Xi Jinping personally opened Shanghai's World AI Conference to launch a rival AI governance body for the Global South, at the moment China's models have closed to within 2.7 points of America's best

For eight years, the World AI Conference in Shanghai was a trade show. Premier Li Qiang gave the opening speeches. Products launched. The Western press mostly checked the specs and moved on.

This year, Xi Jinping took the stage himself.

The first Chinese president to personally open WAIC, Xi used a conference drawing more than 1,100 exhibiting companies 1 to do something no previous edition attempted: pitch the Global South on a Chinese-led alternative to Western AI governance.

The vehicle is the World AI Cooperation Organization, or WAICO, a body China first proposed last year and now wants headquartered in Shanghai. It follows the institutional template of the Shanghai Cooperation Organization, a multilateral body Beijing co-founded and shaped, repurposed for technology governance rather than security 1. The pitch to developing nations is specific: open-weight AI models, cheaper inference costs, and a formal governance role in writing AI rules 1. That combination does not exist in the US "trusted partners" framework or the EU AI Act.

No country has formally announced WAICO membership yet 1. But UN Secretary-General Antonio Guterres attended the opening ceremony alongside Kazakh President Kassym-Jomart Tokayev and Thai Prime Minister Anutin Charnvirakul 1. Having heads of state on the stage is not incidental. It signals that Beijing is making this pitch at the diplomatic level, not the industry one.

Xi told the conference that the world has "entered an unprecedentedly vigorous period of innovation" in AI that brings "great opportunities and governance challenges" 2. He framed the stakes as questions the international community must answer: how humans get along with thinking machines, how to ensure security when algorithms make decisions, how to close a widening technology divide 2.

The timing is not accidental. Stanford's 2026 AI Index puts the performance gap between the best US and Chinese AI models at 2.7 percentage points, down from double digits in previous years 1. At under three, it is competitive.

Huawei made that competitive claim concrete on the exhibition floor. The company publicly debuted the Atlas 950 SuperPoD, a computing cluster of 8,192 Ascend processors delivering 8 exaflops of FP8 compute. What separates the system from a conventional cluster is its UnifiedBus 2.0 interconnect: an all-optical memory fabric that lets all 8,192 chips share a unified memory pool and operate as a single logical machine rather than a room of independent servers 1. The entire software stack is domestically developed and open-sourced 1.

The subtext is export controls. Huawei's access to leading-edge Nvidia GPUs was cut off by US restrictions starting in 2020. The company built around the constraint by compensating for any per-chip performance gap through scale: linking vastly more processors into one machine via a fabric it controls end to end 1. The Atlas 950 is evidence that China can build frontier AI infrastructure without American components.

There is a real limitation. Nvidia's CUDA ecosystem has more than 15 years of development behind it. Huawei's CANN software stack trails in library coverage 1. The hardware gap is closing. The software gap is not.

George Chen, chair of digital practice at the Asia Group, described what has shifted: "WAIC has become more than a technology showcase; it is now a geopolitical stage where Beijing seeks to articulate its vision of AI as both a national priority and a diplomatic instrument" 1.

For builders, the practical consequence is divergence. Companies deploying AI in both Western and Global South markets will increasingly face two rulebooks. The OECD's AI Principles, the EU AI Act, and the G7 Hiroshima Process govern one set of countries. WAICO, if it gains members, governs another 1. Compliance teams will have to design for both.

For investors, the signal is a bifurcating stack. Companies building on US-controlled chips, CUDA software, and Western governance frameworks occupy one half. Companies building on Ascend processors, CANN software, and WAICO terms occupy the other. The Atlas 950 makes the second half investable in a way it was not 12 months ago.

Xi's keynote in Shanghai was not a product launch. It was a declaration that the contest for AI standards has moved past benchmarks. The question was never just who builds the best models. It is who writes the rules for the two-thirds of the world that existing Western frameworks left out. China is now making that case with hardware to back it.