WAIC 2026 Opens in Shanghai on July 17 — Atlas 950 SuperNode and a Token Economy That Grew 1,000x in Two Years

Shanghai will host the 2026 World Artificial Intelligence Conference, or WAIC, from July 17 to 20, organizers confirmed this week. The show will span more than 100,000 square meters for the first time, bring together over 1,100 companies and showcase more than 300 global product debuts, ranging from a flagship homegrown supernode to the world's first AI agent phone.

The hardware headline is Huawei's Atlas 950 SuperNode, the largest AI compute cluster the firm has ever shown in public. It uses a homegrown interconnect protocol called Lingqu, links up to 8,192 NPU cards across cabinets, and reaches a total bandwidth of 16.3 PB per second, a number that places it ahead of several widely deployed overseas designs. Alongside it, MiniMax will debut its M3 multimodal model, StepFun will show a full agent operating system, and several humanoid robots and dexterous hands will walk the floor.

The bigger story, however, sits one layer below the products. According to figures released by China's National Data Bureau, daily token calls inside the country rose from roughly 100 billion in early 2024 to about 100 trillion by the end of 2025, and crossed 140 trillion in March 2026. That is a thousand-fold jump in two years, driven by agents, coding copilots and enterprise search rather than chatbots. Industry forecasts now put China's AI compute economy above 30 trillion yuan by 2030.

That growth has exposed a new bottleneck. At a conference on July 8, a Huawei vice president said token circulation efficiency across Chinese networks still sits around 60 percent, with the remaining 40 percent lost to packet drops, latency jitter and data corruption. His target is to lift that figure above 90 percent before agent traffic scales further. For learners, the takeaway is simple: the next leg of the AI race is no longer about who trains the biggest model, but about who can move tokens cheaply, reliably and at continental scale.