AI diplomacy is really just industrial policy carrying a diplomatic passport. Cut off from the best U.S. chips, China has built an alternative stack and now needs customers for it.
Xi Jinping has called for future AI development to be a “symphony of global collaboration” and he brought 29 nations together to sign an agreement establishing the World Artificial Intelligence Cooperation Organization (WAICO), to be headquartered in Shanghai. UN Secretary-General António Guterres attended the signing, lending the exercise a multilateral gloss and seal of approval Beijing could not manufacture on its own. The declared purpose is to close the AI divide and prevent “new historical injustices”—but as we know, everything Beijing does is carefully considered, calibrated, and executed. At a time when China and the US are slugging it out in cyberspace to reign supreme in AI, it seems, shall we say, disingenuous that China would be offering to level the playing field at this time with the WAICO initiative. And yet, the idea is urgently needed. The key questions are, why now, will the US and other nations participate, will the outcome be meaningful, or will WAICO become another showpiece multilateral organization that China has dreamed up for its own ultimate net benefit (as it did with the Asian Infrastructure Investment Bank (AIIB), New Development Bank, and Shanghai Cooperation Organization)? Beijing has learned that the cheapest way to win a standards fight is to hold it somewhere else. The EU AI Act, G7 Hiroshima process and OECD principles were drafted without meaningful Global South authorship, and the resulting resentment is real. WAICO answers it by creating a parallel forum in which China controls the secretariat, headquarters, budget, and agenda. What a surprise. This is the AIIB playbook applied to rulemaking rather than lending, and it works for the same reason: the incumbents left the field open.
“AI diplomacy” is really just industrial policy carrying a diplomatic passport. Cut off from the best U.S. chips, China has built an alternative stack and now needs customers for it. Xi’s World AI Conference doubled as a trade show: Huawei, DeepSeek, and Moonshot’s Kimi all displayed their latest offerings. Open weights are not charity: the marginal cost of distribution is near zero, and the lock-in accrues one layer down, in chips, cloud, tooling, standards and the engineers trained on them. Xi’s call to oppose “overstretching the concept of national security” in AI is a direct assault on the U.S. control regime, and it lands harder because Washington has made it easy. The June suspension and July restoration of access to Anthropic’s most capable models under Commerce Department controls is precisely the kind of episode Beijing cites to argue that American supply is politically contingent and Chinese supply is not.
The governance philosophy on offer is sovereigntyfirst: states, not citizens, hold the relevant rights, and content control is a legitimate exercise of digital sovereignty. It is also the operating system of digital authoritarianism, and adopting it wholesale is a domestic political choice dressed as a procurement decision. Why should developing countries be sceptical? Not because the grievance is fake—compute is concentrated in a handful of jurisdictions, and the risk of a permanent AI underclass is genuine. The Belt and Road Initiative taught hard lessons about renegotiating a port. A national compute layer, a government services platform and a model stack are stickier than a port, and there is no equivalent of a debt restructuring for a country whose administrative apparatus runs on someone else’s inference. WAICO arrives with no independent redress mechanism—no inspection panel, no ombudsman, no channel through which an affected community can contest a project standardized under its auspices. An intergovernmental body headquartered in, staffed by and funded by a single dominant member, with no external accountability architecture, is not multilateralism. It is a franchise with a flag. The founding membership is drawn substantially from states with acute domestic surveillance requirements, revealing the underlying purpose. If WAICO is ultimately to resemble its stated purpose, the signed agreement should have required such things as a multi-sourced stack, open weights, hardware documentation, exit clauses, data terms prior to deployment, and independent audit capacity. Why not play two suppliers against each other instead of trading one dependency for another? The agreement achieved none of this. The test of Beijing’s real intent is simple enough. Will it accept binding constraints it does not control? Of course, the answer is no. Until it is answered in the affirmative, “AI diplomacy” is best read as little more than market access, standards leverage, and coalition-building offered on generous rhetorical terms with highly dubious likely material outcomes. The corollary should be equally uncomfortable for Washington and Brussels. After all, the need for a WAICOlike entity exists because they all created it.
* Daniel Wagner is managing director of Multilateral Accountability Associates and author of the forthcoming book The AI Insurgency: The Future of Thought, Identity, and Power, to be published in September.
