On July 20, the four-day 2026 World Artificial Intelligence Conference (WAIC) and High-Level Meeting on Global AI Governance concluded in Shanghai. Under the theme "Intelligent Partners, Co-creating the Future," the event's total exhibition area surpassed 100,000 square meters for the first time. Over 1,100 companies gathered to showcase 4,486 exhibits, including 349 global debut products. During the four-day event, more than 400,000 visitors attended in person, and online traffic exceeded 3 billion views. Intended procurement agreements reached 20.36 billion RMB.

President Xi Jinping attended the opening ceremony and delivered a keynote speech. Now in its ninth year, WAIC has reached what industry observers widely regard as a watershed moment in the history of AI development.

📊 WAIC 2026 By the Numbers

· Exhibition area: 100,000+ sqm for the first time
· Exhibiting companies: 1,100+
· Total exhibits: 4,486, including 349 global debuts
· Offline visitors: 400,000+
· Online traffic: 3+ billion views
· Intended procurement: 20.36 billion RMB, with 200+ signed projects
· Countries/regions represented: 38+
· Forum sessions: 58+, featuring 583+ leading figures

Large Models Step Back, Agents Step Forward

The clearest signal from WAIC 2026: the three-year-long model parameter arms race has formally given way to the commercialization of AI agents.

On the eve of the conference, Kimi released its 2.8-trillion-parameter K3 open-source model, causing a stir across the industry. Yet what drew even more attention was Moonshot AI's simultaneous emphasis on agent-scenario optimization — parameters are no longer an end in themselves, but rather the foundation that enables "agents to get things done." IDC predicted during the conference that nearly 50 million AI digital employees will be deployed globally in 2026.

On the exhibition floor, the "war of a hundred models" ceded center stage to a proliferation of AI agents in every corner. Ant Group's Agentar 2.0 platform features 200 built-in role templates spanning finance, legal, supply chain, and more — allowing enterprises to generate custom AI employees with a single click, no programming required. Tencent's WorkBuddy went fully live, demonstrating a 20%+ boost in team administrative efficiency in real-world tests. Alipay's merchant operations agent "Xiaoyu" can independently perform store analysis, marketing configuration, and other complex operational tasks. Agents from Kingsoft Office, iFlytek, and other vendors have already penetrated specific business workflows such as legal affairs and procurement.

"From 'being able to chat' to 'being able to work,' the AI industry is undergoing a generational leap. The essence of the agent shift is AI's transition from conversational capability to action capability."

The Year of Embodied Intelligence: Robots Stop Dancing and Start Working

Embodied intelligence was designated as a core track for the first time. 242 robotics and intelligent hardware companies exhibited, accounting for 23.6% of all exhibitors — an all-time high. A wave of emerging firms including Xingdong Epoch, Lingchu Intelligence, and Songyan Power made their WAIC debut.

But more striking than the numbers was the qualitative shift on the exhibition floor. In previous years, robot zones featured dancing, running, and backflips — audiences judged whether a robot "could move." This year, Zhiyuan brought full factory and logistics warehouse deployment scenarios to the venue, declaring 2026 "the first year of deployment." Its Jingling G2 Max robot, deployed in partnership with JD Logistics in a real-world warehousing scenario, marked the first time a humanoid robot has been deployed in an actual仓储 operation. Qinglang Intelligent built coffee kiosks and retail stores; Qiongche Intelligence demonstrated pharmacy medication retrieval. The focus shifted from individual robot capability demonstrations to concrete operational scenarios.

Cross-regional real-time robot control also became a reality. China Telecom's AI Research Institute (TeleAI) demonstrated a远程操控 system allowing operators in distant locations to control on-site robots for movement, observation, grasping, and transport — capable of supporting industrial inspection, emergency response, and public service scenarios.

Computing Competition Rewired: From Single Chips to Supernode Clusters

Large models may no longer compete on parameters, but computing power has not become less important — the competition has simply been elevated to a new dimension.

At WAIC 2026, Huawei's Atlas 950 supernode made its true-machine debut, capable of搭载ing 1,024 AI chips with a total computing power of 10–20 exaflops. Sugon showcased its Sugon 8000 (Dengfeng) 100,000-card AI supercluster. ZTE's OEX supernode appeared publicly for the first time, featuring an orthogonal backplane-free design that實現s zero-cable interconnection and reduces interconnection costs by 80%.

The concentrated emergence of supernode solutions signals that domestic chipmakers are moving beyond the traditional path of single-process-node追赶, adopting architectural innovation as the core breakthrough and amplifying the role of system-level interconnection within clusters to compensate for individual chip gaps. AI computing is evolving from "single chips" to a systematic "server–supernode–cluster" architecture, with the competitive dimension expanding from computational performance to high-speed interconnect, memory, cooling, software ecosystems, and overall operations.

Global AI Governance Milestone: 29 Nations Found the World AI Cooperation Organization

This year's conference also achieved a historic breakthrough in governance. The World AI Cooperation Organization (WAICO) was officially established in Shanghai — the first intergovernmental organization in the global AI field — with 29 countries as founding members.

China announced a series of inclusive initiatives: 5,000 AI training slots for developing countries over the next five years; the establishment of international AI application cooperation centers targeting ASEAN, the Arab League, the African Union, CELAC, the SCO, and BRICS nations; and the rollout of the "Mazu" meteorological AI solution across 30 countries. As President Xi Jinping stated in his opening address: "The more rapidly AI technology develops, the more firmly we must anchor its direction toward doing good and benefiting humanity."

Conclusion: The Critical Pivot from "Technology Competition" to "Application Deployment"

If the previous two WAIC editions were a "model arms race" where LLM vendors vied to publish parameter metrics and top leaderboards, WAIC 2026 clearly宣告s: the AI industry is shifting from "technology-driven" to "application-driven." Models have not disappeared, nor have they become unimportant — but like electricity, chips, and operating systems, they are being embedded into larger technological architectures.

This year's conference signed over 200 projects with a total value exceeding 50 billion RMB, and 57 core scenarios were deployed. From parameter competition to deployment competition, from intelligence behind a screen to intelligence in the physical world, this "young industry" is coming of age. As the conference theme expresses — Intelligent Partners, Co-creating the Future.