural networks, notes the report.

In addition, Tesla might utilize HBM4 chips in developing AI data centers and its self-driving cars, which currently use HBM2e chips for their pilot programs, according to the report.

The two Korean memory heavyweights have been working on customized HBM4 chips for major U.S. tech companies, as they aim to reduce their reliance on NVIDIA’s AI accelerators. A previous report by South Korean media outlet Maeli Business Newspaper discloses that Samsung has already begun developing “Custom HBM4,” a next-gen high-bandwidth memory tailored specifically for Microsoft and Meta.

On the other hand, the current HBM leader SK hynix, which used to be the exclusive HBM3e supplier for NVIDIA, was reportedly requested by NVIDIA CEO Jensen Huang to accelerate the supply of HBM4 by six months. In October, the company said that it planned to deliver the chips to customers in the second half of 2025.

It is worth noting that both memory giants have teamed up with TSMC on HBM4, as they attempt to incorporate customized features requested by major clients, counting on TSMC to manufacture more powerful logic dies, the component that functions as the brain of an HBM chip.

According to the information provided by SK hynix cited by the Korean Economic Daily, HBM4 offers 1.4 times the bandwidth of the fifth-generation HBM3e while consuming approximately 30% less power.

Read more

• [News] Samsung Reportedly Starts Developing Custom HBM4 for Big CSPs, Eyeing Mass Production by 2025
• [News] SK hynix Reportedly to Tape-out HBM4 in October, Paving the Way for NVIDIA’s Rubin

(Photo credit: Samsung)

According to a Wccftech report, citing Bloomberg, Huawei’s efforts to develop more advanced chips for AI have faced significant setbacks due to U.S. sanctions, making it unable to match the performance of even older NVIDIA products.

The report notes that the sanctions have blocked Taiwan’s TSMC from manufacturing the most advanced chips for China and prevented Chinese chipmakers from acquiring the latest chip manufacturing equipment from ASML, making it hard for China to produce advanced chips. As a result, Huawei’s efforts to develop AI chips are restricted to the 7nm process and are expected to remain so for the next two years, as the report indicates.

Huawei’s AI Chip Push Hindered by U.S. Sanctions on SMIC

The report highlights that NVIDIA’s latest AI chips, the H100 lineup, are manufactured using TSMC’s 4nm process, which is significantly more advanced than TSMC’s 7nm technology.

However, due to sanctions imposed by the Biden administration, Huawei’s main production partner, China’s SMIC, is prohibited from obtaining ASML’s latest extreme ultraviolet (EUV) lithography machines, preventing it from advancing beyond the 7nm process to more advanced technologies, as noted by the report.

ASML is the sole provider of these advanced EUV machines worldwide. SMIC is limited to using older scanners that rely on deep ultraviolet (DUV) lithography. As a result, the report notes that, citing sources, SMIC has been forced to rely on multi-patterning techniques in an effort to manufacture 7nm chips.

According to the report, in chipmaking, multi-patterning involves dividing a mask (which contains the circuit design) into sections, allowing each part to be “printed” onto silicon separately to achieve the resolution required for smaller feature sizes. However, the report notes that this process increases production time and poses quality challenges.

7nm Yield Issues Remain a Challenge

According to the Bloomberg report, SMIC is even struggling to produce 7nm chips at consistent volumes. Citing industry sources, the report highlights that SMIC’s 7nm production lines have been hampered by low yields and reliability issues.

The report highlights that SMIC is not only forced to rely on multi-patterning to circumvent the restrictions but also faces challenges from government pressure to use local equipment. This has left SMIC lagging behind older Western chip manufacturing technologies.

With TSMC gearing up to produce 2nm chips next year, and SMIC still struggling with multi-patterning and domestic manufacturing equipment, Huawei’s attempts to develop its own AI processors face considerable challenges, as the report indicates.

These challenges not only undermine Huawei’s competitiveness in the consumer electronics market, where it competes with rivals like Apple, but also hinder China’s efforts to achieve semiconductor self-sufficiency in the AI era and stay on par with the U.S., according to the report.

Read more

• [News] Huawei May Reserve Kirin 9100 Only for Premium Mate 70 Models Amid SMIC Yield Issues
• [News] Huawei’s Proxy Tactics for Advanced Chips Expose New U.S. Sanctions Loophole

(Photo credit: Huawei)