Openness as Strategy: Beijing, Open Weights, and the Architecture of WAICO

by Emanuele Rossi

China’s AI ecosystem has taken shape at the intersection of technical constraints and diplomatic design. Open-weight models — born partly as a pragmatic response to US export controls on advanced semiconductors, the measure through which Washington has restricted Beijing’s access to chips and manufacturing tools since 2022 on national security grounds — have become the instrument through which Beijing has built a permanent institutional architecture, the World Artificial Intelligence Cooperation Organization (WAICO), capable of translating technological diffusion into influence over global AI governance. The compute constraint meant to hold China back is now the tool Beijing uses to write AI’s global rules — and this shift from necessity to interest is where the coherence of its strategy can be measured. The same openness that made this projection possible now risks losing the exclusivity that made it valuable, as Western actors begin to claim it for themselves.

The World AI Conference in Shanghai, in July 2026, produced more than another statement of intent. Alongside the opening address by Chinese leader Xi Jinping, the Chair’s Statement and two action plans on cooperation and ethical governance, the conference marked the birth of a permanent intergovernmental organisation, headquartered in Shanghai, with its own Council and Secretariat. Twenty-nine countries signed the founding agreement; membership is formally open to all states, with equal voting rights and decisions made by consensus or, failing that, by a two-thirds majority. WAICO’s mandate covers capacity-building, technical standard-setting, interoperability, coordination with UN processes, and — explicitly — open-source ecosystems. The question of open weights ceases to be a technical detail and becomes a matter of multilateral governance. Beijing isn’t just shipping AI models. It’s shipping the rules that govern how they spread.

China’s push towards open-weight models also stems from that concrete compute limitation. Unable to compete on direct computing infrastructure at the scale of major US players, Chinese labs release model weights and offload the burden of distribution to third-party providers, local hosts, and platforms such as Hugging Face, where Chinese models have overtaken US ones in both monthly and cumulative downloads, reaching roughly 41% of the total over the past year, according to the platform’s own data. The model circulates without every request needing to pass through Chinese infrastructure — a dynamic compounded by a performance gap with the US frontier that has narrowed, according to the most recent assessments by the Centre for AI Standards and Innovation, to roughly eight months for the most advanced open-weight models.

Beijing has stopped treating openness as compensation for a computational limitation and has begun treating it as an organised diplomatic lever. Xi’s speech in Shanghai articulated four priorities — openness and cooperation, equitable access, security under human control, UN-centred coordination — alongside explicit criticism of what it characterised as the West’s excessive invocation of national security to restrict access to technology. Behind the multilateral rhetoric lies a selective shaping, rather than an attempt to rewrite the entire architecture of global AI governance: China’s openness isn’t universal. It’s concentrated exactly where technological edge turns into coalition power — technical standards, capacity-building, open-source ecosystems, the domains where Beijing already holds capability and allies.

The comparison with Pax Silica, the US State Department initiative on supply chain security and technological cooperation, sharpens the contrast. WAICO seeks legitimacy through open, formal membership and shared capacity-building; Pax Silica through selective networks of trust and strategic alignment — two distinct principles of legitimacy, not two overlapping initiatives. For now, overlap between the two architectures is minimal — Kazakhstan is the only country present in both — confirming that the competition is being fought through the construction of two parallel ecosystems, between which third governments will have to choose, or try to manoeuvre, with shrinking margins of autonomy.

The openness Beijing promotes concerns the diffusion of Chinese capabilities — models, infrastructure, standards — rather than unconditional market or informational access. Domestically, content controls, data security requirements and restrictions on cross-border flows coexist uneasily with a Western notion of openness: Chinese criticism of Washington’s use of national security sits alongside an equally stringent domestic regulatory framework, oriented towards different ends. Openness therefore remains an instrument of external projection rather than an internal organising principle.

On 10 August, Mark Zuckerberg published a manifesto-letter, The Future Is for Everyone, accompanied by the release of two open-weight Meta models, Muse Glimmer and Muse Spark 1.2. Zuckerberg argues that US labs are at a disadvantage relative to “foreign” ones — a thinly veiled reference to Chinese labs — because of domestic restrictions on training data, and calls on Washington to ease those constraints so as not to cede ground on open weights.

When a central actor in the US ecosystem invokes Beijing’s open-weight advantage to justify a domestic course correction, that lever is no longer China’s exclusive prerogative. China’s competitive advantage on open weights looks less defensible over time, despite its rapid recent diffusion. The institutional infrastructure built around WAICO is a different matter and remains a harder advantage to replicate quickly: an intergovernmental organisation with a headquarters, a mandate and twenty-nine founding members is not improvised with the launch of a single model, however well positioned commercially.

What follows is a shift in the terrain of competition: from the race for the most open or cheapest model to writing the rules by which that openness is administered multilaterally—standards, interoperability protocols, traceability mechanisms. This is ground on which WAICO has already gained an organisational advantage that no single commercial move, however significant, can close in the short term. For third countries, it’s no longer about which model to pick. It’s about which architecture to live inside. The choice increasingly turns on the willingness to embed within one institutional framework rather than the other, with long-term implications for standards, infrastructural dependence, and regulatory autonomy that deserve attention distinct from the model-performance comparisons that have dominated the debate so far.

*****

Emanuele Rossi is an international affairs specialist whose work centres on the Mediterranean and its strategic links to the wider world. He is a Diplomatic Editor at Formiche and a senior analyst at Decode39, and he collaborates with a range of international think tanks and media outlets. The views expressed here are personal.

This article can be reproduced provided the author and the Politeia Research Foundation receive due acknowledgement, and it is not used for commercial purposes. The copyright will remain with the Politeia Research Foundation ©️PRF.