Login
Sign Up
Woofun AI reports that the MWC26 Shanghai exhibition marked a definitive transition from theoretical AI inquiries to tangible, cross-application execution, showcasing devices that operate with minimal cloud dependency. The event displayed a Doubao phone executing complex tasks via single commands, AI glasses projecting real-time translations and summarizing meetings, and humanoid robots achieving nearly 100% success rates on factory lines. IDC predicts that by 2026, shipments of traditional AI terminals in the Chinese market will exceed 300 million units, with the penetration rate surpassing 93% by 2027. To highlight this integration, MWC26 Shanghai introduced the "Mobile AI Innovation Pioneers" special area, demonstrating how smartphones, smart glasses, and smart toys are embedding artificial intelligence into daily life.
The primary obstacle to widespread AI terminal adoption remains the source of computational power, or hash rate, as reliance on cloud infrastructure introduces insurmountable barriers regarding latency, cost, and privacy. Lenovo addressed this through a strategy of localization with its AI Host P7, a palm-sized device weighing approximately 300 grams with a maximum power consumption of 30 watts. Despite its compact form factor, the device delivers a hash rate of up to 190 TOPS, supporting local large models with up to 122B parameters and handling context information up to 128K in length. Even in a completely offline environment, the system performs local inference at a speed of 50 Tokens per second, allowing users to communicate with AI, organize knowledge, and search files without incurring additional costs or consuming cloud-based Tokens.
Staff at the Lenovo booth confirmed that the company aims to generate 80% of the Tokens required for its devices locally, restricting cloud collaboration to only 20% of operations. This strategic pivot reflects a direct response to the current Token shortage, positioning the reduction of Token production costs as a mechanism to control pricing power in an era where Tokens function as the primary currency. The Cyber Machine, another device showcased at Lenovo's booth, reinforces this approach by enabling users to classify, search, and build knowledge bases from uploaded documents, images, audio, and video entirely on the local terminal. With a single command, the system utilizes local AI to create schedules and send emails, eliminating the need to upload private documents to the cloud for analysis by large models.
Woofun AI data shows that AI glasses are evolving beyond simple Bluetooth peripherals with camera capabilities to incorporate practical functions like real-time translation, navigation assistance, and teleprompters. Although current models still rely on Bluetooth connections to smartphones for computing power due to constraints in weight, size, and battery life, industry staff indicated that future improvements in local chip capabilities will enable independent operation without a smartphone. This progression signifies a critical shift in how hash rate is delivered to end-users, moving the computational burden closer to the point of interaction to enhance responsiveness and privacy.
Once sufficient hash rate is secured, the next challenge involves redefining human-computer interaction, which has historically revolved around the lifecycle of opening, using, and closing applications over the past two decades. Fang Fei, President of Honor's product line, outlined three major shifts: the interaction mode moving from graphical interfaces to Agentic UI based on intention expression, the value system transitioning from apps to managing user context, and the distribution logic shifting from consumer focus to intelligent agent focus. Fang Fei stated that mobile terminals in the next decade will cease to be containers for apps and instead become stages for intelligent agents, envisioning an Agentic OS that is user-centered with four core attributes: intention-driven operation, natural interaction, proactive intelligence, and inherent cross-platform capability. In this future architecture, perception becomes distributed, with smartphones acting as the computing hub while AI perception capabilities are spread across watches, headphones, and glasses to free users' hands.
ZTE demonstrated this new interaction paradigm with its AI-powered native phone, the Nubia M153, developed in collaboration with Doubao, which allows users to complete complex tasks by stating requests in natural language. For instance, a user wishing to post a news update about MWC on Weibo can simply ask the phone, which will automatically invoke the Weibo application to write and publish the update without manual intervention. Privacy and security concerns remain a topic of discussion, yet the convenience of such systems is evident, as seen with the AI pet iMoochi at ZTE's booth, which perceives user emotions through touch and sound to respond with eye movements and a unique language.
The evolution from one-dimensional command lines to two-dimensional graphical interfaces and now to three-dimensional embodied interactions suggests that humanoid robots will serve as the "body" of AI, directly embedded in the physical world. Peng Zhihui, Co-founder, President, and CTO of Yuanjing, proposed in his keynote speech that the biggest consumers of Tokens in the future will be embodied intelligent agents operating in the real world. These agents continuously perceive their environment, understand tasks, plan actions, and provide feedback within a task space that encompasses both digital and physical realms, fundamentally altering the scale of computational demand.
This perspective underscores a core bottleneck in the robotics industry: data availability. Han Zheng, Co-founder of Sudu Technology, argued that relying solely on physical robots to collect real-world data is insufficient for training general-purpose large models, and combining small models for specific scenarios fails to achieve the same goal. The industry faces common challenges regarding the effectiveness of simulation data and the efficiency of collecting real-world data, prompting Yuanjing to develop a 'triple intelligence' framework comprising ontology, motion intelligence, task intelligence, and interaction intelligence. This framework aims to determine how robots can integrate into real workflows to create value and interact naturally with humans, marking a critical step toward resolving the data scarcity that limits the deployment of embodied agents.