The day before Xi's speech, Moonshot AI unveiled Kimi K3, the largest open-weight model ever released. Independent benchmarks place it immediately behind the leading American systems, while, for the first time, a Chinese model topped a major international coding benchmark.

The significance is not that China has overtaken the United States; it has not. Moonshot itself acknowledges that user experience still trails the leading American models. Rather, the technological gap has narrowed sufficiently that frontier performance alone no longer guarantees commercial dominance.

Nor is this another low-cost strategy. Kimi K3 is priced close to proprietary American models, and its sheer scale means few organisations will ever run it independently. Even open-weight AI increasingly concentrates economic value in whoever controls inference infrastructure.

Yet the strategic direction is unmistakable. Backed by patient capital and state support, Chinese laboratories are steadily commoditising the foundation-model layer on which much of the American AI business model depends. 

On the eve of Xi's speech, 29 countries, including Russia, Indonesia, Kazakhstan, Pakistan and Laos, signed the agreement establishing the World Artificial Intelligence Cooperation Organization (WAICO), headquartered in Shanghai.

Xi simultaneously announced 5,000 AI training opportunities for developing countries, cooperation centres spanning ASEAN, the Arab League, the African Union, CELAC, the Shanghai Cooperation Organization and BRICS, together with international access to China's AI-powered meteorological platform, MAZU.

This is, in effect, a digital Belt and Road Initiative. Instead of financing ports, railways and power plants, China is financing AI capabilities, developer ecosystems and, perhaps most importantly, the technical standards that will govern AI deployment. Technology leadership increasingly depends not on who builds the best model, but on who defines the standards, attracts developers and becomes the default platform.

The strategy is already generating commercial traction. Chinese frontier models are no longer being evaluated solely by researchers. They are increasingly being deployed by multinational companies. Firms including DoorDash, Siemens and Airbnb have adopted Chinese models because they are increasingly capable, cost-effective and can be deployed within their own infrastructure. This marks an important shift: Chinese AI is no longer simply catching up technologically, it is beginning to establish commercial relevance internationally. For investors, adoption ultimately matters far more than benchmark rankings. 

Xi repeatedly described AI as the new engine of economic growth, extending far beyond chatbots into manufacturing, robotics and industrial automation. Huawei's unveiling of its Atlas 950 SuperPoD during the conference reinforced that ambition.

For investors, this broadens the AI opportunity well beyond semiconductors. Industrial automation, networking, electrification, power infrastructure, cooling technologies and enterprise software increasingly appear positioned to become the next beneficiaries of AI deployment.

Washington, meanwhile, may inadvertently be accelerating the very transition it seeks to prevent. In June, the US Commerce Department ordered Anthropic to bar any foreign national, inside or outside the United States, from using its most advanced models. Unable to verify nationality in real time, the company withdrew them entirely. This was the first attempt to apply export controls to frontier AI software rather than semiconductors. Although the restrictions were subsequently lifted, the episode sent an important signal: access to leading American AI models can become a geopolitical variable overnight.

Within days, Chinese laboratories responded by promoting open-weight alternatives with unrestricted international access. When access to one ecosystem becomes uncertain, governments and enterprises naturally diversify. The hedge increasingly looks Chinese.

The AI investment case is broadening, and exposure needs to broaden with it. Exposure to US AI does not need to be reduced, but it does need to become more selective. The business models facing the greatest pressure are those whose economics rely on the pricing power of proprietary foundation models. While OpenAI and Anthropic remain private, their economics flow through to listed partners such as Microsoft, via its investment in OpenAI, and infrastructure providers including Oracle and CoreWeave, whose growth assumptions depend on continued demand for proprietary models.

By contrast, companies that enable AI deployment should remain beneficiaries regardless of which ecosystem ultimately prevails. That includes cloud platforms such as Amazon, Microsoft and Alphabet; AI infrastructure leaders including NVIDIA, Broadcom, TSMC and memory manufacturers; power and cooling specialists such as Schneider Electric and Vertiv; and utilities expected to benefit from structurally higher electricity demand driven by AI.

At the application layer, cheaper and increasingly capable models are a tailwind rather than a threat, as Airbnb, Siemens and DoorDash are already demonstrating. The listed expression of China's AI ecosystem is, above all, Alibaba, owner of the Qwen family, the world's most widely adopted Chinese open models, and a substantial investor of Moonshot AI, whose valuation reached $20bn in May. Tencent, a repeat investor in Moonshot, offers secondary exposure.[

The AI investment theme is fragmenting, and returns will follow the fragments. In the next phase of the cycle, security selection is likely to matter far more than broad thematic exposure. None of this suggests that China has overtaken the United States. NVIDIA, TSMC, the hyperscalers and America's frontier laboratories remain at the centre of AI innovation. Nor should Beijing's embrace of open-weight AI be mistaken for philosophical openness. It is a calculated instrument of industrial policy and geopolitical strategy.

The most important question for investors is therefore no longer whether China will build a model as capable as America's best. It is whether developers, enterprises and governments increasingly choose to build on China's ecosystem. Technology leadership can be temporary. Ecosystem leadership is far more durable.