Summary of Key Points
AI-powered smartphones are shifting from being mere companions for chatting and engaging in cute interactions to becoming intelligent agents that can perform tasks on behalf of users. This transformation is driven by a combination of policy changes (which have both facilitated the registration of large-scale models on mobile devices and restricted the use of more anthropomorphic emotional features) and the strategic efforts of smartphone manufacturers seeking new ways to thrive amidst declining hardware profits and waning consumer desire to upgrade their phones. The competition now revolves not only around technology (models, computing power) but also around who controls access to users, how traffic is distributed, and who owns the data. Manufacturers have adopted three main approaches: hardware innovation, reconstruction of underlying systems, and ecosystem collaboration. However, the ultimate challenge lies in breaking through the barriers imposed by dominant apps, balancing privacy and commercial interests, and possibly redefining the relationship between humans and machines.
Policy Changes: Setting Boundaries and Defining the Path for AI Phones
Two policies issued on July 15 marked a critical turning point:
- Relaxation of Restrictions: For the first time, large-scale models running on mobile devices (such as Huawei's Xiaoyi and Apple's Smart Services) were permitted to be registered officially, giving AI phones the legal right to operate on users' devices. Previously, due to concerns about data privacy, such models faced significant regulatory hurdles. With these new regulations, manufacturers can now advance their AI technologies with confidence.
- Restriction of Emotional Functions: On the same day, the "Interim Measures for the Management of Humanoid Interaction Services in Artificial Intelligence" were implemented, leading to the discontinuation of certain features like virtual lovers and AI best friends offered by companies like ByteDoubao and Alibaba's Qianwen. The rationale is simple: such functions can easily lead to user addiction, blur the lines between reality and fiction, and fail to generate real revenue (chattings may increase usage time but do not convert into profits). The policy has shifted the industry from creating fictional entities to providing practical services.
The policy approach is clear: while it's easier for AI to mimic human behavior, its ability to perform actual tasks is seen as a more genuine and secure application of its capabilities.
Manufacturers' Desperate Struggles: The Need to Rely on AI Phones
Smartphone manufacturers are rushing into the AI phone market out of necessity rather than choice:
1. Dropping Hardware Profits: The cost of memory chips has skyrocketed (DRAM up by over 50% and NAND by nearly 75% in the second quarter), with storage accounting for 65% of the cost of low-end phones and over 30% of mid-to-high-end models. The traditional model of making profits through hardware sales is no longer viable, forcing manufacturers to explore new revenue models (such as subscription-based services and software royalties).
2. Low Consumer Interest in Upgrades: With screen quality, imaging capabilities, and fast charging reaching their limits, consumers are less inclined to upgrade their phones. Counterpoint predicts that global smartphone shipments will drop to 1.08 billion units in 2026 (a historic low), but AI phones are expected to grow, accounting for 45% of the market by then. This represents the only potential source of new growth for manufacturers.
3. Controlling Data: In the past, smartphone manufacturers were merely intermediaries for app data; user activities on platforms like WeChat and Taobao were collected by those platforms. With AI enabling cross-app functionality (e.g., booking flights, hailing taxis, notifying colleagues), manufacturers can gain direct access to valuable user data, which is much more lucrative and can help retain users.
In short, AI phones that merely engage in trivial interactions are unprofitable and potentially violate regulations, while those that can perform practical tasks represent a lifeline for manufacturers.
Technical Challenges: A Compromise Between On-Device and Cloud-Based Solutions
For AI to be useful, several technical issues must be addressed:
1. Insufficient Computing Power: Even the most advanced smartphones' NPU capabilities will not exceed 100 TOPS by 2026. Running a model with 7 billion parameters for a logical task may take three minutes and still result in errors. Offloading all tasks to the cloud is too costly for manufacturers to sustain.
2. Privacy Concerns: The more AI understands users, the more data it needs. Storing data locally leaves it vulnerable to breaches, while transmitting it to the cloud poses security risks. Current solutions involve partial data anonymization (e.g., keeping sensitive information like payment passwords local and sending only partially processed data like flight requests).
3. Ecosystem Barriers: AI's ability to cross-apps threatens the dominance of established apps. For example, WeChat and Taobao are reluctant to allow AI to directly interact with their user interfaces, as it could disrupt their traffic patterns and business models. Nubia's AI phones have been successful overseas but face resistance in China due to these restrictions. WeChat's recent A2A (AI-to-WeChat) collaboration model shows a limited degree of openness, indicating that apps are reluctant to cede control.
Three Different Approaches: Which Path Will Lead to Success?
Manufacturers are exploring three main strategies:
1. Honor: The company released the Robot Phone, equipped with a four-degree-of-freedom titanium alloy mechanical gimbal that allows for interactive actions and can perform multiple tasks like ordering cakes, hailing taxis, and booking KTVs. The idea is to transform AI from a tool into a partner that enhances user interactions with the physical world.
2. Step Star: This company is developing its own underlying operating system (Step AOS) from scratch, aiming to make AI an integral part of the system rather than a separate component. They believe the future of smartphones lies in tasks-driven systems where users request services and AI manages resources.
3. Nubia: Collaborating with ByteDoubao, Nubia is moving towards A2A (AI-to-app) interactions, allowing apps to directly execute tasks via built-in intelligent agents. This approach has the advantage of compatibility with existing ecosystems but comes with limitations, as it depends on the app's willingness to cooperate.
None of these approaches is definitively superior, but they all face the same question: who will set the rules when AI performs tasks on behalf of users (e.g., in cases like money transfers)?
The Ultimate Question: Are AI Phones a Passing Fancy?
Some argue that smartphones' dominance will decline by 2026, with the rise of PCs, cloud computers, and AR glasses. The ultimate goal of AI should be to provide seamless, proactive services rather than being confined to screens. For example, Honor's use of a mechanical gimbal in its phones is a way to compensate for the limitations of traditional phone form factors. OpenAI's investments in earbuds and glasses suggest a shift towards personal intelligent agents that can accompany users throughout the day without the need for screens.
For ordinary consumers, the key to evaluating AI phones lies in two criteria:
- Whether they can actually perform useful tasks (e.g., booking flights, handling expenses, arranging travel arrangements) rather than just making empty promises.
- Whether they save time or require additional costs from users.
The real question is not whether AI phones have a future but whether users are willing to grant them sufficient permissions and how we can ensure that human autonomy is not replaced by AI.
Smartphones have transformed the relationship between humans and information, and the next generation of AI-powered devices will further redefine our interaction with tools. Ultimately, the success of AI phones depends on their ability to become reliable assistants rather than mere tools for harvesting user data.
In Conclusion: AI phones represent manufacturers' attempts to overcome hardware-related challenges. While policy changes are guiding this transition from casual interactions to practical applications, true success requires solving issues related to computing power, privacy, and ecosystem integration. Only by addressing these challenges can we determine whether AI phones will become a viable solution or just another means of collecting user data.