2025 WAIC's most complete observation: From L4 smart devices to robots that can carry express deliveries, AI has actually been implemented!

The Zhitong Finance App learned that the 2025 World Artificial Intelligence Conference (WAIC) was large-scale and had many participants, focusing on the latest technological developments and commercial applications in the field of artificial intelligence. This article gave an overview of the robot industry, large models and AI applications, hardware innovation, and market and industry ecology at this conference. Recently, there has been a clear rebound in the AI sector. Most companies have positive profits, up to 50% +, and the worst 10% to 20%. Affected by WAIC events exceeding expectations (ticket prices are difficult to buy, suspended after July) and GPT5, the market is continuous. Institutions have adopted a strategy of cutting high and low to simulate positions to maintain 90% of positions. A-share computer/AI applications are in beta, hot spots are rotating rapidly, and there is no sustained alpha return; corporate performance is growing steadily (most revenue growth of 10% +), and the scale of AI revenue is still small and has not reached the explosion stage.

I. Basic Overview of the Conference

The 2025 World Artificial Intelligence Conference (WAIC) was large-scale and had many participants, focusing on the latest technological developments and commercial applications in the field of artificial intelligence. Among them, the robot exhibition area showed outstanding performance. The number of exhibitors, exhibition area and number of visitors all reached record highs — the robot exhibition area increased by about 40% compared to last year, and more than 40 biped humanoid robot manufacturers participated in the exhibition, plus a total of about 60 industrial robot manufacturers, which doubled from last year; the number of dexterous hand manufacturers reached 9, and the number of rotary joint module manufacturers increased significantly, and integration became an industry trend.

II. Exhibition area features and technical highlights

Significant differences in zoning techniques

H1 exhibition area: Mainly by major manufacturers such as Baidu and Ali, the booth area has been expanded, the application display is rich and diverse, and the implementation of end-side models is progressing rapidly. For example, automotive end-side models achieve intelligent interaction optimization, and PC-side linearization end side models improve document processing efficiency, marking a breakthrough in AI from 0 to 1.

H4 exhibition area: focuses on emerging technologies and cutting-edge applications, focuses on showcasing AI innovations in niche fields such as MDSE (Model-Based Systems Engineering) design, shortens the design cycle of complex industrial systems, and lays the groundwork for industrial transformation.

Robot exhibition area: Leading manufacturers (Zhiyuan, Yushu, Fourier, Leju, etc.) have concentrated resources, large booth area, and have entered the commercial mass production stage. The number of manufacturers of core components such as dexterous hands and rotating joint modules has surged, and technological maturity has improved markedly.

3. Robotics industry: from technological breakthroughs to commercialization

1. Product and scenario progress

Industrial scenarios: Wheeled robots cooperated by Zhiyuan and Dema Technology to achieve autonomous package sorting, with an error of ≤ 5 cm; Flying Dragon Technology robots can accurately grab fragile products such as glass and chips with self-developed torque sensors; Kepler biped and wheeled robots have strong battery life and focus on material handling; Galaxy GM's wheeled chassis has high mobility efficiency, and factory material handling efficiency is 3 times higher than manual labor.

Business scenario: The Galaxy General Supermarket sorting robot performed steadily and operated for 8 hours without errors.

Rehabilitation medical scenarios: Fourier medical exoskeleton robots have entered dozens of hospitals in China to help people with walking disabilities recover; their biped humanoid robot is used to help people with cognitive impairment and autism recover, and the next-generation family emotional support robot GR 3 (similar to “big white”) was unveiled for the first time.

Industry data: Leading manufacturers are expected to ship up to 5,000 units in 2025, mostly 500-2000 units, totaling more than 10,000 units.

2. Challenges and technical directions

Currently, most robots still require manual control or lack autonomy, and need to improve the level of “brain” intelligence (multi-dimensional perception, autonomous decision-making, and dismantling of complex tasks). There is still a gap between domestic companies and overseas Tesla, Bigger, etc. in terms of autonomy.

Technical trends: integration of rotating joint modules, improved dexterity; in the core component process, encoders (eddy current encoders, rotary transformers), speed reducers (planetary RV cycloid pin wheel actuators), screws, motors, etc. have become key breakthroughs, involving companies such as Yap Co., Ltd., double ring drive, Hengli Hydraulic, and Buke Co., Ltd.

4. Large models and AI applications: from “quantity competition” to “effective implementation”

1. Big model development trend

From the “100 Model Wars” in 2023 to the convergence of the pattern, the core manufacturers continued to iterate: Shangtang released the “Japan-Nissin 6.5” multi-modal inference model, using a graphic interlaced thought chain design, improving reasoning ability and reducing costs by 70%; mini max launched the open source large-scale hybrid architecture reasoning model M1, making breakthroughs in the field of large video models and intelligent agents.

Domestic startups have achieved breakthroughs in multi-modal reasoning with limited computing power. In particular, progress has been remarkable in understanding long videos, enterprise services, and mobile applications; K2, the overseas version of Kimi's big model, enhances reasoning capabilities through architectural optimization, and has become an alternative to mature overseas models.

2. Application scenarios and industry focus

Industry-wide service agents: Demand for computing power (especially inference computing power) has surged, and large models focus on H parameter capabilities (coding, tool calls, etc.), penetrating from general applications to the industry.

Breakthroughs in various categories: In the medical field, iFLYTEK is deeply involved in AI diagnosis, and the learning machine market relies on “AI personalized coaching” to maintain a high growth rate; in the financial sector, Jinyue Star cooperates with brokerage firms to automate financial report analysis, research and disassembly; in the industrial sector, Baoxin showcased the energy efficiency optimization application of AI in the steel industry.

Industrial Internet Benchmarking Case: Shenyuan Artificial Intelligence and Kaos CosmoPlat, a subsidiary of Haier Group, attracted attention. MasterAgent, the world's first L4 smart parent system, was unveiled. The two sides cooperated deeply around the core scenarios of the industrial Internet to achieve intelligent talent recruitment, automation of the entire customer service process, and intelligent scheduling of enterprise workflows, further unleashing the potential of AI agents in this field.

5. Hardware Innovation and Enterprise Dynamics

1. Smart hardware highlights

Go out and ask about Tik Note: As a personal information terminal, it supports recording and transcribing summaries, mind maps, and voice interaction to build a personal knowledge base.

Kuaishou Smart Animation: Multi-person real-time collaborative AR creation platform. The 10-second video generation cost is less than 10 yuan, and it is quickly transformed into a productivity tool from the laboratory stage.

Titanium Dynamic Technology promotes NAVOS: overseas marketing agent, relies on a large model to provide full-link services to solve the problems of high localization costs and scarce talents.

iFLYTEX multilingual AI smart screen: The transparent screen supports barrier-free communication across languages, and is equipped with a newly released AI voice recorder to enhance scene coverage.

2. Enterprise product iteration

Jinshan released WPHAI 3.0, and the Lingxi product changed from WPS toolbar interaction to an independent portal to an autonomous smart device.

Shangtang Wuneng is an intelligent platform: Based on the expansion of the autonomous driving world model, 100,000 3D assets are used to enhance the robot's vision and motion control capabilities.

6. Market and industrial ecology

1. Market conditions and investment trends

Recently, there has been a clear rebound in the AI sector. Most companies have positive returns, up to 50% +, and the worst 10% to 20%; affected by WAIC events exceeding expectations (ticket prices are difficult to buy, suspended after July) and GPT5, the market is continuous. Institutions have adopted a strategy of cutting high and low to simulate positions to maintain 90% of positions.

A-share computer/AI applications are in beta, hot spots are rotating rapidly, and there is no sustained alpha return; corporate performance is growing steadily (most revenue growth of 10% +), and the scale of AI revenue is still small and has not reached the explosion stage.

2. Industrial Ecology and Policy Support

Changes in the structure of exhibitors: data labeling vendors (such as Aopeng China) customers have switched from model manufacturers to enterprises/governments; application vendors (Youdao, Convinced, etc.) have increased in volume, and the number of pure model manufacturers has decreased.

Local government layout: The Yangtze River Delta region with Shanghai at its core has become the core area of the AI industry, attracting enterprises to land through policies such as computing power vouchers and application subsidies, and using local computing power resources to build industrial clusters.

Ecology between China and the US: The popularity of entrepreneurship is unabated, industrial metabolism is active, computing power is a key investment area, and outstanding talents focus on specific categories such as AI video, coding, and for science.

7. Summary and outlook

WAIC 2025 shows the accelerated penetration of AI from technology research and development to commercialization: the robot industry is moving from “dazzling technology” to practical use, large models are shifting from “parameter competition” to “value creation”, and hardware and applications continue to reduce costs and be close to people's livelihood. Despite challenges such as homogenization of smart devices and insufficient robot autonomy, government support, capital activity, and technological iteration are jointly driving the industry forward. As a young and dynamic industry, AI has broad future prospects.

Note: This article is based on the feelings of analysts from various brokerage firms such as CICC and Cathay Pacific Haitong after attending WAIC. For full feedback or image support, please contact the relevant agencies or check through the official channels of the conference.