WAICO and Pax Silica: Two Architectures of AI Access
METHODS · WORKING PAPER – Global governance. Empirical and analytical; sources current to 19 July 2026. This paper compares institutional architectures, not the intentions or political systems of their members. Neither initiative is treated as operationally mature.
Two international AI structures expanded within three weeks of one another.
On 25 June 2026, the European Commission signed the US-led Pax Silica Declaration, joining an initiative organised around secure and resilient supply chains for AI and semiconductors among trusted allies and partners.
On 16 July, representatives of 29 countries signed the agreement establishing the World Artificial Intelligence Cooperation Organization, or WAICO, an independent intergovernmental organisation headquartered in Shanghai. Its stated mission is to promote international cooperation and global AI governance and to make AI development beneficial, safe, and fair.
It is tempting to call them rival AI blocs. That description is directionally useful and institutionally premature.
Pax Silica is a supply-chain and economic-security initiative. WAICO is a new intergovernmental organisation with a wider governance and capacity-building claim. Their legal forms, memberships, and declared functions differ. What makes them comparable is not institutional symmetry. It is their common object: the conditions under which states can obtain and govern the infrastructure of AI.
The Pax Silica proposition
Pax Silica begins from scarcity and trust.
Advanced AI depends on a chain of concentrated inputs: critical minerals, semiconductor design, fabrication, manufacturing equipment, high-bandwidth memory, networking, energy, data centres, cloud services, and models. A disruption or exclusion at one layer can determine what a state or company can build at every layer above it.
The European Commission described Pax Silica as a commitment to AI and supply-chain security and coordination among trusted allies and partners. Its own rationale connected secure silicon supply chains to European technological sovereignty.
The governing idea is that economic security and national security converge in the AI stack. Access is strengthened by aligning with partners considered reliable; resilience is built through coordinated investment, production, and sourcing.
This is a powerful architecture because it begins below the model. A country may have access to excellent software and still lack the hardware, energy, or capital required to use it at scale. Pax Silica targets the material preconditions.
Its central governance question is therefore: who belongs inside the trusted chain?
The WAICO proposition
WAICO begins from participation and unequal capacity.
China’s official sequence since 2023 has repeatedly presented AI as an international public good, opposed technological monopolies and exclusive groups, and called for a stronger voice for developing countries. The 2024 AI Capacity-Building Action Plan added infrastructure, laboratories, model development, language and data resources, professional training, testing, certification, and governance cooperation.
The founding description of WAICO says it is an independent intergovernmental international organisation following the purposes of the UN Charter, with extensive consultation, joint contribution, shared benefits, and a people-centred approach. China says it will help align AI development strategies, governance rules, and technical standards.
The governing idea is that states excluded from the frontier should not remain permanent consumers of rules and services set elsewhere. Access is expanded through models, training, infrastructure, and an institution in which developing states are founding participants.
Its central governance question is: who has a seat, capacity, and recognised voice in shaping the chain?
Two architectures
| Dimension | Pax Silica | WAICO |
| Starting problem | insecure and concentrated supply chains | unequal access and governance participation |
| Organising principle | trusted allies and partners | consultation, shared benefit, sovereign equality |
| Primary material | chips, minerals, energy, infrastructure, models | capacity-building, models, standards, governance, infrastructure |
| Access logic | strengthen access inside a trusted network | broaden access, particularly for the Global South |
| Security logic | resilience and exclusion of untrusted dependencies | controllability plus opposition to overextended national-security restrictions |
| Institutional form | declaration and coordinated initiative | new intergovernmental organisation |
| Principal unresolved question | conditions and consequences of trusted status | budget, legal operation, standards power, and delivery capacity |
These are declared architectures. The table does not establish how either will operate in practice.
The same state can participate in both logics
The world is unlikely to divide neatly into two sealed groups.
States can seek US or allied chips, European regulation, Chinese open-weight models, Gulf energy and capital, local-language datasets, and UN legitimacy at the same time. India and several middle powers have incentives to avoid exclusive dependence. The European Union may participate in Pax Silica while continuing to regulate US providers, fund its own compute, and engage in universal UN governance.
This produces bargaining space. It also produces complex dependency.
A country may possess downloadable model weights from one ecosystem while relying on chips from another, serving software from a third, cloud capacity from a fourth, and standards recognised by a fifth. “Alignment” at one layer does not guarantee alignment across the stack.
The universal layer: the United Nations
WAICO and Pax Silica do not exhaust global AI governance.
The UN General Assembly established the Global Dialogue on AI Governance, bringing all 193 member states and non-state stakeholders into a universal forum. Its first session took place in Geneva on 6-7 July 2026. The UN also appointed a 40-member Independent International Scientific Panel on AI.
This matters because a two-bloc account can become self-fulfilling. The UN layer offers a venue for common scientific assessment, cross-bloc participation, and principles that are not conditional on joining either a trusted supply chain or a China-founded organisation.
The practical question is whether the universal layer can influence the material layer. Dialogue without compute, standards, procurement, or finance may remain inclusive but weak. Bloc initiatives without universal accountability may become effective but exclusionary.
What should be measured
Press statements and declarations are evidence of positions. They are not evidence of outcomes.
A comparative tracker should record the following for each initiative.
Membership and legal status
founding members, accessions, observers, withdrawals;
binding and non-binding instruments;
decision procedures and voting rights;
secretariat, budget, and dispute mechanisms.
Material capacity
funding commitments and disbursements;
data centres, chip projects, energy and network investments;
training places, laboratories, and local-language resources;
model, cloud, and developer-tool access.
Conditions
security, procurement, export, localisation, and licence requirements;
political or trusted-partner criteria;
data-transfer and audit provisions;
restrictions on secondary sharing.
Authority and remedy
who decides admission or exclusion;
which evidence supports the decision;
whether reasons are disclosed;
whether review, exception, correction, or appeal exists;
whether access can be withdrawn and with what notice.
Dependency and exit
substitutability of vendors and components;
portability of data, models, and applications;
availability of local operators and maintenance;
cost and time required to migrate.
From geopolitics to decision records
The Institute has described admissibility at state scale as the recorded decision to admit, limit, hold, or refuse an AI capability’s access to real use.
The same method applies here. Trusted status in a supply-chain initiative is an access decision. Membership in a standards organisation is an access decision. An export licence, model release, cloud restriction, or public procurement approval is an access decision.
Yet these decisions are commonly narrated as strategy rather than recorded as exercises of authority. The result is power without a reconstructable decision episode.
For each consequential access action, a public record should be able to answer:
What was granted or withheld?
Who had authority to decide?
Which criteria and evidence were used?
Which technical mechanism executed the decision?
Who was affected?
How long does the status last?
How can it be challenged, corrected, or reviewed?
This is the point at which international AI governance becomes a decision-authority problem rather than a contest of communiqués.
The present conclusion
It is too early to know whether WAICO will become a standards body, a capacity-building institution, a diplomatic forum, or a mixture of all three. It is too early to know whether Pax Silica will become an integrated industrial network or remain a flexible declaration among partners with different interests.
What is already visible is the object of competition.
The global AI order will be shaped not only by model performance, but by the organisations that determine access to models, compute, chips, standards, training, and trusted status.
Those organisations should be studied as systems of decision power.
Sources
Ministry of Foreign Affairs of the PRC, Signing Ceremony of the Agreement Establishing WAICO, 16 July 2026.
Ministry of Foreign Affairs of the PRC, Press Conference on WAICO, 17 July 2026.
Chair’s Statement of the 2026 WAIC, 17 July 2026.
European Commission, Commission signs Pax Silica declaration, 25 June 2026.
United Nations, Global Dialogue on AI Governance and FAQ.
United Nations, Independent International Scientific Panel on AI.
This is Article 4 of Open-Weight Power: Models, Access, and the Emerging AI Order.
