If 2023 is the first year of large-scale model technology, 2024 is regarded by the industry as the first year of the outbreak of "AI + hardware". Under the rising wind, a number of Internet giants and 3C manufacturers rushed into the market, and the era of intelligence is in full swing.
Whether it is the ByteDance Doubao team combining the software and hardware of large models with devices such as landing robot dogs, or Meituan grafting related AI businesses with children's wear; Whether it is the advent of iFLYTEK's first intelligent review machine, or Baidu's rumored to provide AI functions for iPhone 16 and other products, superimposed on AI PCs and AI mobile phones that have been hyped in the capital market in turns, the display form of "AI + hardware" is becoming more and more diversified, and it has a certain "stimulant effect" in the consumer electronics market that has been sluggish for several years.
AI is redefining smart hardware and is opening a new cycle of hardware innovation. On the one hand, this is expected to reshuffle the industrial pattern, and on the other hand, it will also objectively promote the ecological prosperity of the AI industry.
However, it is still unknown whether the new entrants represented by the Internet platform can build a new increment with the help of "AI + hardware". Some people in the industry believe that "AI+ hardware" will be a system that integrates multiple factors such as scenarios, technologies, and experiences, and it is not easy to really dig gold in undiscovered high-frequency usage scenarios.
AI reshapes smart hardware
At the end of June, iFLYTEK released its first Spark intelligent review machine. Based on the base capability of the Spark model V4.0, it intelligently corrects multi-disciplinary and multi-question homework, and generates a multi-dimensional learning report in real time, so that teachers can conduct hierarchical guidance when evaluating homework. Previously, iFLYTEK also launched AI headsets and other products, and increased AI office.
This is another AI hardware applied to learning scenarios after Baidu Z30. Previously,Baidu's Xiaodu launched the first AI learning mobile phone"Qinghe",In May this year, a new generation of learning machine products were launchedZ30,Become the world's first learning machine based on the ability of Wenxin's large model。
Baidu and iFLYTEK released large models last year and have continued to iterate. Both companies have a mix of hardware and software. However, some Internet giants that are only good at software have also quickly embarked on the journey of "AI + hardware" this year.
A person close to ByteDance told the reporter of Securities Times E Company that the Doubao team is exploring the combination of software and hardware based on large models, and has been applied on hardware devices such as learning machines, robot dogs, and robots. At the same time, Meituan, which has been fighting offensive and defensive battles with Byte in the local life track, is also exploring the possibility of combining AI and hardware through the "Qiaoyu" AI business, which has reached a cooperation with the children's wearable device manufacturer "Little Genius".
The market regards 2023 as the first year of the explosion of large-scale model software and applications, and by 2024, the transformation driven by artificial intelligence large-scale models is expected to undergo a new transformation and enter a new stage of large-scale outbreak of "AI + hardware". Some institutions estimate that the scale of China's smart hardware market will exceed 1.5 trillion yuan in 2024.
According to the analysis of relevant people from the China Academy of Information and Communications Technology, the large-scale model arms race since 2023 has made the model performance continue to be iteratively upgraded, and the algorithms, computing power, data and other grains and grasses are sufficient, which have become the prerequisites for the application and terminal based on the large model to enter the explosive period. "AI applications and terminals complement each other and innovate each other, which will promote the development of the entire AI industry ecosystem."
In this wave of change, AI PCs and AI phones are bearing the brunt. The two have caused a sharp rise in the A-share market-related concept stocks, verifying the optimistic expectations of the secondary market for "AI + hardware". From the enterprise side, major mobile phone and computer manufacturers have also waved the banner of AI to charge towards the increase in sales, making this "AI + hardware" year very lively.
Taking Lenovo as an example, in the fourth quarter of the 2023/2024 fiscal year, the revenue of 99.4 billion yuan and the net profit of 1.78 billion yuan have finally made the company clear about the upward trend of performance, especially AI PCs have begun to leverage new increments. According to the concept, Lenovo will rely on the new IT technology foundation to build AI-embedded intelligent terminals, AI-oriented infrastructure and AI-native solution services with more secure and cost-effective hybrid AI, so as to meet the explosive growth opportunities in the AI era.
The competition for AI hardware by domestic technology giants and Internet platforms is consistent with the integration strategy of overseas giants such as Apple, Microsoft, Qualcomm, and Samsung, which emphasize AI large model technology and hardware. From the device side, from mobile phones to PCs, AR glasses, home appliances, cars, robots, etc., AI will be ubiquitous and omnipotent.
According to the above-mentioned CAICT, this process will be accompanied by an evolution from 2D to 3D. "PCs and mobile phones represent the era of two-dimensional computing, and new three-dimensional innovative content and applications will appear in the future, such as intelligent robots, intelligent electric vehicles, and various headsets, all of which are expected to become super terminals in the three-dimensional era."
Entry Path Diameter Field
Judging from the path selection of "AI + hardware" for many players to enter the game, it can be roughly divided into two modes: "AI + hardware" and "AI native hardware". The former is dominated by large technology manufacturers and consumer electronics giants, and has the consideration of differentiated layout in the homogeneous competition of consumer electronic products; The latter is more "revolutionary", hoping to achieve new interaction methods and user experience with the help of AI, but it is difficult to say that there is a real breakthrough at present.
In the mainstream path of "AI + hardware", the introduction of AI technologies such as large models into mobile phones is a typical representative. Huawei Mate 60 is equipped with a self-developed Kirin chip and is connected to Huawei's Pangu model. The Xiaomi 14 series is powered by the Qualcomm Snapdragon 8 Gen3 chip and the new surging OS. Samsung, OPPO and others are also exploring large-scale mobile phones.
In the process, the interaction between AI and hardware has been enhanced like never before. For example, Musk has previously stated that he is not interested in simply building mobile phones, but with the advent of the AIGC era and the advancement of technologies such as brain-computer interfaces, how to connect with smartphones will also become a new topic. Therefore, there is a voice that believes that the possibility of Musk linking his X technology business ecology through his smartphone is increasing.
If we distinguish from the depth of hardware manufacturing, we can also see the distinction between different modes. Baidu, iFLYTEK, etc. are more like the "in-person" model, which needs to give full play to the advantages of supply chain and resource integration. Meituan and others adopt a self-developed + cooperation model, which can be regarded as the landing and realization of large models. A considerable proportion of manufacturers choose to "walk on two legs", while introducing one or more large models from the outside, while burying their heads in their own research behind their backs. For example, most mobile phone manufacturers have chosen to access as many large models as possible to improve the ability to generate multimodal AI, and on the other hand, they can use the cloud large models of service providers to train self-developed models to accelerate the improvement of the intelligence level of the models.
Under the parallel development of multiple models, some platforms have made up for their own shortcomings through capital operation. Taking ByteDance as an example, in the first half of this year, the company's acquisition of the well-known open-ear headphone brand Oladance entered the final stage, before the market voice believed that the company was optimistic about the potential of the headphone market segment, and now it seems that the market believes that Byte is supplementing the "AI+ hardware" strategy of the entrance products, behind which is the market potential of wearable devices as an AIGC service platform. In recent years, the byte business system has been changed several times, and it is worth paying attention to whether this move means that the strategy of "focusing on the main business" is undergoing some subtle changes.
"Can it be done" doubts
Some sub-categories of "AI + hardware" have the meaning of saving the declining trend. Typical are smart speakers. As a transitional product of IoT, related revenue has gradually declined in recent years in the financial reports of Baidu and Alibaba. Therefore,Whether it is Baidu Tiantian's new home robot,Or Ali Tmall Genie is equipped with AI's new X6Speaker,It is considered by the market to have the meaning of reversing the previous downward trend。
There are also some sub-categories that are leveraging new market expectations. Taking humanoid robots as an example, AI technologies such as large models can be combined with humanoid robots from various aspects such as voice, vision, decision-making, and control to form a closed loop of perception, decision-making, and control, greatly improving the intelligence of robots, expanding the applicability of robot scenarios, and making the "iPhone moment" of humanoid robots closer and closer. At the same time, the market is also full of expectations for the "Model 3 moment" of humanoid robots, which points to the fact that, like Tesla, it has scientific and technological innovation on the one hand, and also has engineering capabilities, large-scale mass production capabilities, and the ability to expand the market through people-friendly parity.
But this is not an easy task. As far as AI hardware products are concerned, ensuring battery life, computing power, and user experience are all challenges. From the manufacturing side, from design, prototyping, testing to mass production, it requires a long period of research and development and a large amount of capital investment.
Especially for Internet platforms, entering the hardware is often not dominant. WeChat headphones, QQ music audio, Alibaba Cloud mobile phones, 360 smart homes, etc., have gradually disappeared in the market competition. ByteDance's "Xinshi Lab" was also exposed to the development of educational hardware such as office scenes and smart desk lamps, and the response was mediocre in the end. In particular, PICO, which has high hopes from the market, has not been able to achieve great development with the help of the metaverse, and the dream of developing hardware has been called "repeated defeats" by the market.
The above-mentioned person from the Academy of Information and Communications Technology admitted that there is no AI hardware product on the market that surprises most consumers, and the application of AI in hardware products is still in the initial stage of functional stacking. "AI itself is a technology, not a product, and it needs to be combined with specific products to create value for users. This process of integration is not only a simple integration of hardware, but also how to combine AI technology with product experience, which is a system of multiple supports such as scenarios, technologies, and experiences. ”
However, the discovery or invention of undiscovered high-frequency use scenario functions is by no means an easy task. Previously, Kuaishou cooperated with Zebra and Byte made mobile Internet products land in automotive terminals, all of which suffered phased defeats. How to solve the problem of "porting non-rigid scenes" is still an important test for all players.