WAIC 2026: When AI Moves from Chat to Work

On July 17, the 2026 World Artificial Intelligence Conference (WAIC) opened at the Shanghai Expo Center to the synchronized movement of robotic arms. President Xi Jinping attended the opening ceremony and delivered a keynote address — the highest-level endorsement in the conference's nine-year history — signaling that this year's WAIC is far more than an industry expo.

The theme is "AI Partnership for a Brighter Future." But walking through the 100,000 square meters of exhibition space across three venues — World Expo, Zhangjiang, and West Bund — one takeaway dominates: AI is moving from chat to work.

Embodied Intelligence: From Slide Decks to Factory Floors

Embodied intelligence is the single largest thematic track at WAIC 2026, with over 200 companies exhibiting across all three venues. Last year, the prevailing question was "When will humanoid robots enter factories?" This year, the answer is "They're already there."

Huawei unveiled its most ambitious AI computing system yet: the Atlas 950 SuperCluster. Each cabinet houses 64 Ascend 950DT compute cards, scaling up to 8,192 cards. Against the backdrop of ongoing export controls, this fully domestic AI infrastructure stack drew some of the largest crowds at the expo.

Agibot (Zhiyuan) celebrated its 15,000th humanoid robot rolling off the production line, while debuting GE-Sim 2.0, a world-simulation model designed to train next-generation embodied foundation models. Inside a 200-square-meter experience zone, the Expedition A3 Ultra, Sprite G2 Max, and Omnihand 3 Ultra-M took turns demonstrating precision assembly tasks. Just days before WAIC, Unitree Robotics received IPO approval for the STAR Market at an estimated valuation of 42 billion yuan — making it the first pure-play humanoid robot stock on China's A-share market.

Tashit AI went a step further, bringing an actual automotive assembly line into the exhibition hall. Its A1 robot demonstrated wire pulling, routing, and precision insertion, powered by the AWE 3.5 foundation model. Nearly 100 units are already deployed at customer sites including Aptiv and Tianhai Electronics.

Beyond Hardware: LLMs and Chips

On the software side, MiniMax's M3 multimodal large model and StepFun's StepAOS agent operating system both made their global debuts. StepFun also announced STEPX, the world's first native LLM terminal brand, just before the conference — a clear signal of vertical integration ambitions from model to device.

In the chip space, Shanghai Eastern Computing Technology unveiled the DF1000, the world's first software-defined near-memory computing 3D AI chip. Near-memory computing challenges the traditional von Neumann bottleneck by placing compute units as close to data as possible, dramatically reducing the energy cost of data movement. In an era of constrained computing resources, architectural breakthroughs like this may matter more than simply scaling parameters.

Shanghai: A City's AI Ambitions

Behind WAIC lies Shanghai's systematic push to become an "AI capital." The numbers speak for themselves:

• In 2025, Shanghai's 394 large-scale AI enterprises generated 637 billion yuan in revenue, up 39.5% year-over-year
• From 134 billion yuan in 2018, the industry has nearly quintupled in seven years
• Smart computing capacity surpassed 160,000 petaflops, powering AI adoption across industries
• AI talent pool approaches 300,000, roughly one-third of China's total
• 169 large models registered, with open-source communities hosting over 150,000 models

The "SMC Shanghai Foundation Model Innovation Center" in Xuhui West Bund and the "Model Power Community" in Pudong Zhangjiang form an East-West industry cluster — spanning chips, cloud, and devices. Between late 2025 and early 2026, five AI supply chain companies — including MuXi, Biren, Tianshu Zhixin, and MiniMax — went public in rapid succession. Capital markets are voting with their wallets.

Global Governance: From Attendee to Agenda-Setter

The conference's other keyword is governance. WAIC was elevated this year to a "High-Level Meeting on Global AI Governance." Foreign Ministry AI Coordinator Sun Xiaobo outlined three expectations at the pre-conference press briefing: promoting unity and multilateralism in AI, advancing practical cooperation, and implementing the Global AI Governance Action Plan proposed at WAIC 2025.

This marks China's shift from "participant" to "agenda-setter" in international AI rulemaking. Amid intensifying great-power tech competition, WAIC is positioning itself as the third pillar of global AI governance discourse, alongside the UK AI Safety Summit and the Korea AI Summit.

The inaugural WAIC Academic conference, chaired by Turing Award winner Andrew Yao with reinforcement learning pioneer Richard Sutton as international co-chair, received 284 submissions from 11 countries. From industry to academia to governance, WAIC is building a complete three-tier architecture.

Taking AI Beyond the Exhibition Hall

This year's WAIC deliberately softened the insider-only vibe. The City Walk program connects 24 AI experience sites across all 16 Shanghai districts through six themed routes covering AI innovation, science education, urban renewal, and consumer entertainment. Evening programs include youth AI music shows, product beta-testing sessions, and a creators' night featuring directors, sci-fi writers, and content platform representatives.

It's a telling signal: AI is no longer just an industry talking to itself. The organizers are trying to turn the entire city into "a museum without walls."

From chat to work, from exhibition booths to city streets, from technology to governance — in its ninth year, WAIC's central question is no longer "What can AI do?" but rather "How should AI be used, who governs it, and who benefits?" That question is far harder to answer than training the next large model.