On the evening of January 6, US time, a serpentine queue stretched for miles outside the NVIDIA keynote venue. Countless attendees eagerly awaited their turn to enter, marking the countdown to the next day's Consumer Electronics Show (CES) and the opportunity to glimpse into the future.

CES reopened in Las Vegas, USA, on January 7, attracting over 4,500 exhibitors, with Chinese exhibitors comprising nearly 30% of the total.

Similar to last year, AI was omnipresent at the exhibition. Despite the CES 2024 theme being "Dive In," Optical Cone Intelligence observed that AI was prominently featured as the most eye-catching slogan on booths showcasing smart glasses, robots, smart homes, toys, AI PCs, and various other innovative gadgets.

After a day of navigating the exhibition, Optical Cone Intelligence noticed a somewhat "polarized" approach among Chinese manufacturers. Mobile phone manufacturers were notably absent, with Great Wall Motors and Zeekr primarily showcasing automotive products. However, in the realms of smart glasses (including XR devices), robots, and smart homes, Chinese manufacturers were ubiquitous, drawing significant attention away from neighboring booths.

Overall, this year's CES was notably more subdued.

In contrast to last year's ambitious yet ultimately unsuccessful AI hardware like the AI Pin and Rabbit R1, most AI hardware this year is based on solid product logic (or iterative products). AI functionality is no longer the sole selling point; rather, the rapid integration of mature AI and hardware for implementation has emerged as the most common product strategy at CES 2024.

With fewer whimsical ideas, AI is accelerating the upgrade of all existing hardware products.

From rollable screens to AI bird baths and AI plush toys, CES has never lacked bizarre gadgets.

As the world's largest annual AI hardware carnival, CES continues to exhibit boundless imagination. Each year, the exhibition features a myriad of quirky gadgets.

Lenovo, which defined AI PCs for 2024, introduced a series of PCs with "varied screens," including those with screens on the keyboard and those with screens on both the top and bottom. Among them, the most novel is the rollable screen, where the laptop screen can be extended upwards, making it a coder's dream. Additionally, Lenovo unveiled a travel wear set comprising a bracelet and a necklace, allowing users to ask questions and take photos through simple voice interactions, effectively serving as a practical iteration of last year's AI Pin.

In the common AI translation scenario, Timekettle gathered a crowd on its booth to "learn foreign languages" using its AI recorders and AI headphones. Optical Cone Intelligence tested the product and found that its AI simultaneous interpretation function has a latency of around 3-5 seconds. While still slightly "behind the beat," the translation effect, bolstered by large models, is already suitable for daily spoken language.

The most inexplicable product belongs to Birdfy, which launched a smart bird bath. According to the introduction, this product is used for close observation of birds. Users place water and food in the basin, and the AI detects the bird, identifies the species, and automatically records a video.

Not all manufacturers are using AI to enhance hardware intelligence; some are making AI "dumb." Mengyou Intelligent's AI toy Ropet, for instance, will "stare" at you when touched and dance to music.

Yukai Engineering, a Japanese manufacturer, created an AI robotic sloth that, when hung on a bag or arm, looks around at others under AI control.

Jizai's five-legged robot Mi-Mo from Japan, with a desk lamp as its head, also drew attention, albeit with a somewhat eerie appearance. It probably provides emotional value by serving as a "spotlight" wherever the user goes.

The latest development in AI virtual humans is the real-time interactive AI virtual human large screen launched by PERSO.ai in collaboration with Microsoft. Through the debugging of large models, the interaction latency is controlled at 1.5 seconds, indicating that future "sales assistants" in malls will soon feel the pressure.

Moreover, CES 2024 featured novelty items such as AI stringless guitars, pens for real-time transcription of notes, and screens that automatically protrude and recess based on screen content.

However, after several rounds of navigating the exhibition halls, Optical Cone Intelligence's deepest impression remained of the Chinese AI hardware manufacturers in the spotlight.

The battle for AI glasses has already commenced, with Chinese AI hardware dominating CES.

As a race that attracted numerous players in 2024, over 80 Chinese AI glasses (including XR) manufacturers participated in CES. Lei Niao CEO Li Hongwei expressed concern at CES, stating, "The glasses race has never been this lively." The battle for AI glasses has already "erupted" at CES.

In terms of product form, most AI glasses this year chose to align with Meta's offerings. Lightweight (less than 50g), collaboration with glasses manufacturers, edge AI integration, built-in display functionality, and prices comparable to ordinary glasses have become the standard configuration for AI glasses.

Taking Lei Niao glasses as an example, in 2024, Lei Niao glasses were still emphasizing TCL's display technology. This year's Lei Niao V3 not only lowered the price to 1,799 yuan but also "embraced" Alibaba and Qualcomm, implementing AI SoC and a large edge AI model (customized by Tongyi Qianwen). Overall, the product completeness of Lei Niao V3 has surpassed that of Meta's Ray-Ban Stories, excelling in photography, display, AI interaction, pricing, and other aspects.

However, merely surpassing Meta's Ray-Ban Stories is not enough for Chinese AI glasses to take the lead. At CES 2024, Optical Cone Intelligence also observed more manufacturers contemplating the application of AI glasses.

Rokid brought its "complete set" of Rokid Glasses, Rokid AR Lite, and Rokid AR Studio to the CES venue. With different types of product combinations, many attendees were seen wearing glasses and raising their hands, thoroughly enjoying the experience. To test the AI translation function, many visitors even spontaneously "formed teams" to communicate and test in different languages.

Amazon's best-selling Chinese AI glasses giant XREAL introduced an AR+ intelligent driving solution. XREAL aims to utilize the AR display capabilities of AI glasses to supplement missing information from dashboards and HUDs while driving, enhancing driving safety by eliminating the need for drivers to turn their heads to look at the car's infotainment system.

Furthermore, CES 2024 featured new faces with cross-border innovations, including ThundeRobot Technology (a subsidiary of Haier), INMO Technology, Jianan Technology, Liweike Technology, and Shanji Technology.

Despite numerous participants, the AI glasses industry may not have reached its "bloodbath moment" yet. According to JD.com, some Lei Niao V3 models sold out within 10 hours of their release. The high enthusiasm of consumers indicates that startups and upcoming internet giants still have some time to refine their products.

Apart from AI glasses, robots are the second largest area where Chinese manufacturers gathered to "flex their muscles." Following the technological iteration of training robots with large AI models in 2024 and the entry of many players from the automotive industry, Chinese manufacturers staged a "chaotic battle" of robots at CES.

At the "veteran" manufacturers' booths, Optical Cone Intelligence witnessed robot dogs flipping all over the place upon arrival at the booths of Unitree and CloudMinds. At the booths of Lingbao, Jia Su Jin Hua, and Yinhe Tongyong, the displayed products were largely similar to those at the 2024 World Robot Conference.

More eye-catching were the players from the automotive industry who have ventured into the robot field.

Among them, China's two most renowned LiDAR manufacturers, RoboSense and Hesai Technology, both presented the latest advancements in robots. The difference between the two is that Hesai developed LiDAR for robots, while RoboSense directly created a robot. It is understood that RoboSense already possesses an AI development platform and masters three types of components: vision, touch, and joints, along with dexterous hands. The level of product completion and R&D speed are remarkable.

On the other hand, automotive intelligent driving chip manufacturer Black Sesame Technologies collaborated with Fourier Intelligence to introduce a dexterous hand this time. With the assistance of the C1200 chip, the robot's fingers achieve almost human-level flexibility. Perhaps, beyond the field of automotive intelligent driving, Black Sesame has already identified a second growth curve in robots and collaborations.

Focusing on the automotive industry itself, only Great Wall Motors, Zeekr, and XPeng participated as Chinese smart car manufacturers. Among them, Great Wall Motors and Zeekr showcased their most prized intelligent driving and intelligent cabin evolutions from 2024. XPeng primarily exhibited its flying vehicle and land aircraft from XPeng Hui Tian.

In addition to humanoid robots, Chinese manufacturers are actively applying robot technology to other products.

In the highly competitive vacuum cleaner industry, Dreame and Roborock have both installed external arms on their vacuum cleaners. In terms of morphological differences, both utilize a mechanical arm with two "pincers" to address the pain point of vacuum cleaners having to bypass obstacles. Among them, Dreame's mechanical arm can lift objects weighing up to 400g, enabling the vacuum cleaner to complete tasks such as arranging shoes, socks, tissues, and rags.

Lastly, there are the Chinese home appliance manufacturers that "impress with their size." Hisense and TCL both introduced new TV products with "overwhelming" giant screens at the exhibition. Focusing on the implementation of AI applications, both manufacturers have integrated AI dimming, AI assistants, and AI-connected smart home systems into their products. Notably, TCL had the largest booth at CES this year, showcasing a trend towards whole-house intelligence. It seems that TCL's home appliance strategy is indeed converging with that of Haier Smart Home.

Overall, actively showcasing hardware combined with AI is the secret to Chinese manufacturers becoming the focus of CES. By integrating AI capabilities, Chinese manufacturers are launching an "abundant" AI hardware product line, shifting the main battlefield of the AI industry towards manufacturing logic.

It is foreseeable that under the competitive model of Chinese manufacturers, 2025 may be a year of full bloom for AI hardware. As AI "rapidly advances" across various industries, the innovation logic of AI hardware will also evolve accordingly.

AI has entered the stage of combined innovation, with large AI models accelerating their implementation through hardware.

Based on the maturity of AI tools, AI development platforms, AI chips, and other supporting products, the pace at which AI hardware manufacturers create products is becoming faster and faster.

Among the AI computing power vendors at the bottom layer, NVIDIA has taken the lead in stepping on the gas.

The 50-series graphics cards released this year did not receive a significant boost from NVIDIA. Excluding NVIDIA's somewhat inaccurate comparison methods (such as testing AI with different precisions on both sides), it can be seen that the 50-series graphics cards achieved a roughly 30% performance increase through material stacking. The substantial difference in frame rate performance during the demonstration primarily stemmed from the optimization of Transformer for DLSS (Deep Learning Super Sampling) technology. NVIDIA has unleashed the predictive capabilities of generative AI, transforming the computation of images from "inserting one frame between two frames" to "inserting three frames between two frames."

After generative AI unleashed the world's predictive capabilities, NVIDIA also introduced model "pre-made dishes" to the entire AI application community. Among them, the Cosmos World Foundation Model launched by NVIDIA this time can provide simulations based on the real physical world for AI models, which can be understood as an AI pre-training framework for intelligent driving and robots.

Similarly, Qualcomm also focused on showcasing its AI SoC platform at CES 2024. Not limited to cars and PCs, relying on edge AI implementation capabilities and SoC power consumption control, the latest generation of AI glasses predominantly uses Qualcomm's AR1 chip.

"Compared to the Spreadtrum chips commonly used in the industry before, the Qualcomm AR1 chip is significantly superior in terms of process, connectivity, power consumption, and other aspects," said Li Hongwei.

At the software level, China's large AI foundation models have also achieved astonishing implementation speeds after two years of development.

At CES, we witnessed BMW's subsidiary Smart collaborating with ByteDance's "Doubao Large Model" to implement an intelligent cockpit. Last year's Lei Niao glasses, lacking AI capabilities, integrated AI functions this year through cooperation with Alibaba's Tongyi Qianwen. Outside of CES, there were also collaborations between Rokid and Lenovo's AI PCs with Tongyi Qianwen, as well as Tencent's Hunyuan Large Model launching an intelligent cabin solution for automobiles.

The implemented AI models also differ greatly from the previous method of accessing cloud-based large model APIs. AI hardware manufacturers only need to train a small model with recognition intent to differentiate themselves. The overall implementation speed is so rapid that a member of the ByteDance AI team told Optical Cone Intelligence, "If the other team is complete, and there is no need to retrain the model, the simple deployment of the button intelligence agent can be completed in one day."

With the assistance of chips and AI, Chinese AI hardware manufacturers, who have long "mastered" supply chain advantages, have been able to quickly roll out their products.

Let's take the battery life of AI hardware as an example. Once the development of silicon-carbon anodes for automotive power batteries reached maturity, the mobile phone industry swiftly adopted large batteries with capacities exceeding 6,000mAh within a year. By 2024, however, the vacuum cleaner industry was still caught up in the competition for movable cleaning wheel assemblies, prematurely branding them as "robotic arms." This year's CES embraced the embodied intelligence industry, showcasing dexterous hands as the next frontier.

Consequently, AI hardware is poised to enter a phase of combined innovation.

Currently, AI glasses and AI toys are established products that no longer need to prove their feasibility. AI enhances productivity, and AI PC manufacturers can harness the ecological strength of the Agent platform. For smart cars, intelligent driving, and intelligent cabin manufacturers, the focus has shifted from competitiveness to user experience.

At the 2025 CES, we will witness the dawn of the era of inclusive AI.