Since entering 2018, the domestic smartphone market has continued to slump. Under pressure, many mobile phone manufacturers have to focus on innovation and try to find new ways out. As large AI models continue to emerge from the industry in 2023, AI and pan-AI hardware such as AI PCs, AI mobile phones and other related directions have increasingly attracted the attention of the industry. The "accidental intrusion" of AI is like a catfish, quickly stirring up This has affected the fighting spirit of mobile phone manufacturers.
At the beginning of 2024, mobile phone manufacturers such as OPPO and Meizu have responded to the call for AI mobile phones and announced the promotion of the development of AI mobile phones. Huawei, Xiaomi, Honor and other manufacturers have also secretly laid out business related to AI mobile phones, which shows that the industry attaches great importance to AI. . Among them, OPPO founder Chen Mingyong announced that the industry has entered the first year of AI mobile phones. Behind the excitement in the industry, there is actually hidden concern and emphasis on the replacement of mobile phones by major manufacturers.
Smartphones that accelerate the iteration towards AI
With the development of large AI models, generative on-device artificial intelligence has received increasing attention, and smartphones have become one of the best carriers of on-device artificial intelligence. The so-called AI mobile phone is the latest mobile phone with end-side generative artificial intelligence. As the forefront of developing new productivity, it has increasingly become a new direction for mobile phone iteration to accelerate under the new situation. Starting from the second half of 2023, manufacturers represented by Google, Samsung, and OPPO have begun to build large models into mobile phones, completing the transformation from cloud AI to terminal AI, and the AI mobile phone wave has officially begun.
First of all, on-device AI can save money. As early as September last year, Qualcomm issued a report in which it succinctly pointed out that hybrid AI is the future of AI. The core of this report is that Qualcomm uses the "downtime" phenomenon of large-model cloud reasoning to once again emphasize to the market that hybrid AI where the terminal and the cloud work together is the real first choice. The so-called hybrid AI means that in a cloud-centered scenario, the terminal will share the AI workload from the cloud when possible according to its own capabilities; in other scenarios, the computing will mainly be terminal-centered, and will be transferred to the cloud when necessary. Offload tasks.
Qualcomm believes that eventually smartphones will adopt a model that combines the cloud with edge terminals such as PCs and smartphones like traditional PCs, while generative AI will form a combination of cloud and terminals, resulting in two trends: AI computing Power will shift from being concentrated on the cloud in the past to a model of flexible distribution between the cloud and terminals; large models will also accelerate their penetration into terminals. In fact, when large AI models enter the terminal, they not only run fast, but can also save mobile phone companies money.
According to data released by vivo's vice president, the cost of a single conversation with vivo's large model is about 0.012-0.015 yuan/time, and the current number of vivo users nationwide is about 300 million. Calculated based on 250 million daily active users (assumed future penetration rate ceiling) and 10 conversation requirements per person per day, the daily cost of cloud inference for vivo's large model will be 30 million to 37.5 million yuan. Assuming a 50% offload rate and half of the inference is offloaded to the device side, vivo will save RMB 5.5-6.8 billion in cloud computing costs every year. For the sake of money, mobile phone manufacturers can’t do anything without developing on-device AI.
Secondly, the changes brought about by AI mobile phones have an impact on the whole body, and may eventually affect the functions and structure of smartphones, and even the gross profit margins of mobile phone manufacturers. Under the influence of the AI revolution, the architecture of mobile phone SOCs has undergone profound changes. In addition to the traditional CPU+GPU, the neural network module (NPU) responsible for AI calculations has increasingly become an important computing unit in smartphones. increasingly important.
With the addition of this new module, the production cost of the smartphone itself has also increased, and the related peripheral supporting chips have further promoted the rise in the cost of smartphones, thus forming the "AI mobile phone needs to increase computing power - triggering SoC and memory "Comprehensive upgrade - memory and chip demand surges, prices rise - mobile phone manufacturers are under profitability pressure" closed loop. Under the pressure of this closed loop, smartphone manufacturers must either raise prices to obtain product premiums to increase overall product profit margins; or they may temporarily bear costs and achieve small profits but quick turnover by expanding scale. In short, under the new situation, AI is rapidly penetrating terminals.
OPPO, unwilling to lag behind
While many manufacturers are stepping down to participate, OPPO, which has always emphasized being the last to strike, is not far behind and is determined to participate in this feast.
First of all, OPPO is accelerating the integration of internal AI research and development resources, using the whole company's efforts to integrate the company's power in the cloud and terminals, paving the way for the implementation of large AI models in terminal mobile phones. As early as this year's annual meeting, OPPO founder Chen Mingyong emphasized the need to integrate OPPO's global R&D resources for AI and "use the power of the entire company" to seize the opportunities of the times. To this end, OPPO has made major adjustments to its personnel structure, integrating all AI-related functional departments into one entity department, and setting no upper limit on personnel and capital investment.
In addition to personnel research and development, OPPO has also technically deployed three-level large models, and implemented joint deployment of the Titan ultra-large model, Turbo large model, and Tiny lightweight model on devices and clouds. With the support of these AI resources and large models, OPPO's newly released Find X7 series mobile phones have implemented many new AI functions such as AI elimination function, AI call summary, AI ultra-clear photo taking, etc., and are working hard to become system-level agents.
Secondly, OPPO works closely with chip manufacturers such as MediaTek to carry out in-depth customization around AI chips to provide powerful computing power for the terminal-side deployment of large AI models. According to industry insiders, in the short term, the development of AI mobile phones still faces constraints in terms of algorithm efficiency and accuracy, improvements in chip computing power, and hardware energy consumption. Therefore, in the long run, if you want to break through the development bottleneck of AI mobile phones, you must start from these aspects, and the solution to these problems depends on the development of chip technology.
In order to promote the development of AI mobile phones, OPPO, in addition to self-developed chips, has also jointly developed a new generation of Dimensity 9400 chip with partners such as MediaTek. It uses a new Tidal engine to truly achieve the transition from system cache to CPU exclusive cache. Dynamically allocating resources enables an unprecedented leap in chip energy efficiency.
Is the AI phone a pioneer or a gimmick?
Regarding the wave of AI mobile phone craze in the mobile phone circle, industry insiders have mixed reviews. Some people think that AI mobile phones are just gimmicks used by many manufacturers to attract users, and there is nothing new; others believe that domestic The great development of various aspects of technology has prepared the conditions for the birth of AI mobile phones, and now is the time to take the lead. So, for OPPO, is the introduction of AI mobile phones a first move or a gimmick? It also needs to be considered based on various aspects of the situation.
First of all, from the perspective of OPPO’s own situation, due to its low development in high-end and IOT, it is more dependent on existing business, and OPPO has the motive to urgently want to be surrounded by conflicts. Since Huawei was banned in 2019, OPPO has also been eager to try out in the high-end market. However, it does not have much of a presence today. In order to impact the high-end market, OPPO announced the launch of self-developed chips in 2019. In 2020, it launched OPPO Find, which was previously suspended, and has also accumulated patents in many folding screen technology fields such as water drop hinges.
However, OPPO's road to self-research of chips has been full of twists and turns, and it announced its closure as early as last year. What makes OPPO even more worried is that Huawei has announced its "resurrection" since last year. It has not only returned to the candy bar market, but also occupied a large share of the folding screen market, posing a "threat" to OPPO again; And compared with Xiaomi, OPPO, which has always focused on offline stores, has closed many stores due to the severe impact of the previous epidemic.
Under heavy siege, OPPO executives announced major business and organizational adjustments and called back Zuohu Liu, who had been fighting overseas, in an attempt to find a breakthrough. From these aspects, OPPO does have sufficient motivation to participate in AI mobile phones.
Secondly, from a capability perspective, OPPO is still unable to make AI phones. At present, AI mobile phones are mainly reflected in two aspects: hardware and software. At the hardware level, mobile phone companies have little room to play, especially those that rely on purchasing upstream chips. What they can freely play is at the software and application level, which naturally depends on the application ecosystem built by previous manufacturers. Is it strong enough? As mentioned earlier, OPPO’s chip business has been forced to shut down due to various pressures, and the terminal IOT ecosystem has not improved significantly. Under such circumstances, it has naturally become more difficult for OPPO to make an AI mobile phone.
Taken together, OPPO does have plans to take the lead, but in terms of current capabilities, it may still be beyond its capabilities. However, it is probably an exaggeration to say that it is a gimmick.