Written by | Wen Yehao

Edited by | Wu Xianzhi

During a post-launch communication session, a mobile phone manufacturer remarked, "I have a special AI meeting every week." This statement underscores the growing importance that mobile phone manufacturers are placing on AI functions, which need to align with users' intuitions.

Technological advancement is relentless. Each time we believe we've hit a roadblock, a new force emerges to break through existing paradigms, leading humanity out of stagnation and fostering new opportunities.

In the recent flurry of year-end smartphone releases, brands like OPPO and Xiaomi have showcased breakthroughs in ultra-narrow bezels, which have required significant investments in time and resources. These advancements involve intricate glue injection processes and chip-level dust-proof measures, demanding substantial investments from manufacturers. However, these enhancements often go unnoticed by users.

As the hardware development of smartphones nears its physical limits, with screens, cameras, and processors offering limited room for improvement, the industry felt on the brink of despair, fearing the end of the smartphone era. Yet, the advent of edge AI has opened new avenues.

While AI integration into scenarios like imaging and voice assistants began around 2017, 2024 can be deemed the true "first year" of AI phones. This year, the mobile phone industry has rebounded, and the wave of AI has transitioned from laboratories to practical applications, coinciding perfectly with the timing of AI phones.

Against this backdrop, AI phones have evolved from being a "bonus" feature in select products to a "must-have" across all models. This transformation has not only propelled the industry forward but also ignited new business ecosystems.

AI Phones Need to Prove Their Worth

AI stands as one of the driving forces behind the mobile phone industry's overall recovery in 2024.

According to Canalys, the penetration rate of AI phones is expected to reach 17% in 2024 and accelerate further in 2025, pushing the global penetration rate to 32%. AI phones have achieved in just over a year what foldable phones, considered the industry's "elder brother," failed to do in six years.

Despite the rapid growth in data, the statistical criteria are nuanced. Amid manufacturers' vigorous product promotions, virtually every newly launched model is now associated with AI. AI has almost become a nuclear option for flagship phones – it may not always be used, but it must be present.

This stems from the smartphone industry's desperate need for a new gimmick to rejuvenate itself, particularly after the foldable phone trend faltered.

This story began a few years ago when the smartphone industry narrowly averted a crisis. Moore's Law, which had propelled the consumer electronics industry forward, hit a bottleneck. This directly contributed to the failure of mobile phone and software manufacturers' strategies to promote hardware upgrades.

In a panic, manufacturers even replicated ray tracing, a technique used on PCs to create additional barriers, to mobile phones. However, the mobile phone market is not as tightly bound to gaming and productivity as the PC market, leading to underwhelming results. Other rescue attempts, such as focusing on imaging and foldable screens, had limited impact, primarily serving niche demands rather than driving large-scale replacement waves.

However, AI is different. Edge AI differs significantly from past explorations of smartphone performance. An industry insider told Photon Planet that, from a hardware perspective, AI models rely less on small cores and place greater emphasis on NPU capabilities, ideally supported by large memory and high memory speeds.

Therefore, for smartphones to truly become AI phones or provide a superior AI application experience, hardware reconstruction is inevitable. This naturally implies a need for phone replacements. To effectively communicate this, it's crucial to convince the broader consumer market to invest in AI phones.

According to Photon Planet's mid-year research this year, AI phones were largely ignored in stores, with both customers and store staff showing little enthusiasm. A recent SellCell survey revealed that 73% of Apple Intelligence users and 87% of Galaxy AI users believe that AI offers little to no value in their usage.

This is understandable as functions like summarizing PPTs, call summaries, and image generation are not useful for most users. No matter how much they are iterated and upgraded, AI phones are ultimately constrained by niche demands, potentially repeating the foldable phone's mistakes.

Therefore, demonstrating the value of edge AI to users as soon as possible has become a top priority for mobile phone manufacturers.

Looking at the new phones released in the second half of this year, from Apple, Samsung, Honor, Xiaomi, OPPO, to Vivo, promotional efforts have been abundant. During launches and in store materials, these brands heavily promoted their AI capabilities, building upon previous AI-based image editing and conversational features with numerous system-level deep applications.

Among these, the most imaginative and attractive is the "autonomous driving" realized by AI screen reading. By accessing screen information, AI interacts with applications automatically to execute user commands – the "one-sentence coffee order" demonstrated by Zhao Ming at the Honor launch event is based on this capability. In the consumer context, Honor, which was the first to unveil this trump card, has successfully captured users' attention.

This signifies that at this stage, AI phones have evolved from integrating AI toys and tools to deeply penetrating mobile phone interactions and system levels, demonstrating a disruptive impact on traditional phones. This deeply embedded AI interaction has the potential to drive a true reshuffle of edge AI and may foster new business models for mobile phone manufacturers.

AI as a Lifeline

"Performance overcapacity is a myth in the AI era. Gaming and photography scenarios have extremely high performance requirements, and stacking hardware is not entirely futile. Advances in chip manufacturing processes actually provide broader space for AI."

A mobile phone manufacturer stated that edge AI places high demands on systems, even higher for consumer products. For example, heat dissipation, which seems simple, is actually quite challenging to implement. The mobile phone industry is only just beginning to address underlying memory usage issues. Over the past year, many mobile phone manufacturers have made numerous optimization efforts for edge deployment, often overlooked due to weak consumer perception.

The biggest bottleneck in edge deployment lies in the utilization of mobile phone computing resources, particularly memory usage. As more edge functions are added, more computing resources are occupied. Since hardware manufacturers generally do not reserve specific memory for these functions, most manufacturers dynamically allocate resources based on the demands of edge AI functions.

Taking OPPO as an example, its previously launched edge architecture "AI LoRA" employs a dynamic allocation solution. Specifically, the architecture provides three edge functions, with one occupied by the base model and the other two adjusted dynamically based on actual conditions. It is reported that this solution can save up to 75% of peak memory usage.

The gradual maturation of architectures signifies the initiation of large-scale ecosystem transformations. A mobile phone manufacturer commented that "next year's launch events will likely be even longer." This is because descriptions of hardware will be replaced by AI functions, which rely on user experience and require more time to present.

In addition to mobile phone manufacturers, large model companies like Zhipu have also launched quick execution of complex tasks through natural language commands based on agent task planning and screen information understanding capabilities at the end of the year.

An industry insider stated that the relationship between mobile phone manufacturers and application developers is not entirely adversarial but rather cooperative at this stage. "For developers, just because users download an application doesn't mean they use it. Edge AI can help increase the frequency of usage, and mobile phone manufacturers also need to cooperate with developers to make edge AI assistants smoother to use."

Application vendors' most obvious advantage lies in consumer data within their vertical fields. Taking photo-triggered services as an example, connecting to OTA alone requires tens of millions of vertical data points for specialized training. Mobile phone manufacturers lack both relevant data and the need to reinvent the wheel, thus forming stable cooperation with developers.

"In fact, direct access, such as for navigation and shopping, is welcomed by application developers because mobile phone manufacturers directly push services to users. A mobile phone manufacturer's responsible person said that existing mobile services are already saturated, and they are now more focused on how to quickly connect users to services, especially when they need them. The shallower the entry point, the higher the access frequency will be.

Even so, in the demonstrations of AI functions in new phones at the end of the year, we can still see that the discourse power of mobile phone manufacturers is growing with the acceleration of edge deployment. When Zhao Ming demonstrated ordering coffee from Luckin using Honor or when Magic OS 9.0 launched one-click shopping price comparisons, few people considered whether to open WeChat Mini Programs, Meituan, or Ele.me. Is this based on user habits or influenced by a certain degree of commercial weighting?

Undoubtedly, relying on direct access, the smart assistants of various mobile phone manufacturers will potentially change existing mobile phone interaction forms – from mobile applications to smart assistants and from simple touch interactions to a variety of interaction forms including touch, natural language, and intent recognition. This means that the "entry point" status of mobile phone manufacturers is further strengthened, and their control over application distribution will increase in the future – this will be another battle for the "traffic entry point" after pre-installed applications.

Can User Habits Be Changed?

Whether users are willing to pay for AI phones is the ultimate measure of success.

Earlier, SellCell surveyed 1,000 iPhone and Galaxy users each, and the results overall showed a mismatch between purchase decisions and usage habits. While most people are skeptical about the value of AI functions and have low willingness to pay, nearly half of users consider AI functions an important decision-making factor. In other words, users believe AI functions are important but have poor actual usage experiences.

The reason is that large models originally designed for PCs do not fully adapt to mobile environments. Most existing large model vendors rely on PC browsers to deliver services, emphasizing productivity tools, which do not fully align with mobile scenarios. For example, in public settings, voice interaction is clearly inferior to touch interaction, and the frequency of use for document summarization and article writing is lower than that for generating social media posts.

Recently, well-known Bloomberg technology reporter Mark Gurman stated that iOS 19 will abandon access to ChatGPT and instead adopt a proprietary model.

Another challenge for AI phones comes from user habits. Nowadays, with the traditional mobile phone operation logic already well-established, AI functions have indeed greatly simplified user habits. However, adapting to new interactions still requires time from users. Just as it took several years for users to accept the mobile ecosystem and new touch interactions when transitioning from feature phones to smartphones.

Currently, mobile phone manufacturers are continuously enhancing AI functions, such as AI screen reading, perhaps hoping to complete scenario or business model explorations ahead of the next era.

It remains uncertain whether technology can drive changes in user habits. After all, mobile phones are tools over which users have a strong desire for control. AI may bring excessive 'intelligence,' making users feel a loss of control. Gradually guiding users to accept and rely on AI remains a market challenge that manufacturers need to overcome.