The 2026 World Artificial Intelligence Conference and High-Level Meeting on Global AI Governance opens July 17 in Shanghai and runs through July 20. For the first time in the conference’s history, Chinese President Xi Jinping will deliver the opening keynote. That is not a protocol detail — it signals that Beijing is moving AI governance from bureaucratic process to top leadership priority, on a global stage.
More than 1,100 enterprises will participate, over 3,000 exhibits go on display, and more than 300 AI products will make their global debut across more than 100,000 square meters of exhibition space — a record for the event.
This is a research-based preview. Facts are drawn from Global Times, South China Morning Post, CGTN, and George Chen’s reporting on Substack. We will cover the actual announcements after the conference concludes.
What Xi Jinping Is Expected to Say
Pre-conference reporting, citing official statements from Beijing, said Xi’s address would systematically lay out China’s policies, positions, vision, and proposals on AI development and governance. China’s Ministry of Foreign Affairs subsequently confirmed Xi delivered the keynote on July 17. Based on pre-conference reporting, two specific initiatives were expected:
1. “China’s AI Empowering the World Case Studies, Shared Benefits for the World” — a compilation of cooperative AI projects spanning more than 20 countries, covering healthcare, agriculture, manufacturing, education, disaster response, and digital public services. This is China positioning its domestic AI deployments as reference implementations for developing nations, particularly in the Global South.
2. “Action Plan for AI Cooperation and Development” — a framework spanning eight areas (including policy dialogue, technology cooperation, and infrastructure development) promoting inclusive access to computing power and shared open-source ecosystems. The framing is explicitly multilateral and non-exclusionary, contrasting with US export controls on advanced chips that limit which countries can access cutting-edge compute.
The context matters. Days before WAIC, the UN’s Global Dialogue on AI Governance convened in Geneva on July 6–7, where China pursued direct multilateral diplomacy under an “AI for Good and AI for All” framing while the US sent lower-level representation focused on market-based cooperation rhetoric. Xi’s keynote turns WAIC from a technology showcase into a diplomatic instrument: China is offering a governance framework that developing nations can adopt as an alternative to the US-led approach.
Hardware: Huawei Atlas 950 Debuts
The Huawei Atlas 950 SuperPoD made its public debut at WAIC — Huawei’s own billing, echoed by Digitimes and Huawei Central, calls it the industry’s largest-scale AI computing supernode: a cluster of up to 8,192 Ascend NPUs delivering 1 EFLOPS (FP8) / 2 EFLOPS (FP4) of compute and 256TB of unified memory.
Why this matters for builders: DeepSeek pricing. DeepSeek indicated that pricing for its V4-Pro model could fall meaningfully in H2 2026 once Huawei’s Ascend 950 chips ship at scale — Huawei is targeting roughly 750,000 units of the Ascend 950PR chip shipped in 2026, with mass production starting in April and full-scale shipments in H2 2026. Builders currently evaluating Chinese open-weight models as cost-efficient alternatives to GPT-5.6 Sol or Opus 4.8 should track Atlas 950’s production timeline as a price trigger.
This also matters for AI sovereignty watchers: Huawei’s Ascend chips are the domestic alternative to Nvidia hardware, and their capability trajectory directly affects whether China can continue frontier AI development under US export controls.
Product Launches Worth Tracking
With 300+ global debuts, most will be noise. Notable confirmed launches:
- MiniMax M3 — a multimodal large model from MiniMax, the Chinese lab whose earlier MiniMax-01 (456B total parameters, 4M-token context) set a long-context record in early 2025. M3 itself is a 428B-parameter (23B active) model with native text/image/video multimodality and a 1M-token context window.
- ZTE Nubia AI Agent Phone — one of multiple competing claims to “world’s first AI agent smartphone” at WAIC (StepFun’s rival device makes the same claim); notable because Doubao just disabled custom persona features to comply with China’s new AI companion law (effective July 15) — see below.
- Humanoid robots and AI dexterous hands — WAIC will likely surface the current state of China’s humanoid robotics push.
- Huawei Ascend chip solutions — beyond Atlas 950, Huawei is expected to announce new products across its Ascend AI chip line.
- Jieyue Agent Operating System — StepFun’s Step AOS, an intent- and task-driven agentic layer rather than a traditional file/app-centric OS.
The Governance Divergence Builders Should Understand
China’s AI regulatory trajectory in mid-2026:
- July 15: AI companion rules take effect — ByteDance disables Doubao custom personas, Alibaba disables Qwen agent features (see our China AI companion compliance guide)
- April 10, 2026: “Interim Measures for the Administration of AI Anthropomorphic Interaction Services” issued by the Cyberspace Administration of China and four other ministries — China’s first dedicated regulation targeting AI companion/emotional-interaction services, effective July 15, 2026
- WAIC July 17-20: Xi proposes a multilateral AI governance framework to replace or parallel the US/EU approach
The pattern is a governance system that is simultaneously strict on domestic emotional/social AI (protecting users from parasocial dependencies) and expansionist on technical AI deployment (promoting infrastructure access globally). If you build AI products that touch Chinese users or rely on Chinese AI providers, these two tracks move in opposite directions for different parts of your stack.
What This Means if You Build Global AI Products
If you use Chinese AI models — DeepSeek, Qwen, Doubao, MiniMax — track WAIC announcements for new API capabilities, pricing changes, and regulatory constraints that may affect what those models can and cannot do.
If you ship to Chinese users — the AI companion law is active now. Review whether any of your product’s agent features (custom personas, sustained emotional interaction, long-memory conversational agents) fall under the new rules. See our companion compliance guide for the technical definition.
If you compete with Chinese AI products globally — Xi’s governance pitch targets markets where Chinese AI infrastructure may be adopted as the default. In those markets, Huawei chips, DeepSeek models, and Doubao-derived products may be the dominant stack. Building for infrastructure portability matters more than it did in 2025.
If you are watching the US-China AI divide — the WAIC keynote will likely produce governance proposals that the US and EU will need to respond to. This has direct implications for export control policy, chip access, and which models are available on which platforms. The governance gap is widening; builders who depend on access to both markets need contingency plans on both sides.
What to Watch Starting July 17
- Xi’s keynote text, specifically the governance proposals and how they frame China’s AI development as inclusive versus exclusionary
- Whether any specific new regulations are announced targeting global AI governance (data sovereignty, cross-border transfer rules, compute access requirements)
- Huawei Atlas 950 production timeline — this is the trigger for DeepSeek price drops
- New model releases from Chinese labs that may surface at WAIC before official API availability
- Any signals about China’s position on MCP, the open agent protocol, or cross-border AI interoperability
We will publish a post-WAIC builder digest when the conference concludes on July 20.