The real purpose of Waico, China’s AI club

THE announcement of an American-Chinese artificial intelligence dialogue on Sep 26 suggests that the two nations are making a pragmatic effort to erect guardrails around an accelerating tech war.

It follows China’s launch of the World Artificial Intelligence Cooperation Organization (Waico) in July. The following day, Xi Jinping, China’s president, opened the World AI Conference calling for a “just and equitable” system of global AI governance and warned against “overstretching the national security concept” in the field – a clear rebuke of American policy.

The temptation in the West is to dismiss the founding of Waico as theatre. Although 29 countries have signed the agreement, no G7 country joined, nor did any nation from the EU.

It has no real budget, no enforcement powers and, most damningly, the countries that have joined don’t have AI chips. Its charter pledges “extensive consultation, joint contribution and shared benefits”, a generic wording familiar from China’s Belt and Road Initiative.

On these terms, Waico falls short as an institution. None of its qualities suggest that Waico can act as a global regulator.

But it was never meant to. Beijing’s appeal in AI is that Chinese alternatives to leading US models don’t come with the security checks and balances of their American counterparts, and that they are less expensive for most commercial tasks.

Waico is the diplomatic wrapper around that appeal and China’s attempt to create a structural advantage.

More than the sum of its parts

Three features make China’s AI diplomacy more formidable than the list of Waico’s founding members suggests.

Start with the grievance the organisation aims to answer. Over the past year, most governments, including rich ones, have discovered that Washington is the arbiter of access to this age-defining technology.

In June, the American government barred Anthropic, a leading AI lab, from serving Fable 5, its most capable model, to foreigners.

The top-of-the-head response – pursuing sovereign AI – is unaffordable. The 150 or so countries that will never build a frontier AI model must choose who to depend more on: America or China. To the undecided, Waico offers one option: capacity building and development first, with no lectures about values.

The second strength is that Beijing has impressive products. Chinese makers of open-weight models, including DeepSeek, Alibaba’s Qwen, Moonshot and Z.ai, now trail the American frontier by months rather than years, and undercut it, for some tasks, by as much as 60 to 90 per cent.

Xi’s accompanying pledges, including cooperation centres with Asean, the African Union, the Arab League, Brics and the Shanghai Cooperation Organisation, are aimed squarely at capturing those markets.

The third element may prove the most consequential: AI standards. If Chinese companies get to implement a technology first, their engineering choices are likely to become the local default.

Consider other fields, such as autonomous vehicles. Chinese operators have been able to enter the robotaxi market with few rules for robotaxis, in effect giving the country a chance to write them.

The same logic applies to AI, including the documentation of models, their evaluation and procurement, as well as things such as developer tooling. Once such standards are set, they are hard to unwind.

Through Waico, China is making itself the AI standard bearer for the rest of the world.

Competing where it counts

China’s AI diplomacy has shortcomings. The most significant weakness is that Waico cannot, at least in the short term, supply advanced chips, the essential commodity in AI.

Huawei, a Chinese tech giant, makes Ascend chips that have advanced impressively under Western sanctions, but still lag behind those of Nvidia by years.

Chinese AI labs are themselves rationed. ByteDance, maker of TikTok, reportedly suffers bottlenecks in chips; Z.ai meters its services; and Moonshot’s latest model has a waiting list. And so adopting more sophisticated Chinese AI models will often mean running them on American silicon.

Washington’s instinct is containment, including blacklists for Chinese labs, warnings of export controls and threats of sanctions over the distillation of models.

Despite this, Chinese models are spreading rapidly because they are often cheaper and because America’s own access policy is erratic. The three weeks in which Fable 5 was withheld did more for Chinese adoption than any Shanghai communique.

The better response is to compete where China is competing. America should publish capable open-weight models rather than ceding the layer entirely. (Some of this is happening: Nvidia has started an initiative for American open-weight models.)

It should also make access to proprietary models predictable and rule-bound instead of discretionary, since unpredictability is now America’s costliest export.

The West could also finance compute in poor countries and treat the developing world’s demand for a role in decision-making seriously.

Unlike the US-China AI dialogue, Waico will probably never shape legislation, much less govern anything. But it does not need authority to matter. All it needs is to be the place where the default gets set. The West still has time to ensure that it is not.

The writer is manager and editorial lead of tech frontiers at Economist Enterprise