WAIC 2026: Farewell to Parameter Race, AI Agents and Embodied Intelligence Usher in a New Era of AI Deployment

The 2026 World Artificial Intelligence Conference (WAIC) was grandly held in Shanghai, attracting over 500 exhibitors and marking the AI industry's official entry into the 'second half': moving beyond mere parameter competition in large models toward the commercialization of AI agents and the practical implementation of embodied intelligence.

From 'AI Features' to 'Native Systems'

The most prominent trend at this year's conference was the central role of AI agents, now seamlessly integrated across hardware and software. In end-user devices, AI smartphones have entered their second phase—evolving from traditional phones with bolt-on AI features to 'native AI agent phones.' Honor, StepFun Celeste, and ZTE Nubia all unveiled new models equipped with AI agent operating systems, emphasizing the concept of 'letting the phone do tasks on its own.' These devices can execute long-horizon, complex tasks across multiple apps, such as automatically placing delivery orders or synchronizing multiple bookings.

Additionally, AI glasses and earbuds are becoming lighter and more function-specific, such as iFlytek's translation glasses and Alibaba Qwen's AI agent earbuds, integrating AI into personal memory and real-time communication scenarios. On the enterprise side, AI agents are emerging as 'AI colleagues' or 'AI employees,' assisting not only with text processing but also deeply engaging in decision-making and business process reengineering.

Zhang Yu, Vice President of Dmall, pointed out that the retail industry is undergoing a leap from 'delivering software' to 'delivering outcomes.' Through its self-developed intelligent retail brain 'D-BRAIN,' the company orchestrates clusters of AI agents to achieve efficient closed-loop operations—from product selection and pricing to supply chain execution.

The Year Robots 'Truly Work'

Embodied intelligence made a dramatic leap in scale at this year's event, with 208 embodied AI devices on display. Unlike previous years' demonstration-only robots, 2026 is being hailed as the 'Year of Real Work' for robots. They are now officially 'on the job' in pharmacies, restaurants, hotels, and retail stores, performing tasks such as sorting, packaging, cooking, and even traditional Chinese medicine massage.

Technologically, embodied intelligence has evolved into a 'four-chain resonance' of 'whole machine—brain—dexterous hand—data.' The 'dexterous hand' has become a core component, with significantly enhanced tactile perception and fine manipulation capabilities, now capable of tasks like painting and threading needles. Meanwhile, data collection has turned into the 'oil war' of embodied AI, with major vendors racing to build data factories, using real-world robot training and simulated data to overcome the scarcity of labeled data.

From 'Selling Compute' to 'Token Factories'

At the infrastructure level, the metric for computing power has fundamentally shifted—'Token' has replaced 'FLOPS' as the new unit of account. AI vendors are transforming into 'Token factories' or 'Token stores,' competing on the ultimate cost efficiency: 'how many effective Tokens can be produced per yuan, per watt.'

Simultaneously, the narrative of domestication has become mainstream, with multiple vendors emphasizing 100% domestically produced chips and full systems. Giants like Huawei, Alibaba, and ZTE have launched 'ultra-node' products, leveraging system-level innovations to close the gap in single-chip performance. Some have even unveiled '100,000-card AI superclusters,' supporting the development and inference needs of 3-trillion-parameter-scale large models.

The Rise of AI-Native Organizations

According to Chinese media outlet Zhidx, this year's WAIC reflects a comprehensive awakening in the AI industry: large models are no longer just Q&A tools but are fundamentally reshaping organizational productivity through AI agents and embodied intelligence. Experts predict that within the next 12 to 24 months, AI-native organizations will transition from concept to reality, and human-AI coexistence will become the norm. In this era of rapid AI advancement, the ability to deliver verifiable task outcomes has become the new standard for evaluating AI's value.