WAIC 2026 Opens: 300 AI Debuts, the World's First AI Agent Phone, and China's Boldest Claim on AI Leadership

On the same day Google launched Gemini 3.5 Pro, China opened its most consequential technology showcase in years. The 2026 World Artificial Intelligence Conference kicked off in Shanghai on July 17, filling more than 100,000 square meters of exhibition space across four days with over 1,100 companies, 3,000 exhibits, and more than 300 products making their global debuts — a number that by itself exceeds the total product announcements from any single Western AI event this year.

Xi Jinping’s in-person keynote — his first at WAIC since the conference began in 2018 — underscored the degree to which Beijing now treats AI leadership not as a technology policy goal but as a core national priority equivalent to semiconductor independence and space access. The products that debuted on day one put concrete hardware and model releases behind that political framing.

The World’s First AI Agent Smartphone — Twice

The headline consumer product at WAIC 2026 is the concept of the AI agent smartphone, and China has decided to ship two of them simultaneously. ZTE’s Nubia subsidiary debuted the NaviX Ultra under the claim of being the world’s first AI agent smartphone. StepFun — the AI lab backed by Zhang Yiming, ByteDance’s founder — made an identical claim for its own device.

The distinction between an AI agent phone and a conventional AI-enhanced smartphone is architecturally significant. Current AI phones primarily run language model inference on-device to answer questions or generate text. An AI agent phone integrates a GUI agent — a system that can visually parse the device’s screen interface and execute multi-step tasks across applications without explicit per-step user instructions.

The Nubia NaviX Ultra runs ByteDance’s Doubao AI assistant at the operating system level. In demonstrations, the device was shown booking restaurants, drafting and sending messages across multiple apps, setting up reminders based on conversational context, and executing sequences of actions that would normally require three to five separate app interactions. Ni Fei, president of Nubia Technology, positioned it as a mass-market flagship — not a developer preview — targeting the same consumer pricing band as a flagship Android device.

The StepFun device operates on a similar GUI agent architecture with StepFun’s own foundation models, positioning it in competition with Nubia for the same “world’s first” designation. Both companies appear to have targeted WAIC as the symbolic stage for a product category that, if it lives up to demonstration capabilities, could represent the most significant shift in smartphone interaction since the introduction of touch interfaces.

Huawei’s Atlas 950: AI Infrastructure at National Scale

Huawei’s contribution to WAIC 2026 was positioned not at the consumer end but at the infrastructure foundation. The Atlas 950 Super Node is an AI data center computing platform described as the industry’s largest-scale super node — a claim that, given U.S. export restrictions on NVIDIA’s most advanced GPUs, Huawei is free to make without a direct comparison partner from American chipmakers.

The Atlas 950 is significant because it represents Huawei’s continued demonstration that China’s domestic AI chip ecosystem can support frontier AI training and inference at scale without access to NVIDIA’s H100 or H200 accelerators. The restrictions that followed the 2022 BIS rule updates, and their subsequent tightening, created a structural incentive for Chinese companies to develop indigenous AI hardware — and WAIC 2026 is the clearest showcase yet of what that investment has produced.

For international observers, the Atlas 950 matters less as a direct competitive threat to American data center hardware and more as a data point about the durability of U.S. export controls as a technology containment strategy. If China can train and deploy frontier models on domestic hardware at commercial scale, the strategic assumption underlying export restriction policy — that denying access to advanced chips would slow China’s AI progress — requires reassessment.

MiniMax M3: The Multimodal Push

Among the model releases at WAIC 2026, MiniMax’s M3 attracted the most developer attention. MiniMax — a Shanghai-based AI lab that has grown rapidly into one of China’s best-funded model companies — debuted M3 as a multimodal foundation model with a 1-million-token context window, targeting enterprise customers in legal, finance, and content generation verticals.

M3’s technical positioning emphasizes long-form reasoning, coding performance, and multimodal input processing including image and video understanding. The 1M-token context window is notable: it puts M3 in the same tier as Gemini 2.5 Pro (though below today’s Gemini 3.5 Pro launch) while remaining accessible via MiniMax’s own cloud infrastructure, free from the U.S. export control questions that complicate use of some Western frontier APIs in China.

MiniMax’s growth trajectory — the company was valued at over $5 billion in its most recent funding round — reflects broader investor conviction that China’s domestic AI market is large enough to support multiple frontier model companies competing independently of Western equivalents.

The Geopolitical Stage

WAIC 2026’s scale — 100,000 square meters, 1,100 exhibitors, 300+ product debuts, 9 Turing Award laureates delivering academic presentations — is partly about technology and partly about narrative. China is staging the largest AI showcase in the world on the same week the Western frontier releases three competing flagship models, and doing so with products that span consumer devices, enterprise infrastructure, foundational models, and governance frameworks.

Xi’s keynote, delivered to an audience of government officials and industry leaders from dozens of countries, emphasized China’s commitment to open global AI governance frameworks and positioned Beijing as an alternative to what it characterizes as the United States’ unilateral approach to AI standards-setting. The UN’s High-Level Meeting on Global AI Governance is running concurrently with WAIC in Shanghai, giving China a diplomatic venue to advance its governance proposals alongside its product showcases.

The through-line of the 2026 AI race, as it has emerged over the first half of the year, is not a Western monopoly on frontier capability facing an aspiring challenger. It is a two-superpower competition in which domestic Chinese AI infrastructure, models, and products are increasingly credible at the frontier level — and in which the governance and standards layer is as contested as the technical one.

108 Chips and 261 Large Models

A statistic cited in WAIC’s official preview that deserves separate attention: the conference will feature 108 chip products and 261 large language models in its exhibits. These numbers are not all frontier-class — many are purpose-built or domain-specialized rather than general-purpose frontier models. But the sheer volume is a meaningful indicator of the depth of China’s AI development ecosystem.

The chip count is particularly significant given export restrictions. Exhibiting 108 chips — spanning training accelerators, inference chips, edge AI processors, and memory architectures — suggests that the domestic Chinese semiconductor industry has diversified and deepened its AI-specific offerings substantially since the original export controls were imposed. Whether any of these chips can match NVIDIA’s most advanced offerings in raw performance is a separate question from whether they can collectively support a self-sufficient national AI compute ecosystem.

WAIC 2026 runs through July 20. Major product demonstrations, academic sessions, and governance meetings are scheduled across the full four days, with more debuts expected in the coming sessions.