Summary of Key Points
Xiaomi has released three Xuanjie chips (O3, O100, D100), which address three major challenges in AI performance at the edge (smartphones, autonomous driving, etc.): energy efficiency, bandwidth, and computing power. These chips form the foundation for a comprehensive AI computing ecosystem that encompasses humans, vehicles, and homes. While Xiaomi has demonstrated that its hardware can support AI operations more efficiently and with less power consumption, it has not yet clarified what kind of "AI-powered smartphones" these chips will make possible—there is a lack of a unique user experience, and the "last mile" in AI application development has not been overcome.
1. The Three Chips: Each Solving a Specific Problem
Xiaomi’s three chips act as three types of "assistants" for AI, each tackling a different issue:
- Xuanjie O3 (for smartphones): Making AI more energy-efficient and faster
This chip distributes AI tasks among different components: simple tasks (such as voice assistant responses) are handled by the CPU, gaming-related AI (e.g., image enhancement) by the GPU, and complex tasks (e.g., AI-generated images) by a dedicated NPU (Neural Processing Unit). It also uses compression technology to reduce the size of AI models to an average of 2.6 bits, resulting in 45% faster decoding performance and 26% lower power consumption. The O3 will be available with the Xiaomi 18 Fold foldable phone in September.
- Xuanjie O100: Breaking the "memory bottleneck"
The bottleneck in edge AI is not insufficient computing power but slow data transfer. The O100 uses 3D stacking technology to integrate the storage and computing chips, increasing data transfer speeds to 1.22 TB/s (equivalent to downloading 305 high-definition movies per second), and significantly speeding up AI text generation to 330 words per second.
- Xuanjie D100: Providing sufficient computing power for autonomous driving and large models
This 3nm chip supports up to 160GB of memory and can run large models with 200 billion parameters (similar to ChatGPT). It is primarily used for autonomous driving and complex local AI tasks and is expected to be commercially available in 2027.
2. Hardware is Ready, but the Definition of AI-Powered Smartphones is Still Unclear
Xiaomi claims that its smartphone hardware is "AI-ready," but when asked what an AI-powered smartphone is, the company’s responses are vague. It either says it can only be announced after compliance issues are resolved or that it’s difficult to define. The real issue lies in the lack of a unique user experience. For example, current AI features are often standalone: image editing, summarizing texts, or using voice assistants, which other smartphones can also perform. There is no "system-level" AI interaction that seamlessly integrates multiple tasks (e.g., automatically booking a flight and setting reminders). Additionally, it’s not clear whether Xiaomi’s smartphones would still be considered AI-powered if they used different chips (e.g., Qualcomm’s).
3. The "Application Barrier" is More Difficult to Overcome than the "Memory Barrier"
Although Xiaomi has made significant progress in improving data transfer speeds, the real challenge for AI-powered smartphones is the "application barrier":
- Existing AI features are more like enhancements rather than game-changers; for instance, AI image editing makes the process faster but doesn’t fundamentally change user habits.
- System-level AI requires collaboration among multiple components, addressing issues such as model reliability, application permission management, and privacy protection, which cannot be solved by chips alone.
- Users need a compelling reason to upgrade to AI-powered smartphones. If they only offer slightly faster performance and lower power consumption, they may not be willing to pay a higher price.
4. Xiaomi’s Next Step: From "Functionality" to "Usability"
The Xuanjie chips lay the foundation for Xiaomi’s AI capabilities, but to create commercially successful AI-powered smartphones, the company needs to do three things:
1. Provide a unique experience: Features that are exclusive to Xiaomi’s chips, such as running large models locally or generating high-quality content offline, or providing faster response times in autonomous driving (which is crucial for safety).
2. Define the standards for AI-powered smartphones: Establish clear criteria, such as the ability to run models with 10 billion parameters and perform cross-application tasks, to highlight the benefits of Xiaomi’s chips.
3. Integrate the ecosystem: Ensure that Xiaomi’s operating system, models, and applications work together seamlessly, allowing AI to adjust settings based on user habits or switch between devices (e.g., between a car and home).
In summary, while Xiaomi has demonstrated its ability to develop AI chips, it still needs to show that users truly need its AI-powered smartphones. This is the key to success.
In One Sentence
Xiaomi’s Xuanjie chips have established the hardware foundation for AI, but to convince users to buy AI-powered smartphones, the company must provide a compelling user experience and establish clear standards for what such devices should offer.