Smartphones supporting on-device AI are expected to reach 47% by 2028, and Hybrid-AI will accelerate the popularization of AI functions on smartphones

Since the end of last year, AI functions have officially started to support running on the end side of smartphones. Mainstream smartphone manufacturers have also launched AI large models with different parameters, which support deployment on the end side and the cloud side. Chip suppliers MediaTek, Qualcomm, Google and Samsung have successively launched the latest high-end chips, which support AI large models with 7 to 10 billion parameters. Smartphone brands have launched smartphones with on-device AI capabilities in conjunction with corresponding chips, such as vivo's X100 series, OPPO's Find X7 series, Honor's Magic 6 Pro, Samsung Galaxy 24 series and Google's Pix 8 series. As time goes by, the number of smartphones with on-device AI capabilities is also increasing, but currently they are all flagship product series of each company, equipped with the latest flagship chips and memory capacity starting from 8GB. Chip computing power and memory size are crucial to the on-device AI experience. The penetration rate of smartphones supporting on-device AI capabilities is also limited by chip computing power support and memory capacity.

Smartphones supporting on-device AI are expected to grow rapidly every year starting in 2024. In particular, on-device AI will expand to high-end and mid-range smartphones, and the penetration rate of on-device AI smartphones is expected to rise significantly in 2026. Due to the proactive initiatives of chipset companies, on-device AI features are expected to be deployed downwards from 2025 onwards, initially only in flagship chips. Eventually, even mid- and low-priced smartphones are expected to take advantage of basic on-device AI features such as photo editing, camera functions, and translation functions, which will encourage users' interest in using artificial intelligence and accelerate market growth. In 2024, the total shipments of on-device AI smartphones are expected to be 119 million units, accounting for 10% of the total shipments. After that, sales are expected to grow steadily year by year, reaching 606 million units in 2028, with a compound annual growth rate of 38%. From 2026 onwards, the penetration rate is expected to increase to 47% in 2028.

Source: Omdia

Hybrid-AI will accelerate the application and popularization of AI functions on smartphones

Both On-device AI and Cloud-AI solutions have obvious advantages and disadvantages. First, the On-device AI solution can achieve low-latency response and personalized AI services with the chip computing power on the end side, while retaining privacy data on the end side, eliminating users' concerns about data security. However, currently, smartphones that support On-device AI require high-specification hardware, which has high hardware costs, and the cost of purchasing devices for users is naturally high.

Cloud-AI has a powerful computing center to support it, and can run models with larger parameters to obtain more accurate generative AI content. However, the user's AI experience is affected by the quality of the network signal. In addition, the generated content is general, with less personalized content, and users are more concerned about data security. Therefore, the Hybrid-AI solution has become an inevitable path for the development of AI smartphones. The use of Hybrid-AI can take into account the personalization, universality, data security and scenario diversification of AI functions. Tasks involving user privacy data are kept on the end side, and other general AI services are placed in the cloud. Such a solution reduces the hardware requirements for the device to run AI capabilities, and will accelerate the penetration rate of AI functions on smartphones. At Apple's Developer Conference in 2024, Apple demonstrated the prototype of the future application of hybrid AI in terminal devices. Through the combination of 3 billion end-side models, private cloud computing and third-party public cloud computing, it can quickly penetrate AI capabilities into various terminal devices.