A graphic illustration titled "When the South Writes Its Own AI Rules: Meet the World Artificial Intelligence Cooperation Organization". Photo: Jeff Xiong.
By Jeff Xiong – Jul 20, 2026
Meet the World Artificial Intelligence Cooperation Organization
On July 16, 2026, in a ceremony hall along the Huangpu River in Shanghai, diplomats from twenty-nine countries across Asia, Africa, Latin America, and Europe walked one by one to a signing table. Among them were China, Russia, Indonesia, Brazil, South Africa, Malaysia, Kazakhstan, Laos, Pakistan, and Cambodia. The document they signed established the World Artificial Intelligence Cooperation Organization (WAICO), an independent intergovernmental body headquartered in Shanghai. China’s Foreign Minister Wang Yi signed on behalf of the host government, as twenty-nine founding members signed in turn to sustained applause. The next morning, the World Artificial Intelligence Conference (WAIC) 2026 opened across the river, and the keynote that followed placed this modest signing ceremony in a longer arc.
In his keynote on July 17, President Xi Jinping told the opening session that “the World Artificial Intelligence Cooperation Organization came into being in Shanghai; the conception of a year ago has already been translated into actual results.” He called it “a major measure by China to respond to the call of the Global South and unite the international community to actively promote AI development and governance,” one that “will become an important milestone in the history of AI development.” Then, in the fourth of his four proposals, he made a political commitment that is easy to miss amid the ceremonial language: “We should carry out broad international cooperation, help Global South countries strengthen capacity building, bridge the digital-intelligence divide, promote sustainable development, and avoid creating new historical injustice in the field of artificial intelligence.” A signing ceremony, a milestone, and a warning against new historical injustice — the question is what, exactly, these twenty-nine countries have signed onto.
Not Just Another International Organisation — WAICO’s Sovereign Cooperation Frame
To see what WAICO is, it helps to see what it is not. The dominant existing platform for multilateral AI cooperation is the Global Partnership on AI (GPAI), launched in 2020 under the joint initiative of the OECD and Canada, with France as an early co-lead and the United States and other wealthy democracies as its core membership. GPAI’s orientation tilts toward frontier-risk research, safety standards, and the responsible-stewardship vocabulary familiar in Brussels and Washington. WAICO, proposed by China at WAIC 2025 and signed into existence a year later by twenty-nine founding members all drawn from the Global South, tilts in a different direction: development, capacity building, and inclusive access. The contrast is structural. GPAI’s cooperating parties are wealthy OECD economies; WAICO’s founders are developing states for whom the first question is not how to govern frontier systems they do not yet build, but how to acquire the conditions to build, deploy, and govern any system at all. GPAI’s cooperation logic is to output platforms and rules; WAICO’s is to jointly build infrastructure, human capital, data, and governance. GPAI’s sovereignty stance presupposes first-mover advantage; WAICO’s explicitly names “respect for sovereignty” as a basic principle. Two organisations, two geographies, two theories of what AI cooperation is for.
The intellectual frame for that difference was articulated at WAIC 2026 by Shen Yi, a Fudan University scholar whose keynote, AI as Sovereign Infrastructure Through the Lens of Accessibility, has circulated in Chinese policy circles as a conceptual map for the new organisation. The central move, in Shen’s telling, is a shift in cooperation grammar: from granting opportunities for use to jointly building capabilities. The first grammar describes a world in which wealthy firms hand developing countries access to a finished platform; the second describes a world in which infrastructure, human capital, data, and governance are built together, with local societies retaining real choice and control. Shen identifies three foundations on which the second grammar rests: infrastructure readiness, cooperation-model readiness, and governance-principle readiness. The phrasing is deliberately modest — not a finished blueprint, but the preconditions for any blueprint at all.
Shen is equally explicit about what the second grammar is not. He names two deviations to avoid. The first is the export of closed platforms, which locks recipient countries into a single supplier and erodes their technological autonomy and development choices. The second is duplication and isolation — reading sovereignty as a licence to rebuild everything from scratch, with the cost waste and technical stagnation that follow. The alternative he proposes is compressed into one line: autonomy on the basis of open cooperation; diversity on the basis of interoperability; local capacity on the basis of regional sharing; sovereign equality on the basis of international rules. AI sovereignty, in this account, is not closed nationalism but a development capacity — the ability of a country, under its own laws and culture, to determine how data are used, how models are deployed, and how critical systems operate.
That is also the political meaning of Xi’s fourth proposal — quoted verbatim above — to “avoid creating new historical injustice in the field of artificial intelligence.” The phrase is deliberately historical. It treats the risk of AI-driven dependency as the latest chapter in a much older story, the story of colonial-era technology dependence, and it commits the new organisation to a different ending. Shen’s framework names what that different ending requires in concrete terms: AI treated as sovereign infrastructure that is accessible, usable, affordable, maintainable, and auditable, with local societies keeping real choice and control. For a Global South reader, the value is not abstract. It is the difference between buying a finished system from a foreign vendor and participating in how that system is deployed, what data it learns from, and who maintains it. WAICO, in other words, is not only a different organisational form. It also carries a different yardstick for what counts as AI progress.
An International Public Good — WAICO’s Yardstick for AI
That yardstick, in official Chinese language, is the framing of AI as an international public good. The wording appears in two tier-one documents a year apart. In his July 17 keynote, Xi said that “AI can be an international public good that benefits all humanity,” and that “as a responsible major country, China has always been committed to being a provider of international public goods in the field of AI.” Almost a year earlier, on August 28, 2025, a policy interpretation published by the National Development and Reform Commission (NDRC) by Guo Jianmin, deputy director of the Center for Innovation-Driven Development, put it in the same register: AI, he wrote, should be “an international public good that benefits all humanity,” and the goal is to “build an open ecology of equal, trusted, diverse, and mutually beneficial AI capacity building.” The point of noting the same wording in two documents is not that officials are quoting each other. It is that “international public good” is a stable category in this policy vocabulary, not a one-off slogan.
Stable, but also contested — and Shen Yi’s keynote names the contest directly. In the slide of his talk titled “Redefining the Metrics for AI Progress,” he writes: “Model size, investment volume, and valuation may indicate competition intensity. They cannot establish development quality.” The sentence is short, and it is doing a lot of work. It does not deny that there is a technology race. It denies that winning the race, measured in parameters and valuations, proves anything about whether a society is actually being served. The alternative metrics Shen proposes are social: the efficiency of public administration, the depth of integration between AI and local agriculture, health, and education, the measurable improvement in ordinary people’s lives, and the public value AI produces. The argument is not that parameter races are trivial. It is that they are not the yardstick by which a Global South country should decide whether AI is working for it — and choosing that yardstick for yourself is the first move in writing your own rules for AI.
What does that yardstick look like once it is operationalised? The NDRC’s Artificial Intelligence Cooperation Development Action Plan, released on July 17, 2026 at WAIC, offers one answer in its fourth of eight actions, the “AI Deep Empowerment Action.” The text calls for “promoting AI application and empowerment in science, manufacturing, healthcare, education, agriculture, and governance, so as to use AI to promote economic development, improve social governance, and increase people’s well-being.” Six sectors. Each one is a place where a Global South ministry can already tell you, in concrete terms, what improvement would look like — a clinic diagnosing faster, a district agricultural office forecasting drought, a school adapting a curriculum to a local language. The list is not a survey of cutting-edge research. It is a list of places where AI either shows up in ordinary lives or does not. That is what “development quality” looks like once it is detached from the league table of the largest models.
The eighth and final action, the “AI for Good Action,” names the ethical floor of the same yardstick: “eliminate racism, discrimination, and other forms of algorithmic bias, and safeguard fairness and non-discrimination,” alongside a commitment to “contribute public science products for the Global South and serve the UN 2030 Sustainable Development Goals.” Here, too, the wording has a multilateral anchor. The Global AI Governance Action Plan, issued by China’s Ministry of Foreign Affairs on July 26, 2025 — almost exactly a year before WAICO’s founding — set out six basic principles: “good for the people, respect for sovereignty, development orientation, safety and controllability, fairness and inclusive access, and open cooperation.” Three of those six — good for the people, development orientation, and fairness and inclusive access — map directly onto the yardstick. The yardstick is not suspended in air. And it has, over the past year, begun to acquire a policy spine.
Global South Struggles for Socialism and the Workers’ Side of Marxism
From Principle to Action — WAICO Is Not an Empty Shell
The spine begins earlier than most observers would guess. On August 28, 2025 — almost a year before the Shanghai signing ceremony — Guo Jianmin, in his NDRC policy interpretation, listed the priority tasks for the “AI Plus” global cooperation action. The first read: “Next, build the World Artificial Intelligence Cooperation Organization well; actively implement the Inclusive AI Capacity-Building Programme; promote the open-sourcing of AI; help Global South countries strengthen capacity building.” A reader of that sentence in August 2025 would have found WAICO referred to as a thing to be built well, not yet as a thing that existed. By July 2026, it had a constitution, a headquarters, and twenty-nine founding signatures. The “empty shell” critique, which has become the default dismissal of new multilateral initiatives, has to deal with the fact that WAICO was named as a national agenda item before it was named as an organisation.
The shell, in any case, is not empty. The Artificial Intelligence Cooperation Development Action Plan released the day after the signing ceremony sets out eight actions, each with verbs and deliverables. Action 1 builds cross-border trusted data spaces and multilingual corpora. Action 2 connects computing infrastructure and offers affordable intelligent computing services to developing countries. Action 3 builds an international open-source community and supports localised innovation on top of open models. Actions 4 and 8 — deep empowerment and AI for good — were discussed above. Action 5 co-cultivates talent through joint training, professional standards, and attention to vulnerable groups. Action 6 co-builds standards and rules and pushes for international standard-setting. Action 7 builds safety governance, threat-information sharing, and incident-response cooperation. Read as a single document, it is a work programme with an addressee and a timeline. Read against the yardstick of the previous section, it is the operational shape of “AI as international public good” — not a slogan but a list of things to do.
Three of the deliverables are concrete enough to track over the next two years. In his keynote, Xi announced that, over the next five years, China will provide 5,000 AI-themed research and training slots for developing countries; that it will build international AI application cooperation centres with ASEAN, the Arab League, the African Union, the Community of Latin American and Caribbean States, the Shanghai Cooperation Organisation, and the BRICS countries; and that it will work to deploy the meteorological early-warning system called Mazu in thirty countries. The number thirty here is not the same number as the twenty-nine founding members of WAICO — the Mazu platform is an application, not a membership list, and the thirty countries are early-warning users, not necessarily all signatories. The choice of Mazu as a flagship is telling. Named after the sea goddess revered along China’s southern coast, the system packages meteorological satellites, radar, physical models, and AI into a “measure-report-warn” integrated service that has already been delivered, in a 2.0 version, to Djibouti, with forecast resolution sharpened from nine kilometres to three, according to Chinese reporting that also documented Wang Yi’s signing of the WAICO agreement. In Xi’s words, the platform will “guard the lights of ten thousand homes and shelter the four seas.” A disaster early-warning system is not the most photogenic AI application. It is, however, exactly the kind of thing a Global South minister of finance can cost out against the lives and property it saves.
The four data points now line up along a single timeline. On July 26, 2025, the Ministry of Foreign Affairs published the Global AI Governance Action Plan with its six principles and thirteen articles. On August 28, 2025, the NDRC interpretation named “building WAICO well” as a priority task. On July 16, 2026, twenty-nine countries signed the WAICO agreement in Shanghai. On July 17, 2026, the NDRC released the eight-action plan and Xi announced the three deliverables. “The conception of a year ago has already been translated into actual results,” Xi said in the same keynote, calling it “an important milestone in the history of AI development.” Principle, agenda, action, deliverable — in that order, across twelve months. The milestone claim is in quotation marks because it is Xi’s claim, not this article’s. What this article can say, without quotation marks, is that the policy chain is now dense enough to be trackable, which is a different thing from being immune to criticism. The most common criticism is not that the chain is incomplete. It is that the countries involved are, in a phrase that has circulated online, not yet able to “play with AI” at all.
“Got Electricity?” — The Colonial Mind and the Global South’s Narrow Window
Some voices have mocked these countries — got electricity to play with AI? — as if technical realism were a clever punchline. It is not a clever punchline. It is the contemporary echo of a much older claim: that the colonised are not owed modern capabilities, that their ambition outruns their station, that the proper response to their asking for technology is to ask whether they have even mastered the previous century’s. The phrase can be said coolly, the way colonial claims were always said coolly. But underneath the coolness is the same assumption — that the distribution of capability in the world today is the natural distribution, and that those at the bottom are there because they have not yet earned their way up. That assumption is worth sitting with for a moment, because the historical record says otherwise.
Shen Yi’s keynote puts the same assumption in a longer time frame. The development gap, he writes, “is a historical debt.” The twentieth-century national liberation movements won political independence and entered a long struggle for the right to development, to industrialisation, and to choose a path of modernisation. In the digital age, he argues, data, computing power, algorithms, cloud platforms, and technical standards have become the new conditions of production. These conditions, in his words, “are deeply connected to colonial patterns of resource extraction and knowledge monopoly,” and “the development gap should be understood as a historical development debt requiring collective redress.” The slide’s core claim is the one that gives this section its weight: “Digital sovereignty is the continuation of the liberation project in the digital age.” Behind that sentence sit numbers. According to the International Telecommunication Union’s Facts and Figures 2025, released in November 2025, 2.2 billion people remain offline — ninety-six per cent of them in low- and middle-income countries. Ninety-four per cent of people in high-income countries use the Internet, against twenty-three per cent in low-income countries — a gap of seventy-one percentage points. 5G coverage reaches eighty-four per cent of the population in high-income countries and four per cent in low-income ones — an eighty-point gap. These are not the numbers of countries that have failed to try. They are the numbers of countries that have inherited a bill.
Shen’s framework then names the specific shape of the inheritance, what he calls the five layers of divide in the generative-AI era: basic connectivity, the affordability of computing and cloud services, local data and language resources, education and talent, and the capacity to deploy, evaluate, and govern models. None of the five is solved by a mockery. All five are addressable by cooperation — by the joint construction of infrastructure, human capital, data, and governance that WAICO’s eight actions describe. The institutional answer to the mockery, in other words, is already in writing. The NDRC’s eighth action commits the new organisation to “eliminate racism, discrimination, and other forms of algorithmic bias, and safeguard fairness and non-discrimination.” The wording is not a slogan about inclusion. It is a policy text that treats the mockery’s underlying assumption — that some peoples are owed less capability than others — as a problem to be eliminated from the systems the organisation builds.
From the second half of 2026 onward, a sequence of international conferences will begin to push the content of the Artificial Intelligence Cooperation Development Action Plan into specific cooperation projects with specific Global South countries. The countries that signed the WAICO agreement will not each receive a large portfolio. In the first year or two, most will be involved in one or two flagship projects — perhaps a regional cooperation centre, perhaps a Mazu-style application adapted to local climate, perhaps a cohort of trainees under the 5,000-slot programme or a local-language dataset built on an open-source model. The window for shaping what those flagship projects become is narrow. Once the pattern of cooperation hardens around a few early templates, the cost of changing course rises. Whether a particular Global South government, party, or popular movement treats that as their problem, and how they choose to engage with it, is not something this article can decide. The window is there. What each country makes of it is its own affair.
(Substack)
