Collaborative Artificial Intelligence: The New Pluralism
Jacob Taylor: Fellow at the Brookings Institution’s Sustainable Development Center and AI Fellow for 2025.
Joshua Tan: Co-founder and Director of Research at Metagov.
Recently, an international coalition of artificial intelligence labs and cloud service providers achieved something practically thrilling: they pooled their computing resources to present us with Apertus, a massive open-source language model (LLM) made in Switzerland, freely available to users worldwide. Queries directed to Apertus can be processed through Amazon Web Services in Switzerland, Exoscale in Austria, AI Singapore, Cudo Compute in Norway, or the Swiss National Supercomputing Centre, or Australia’s national computing infrastructure. Does this project point the way forward for international collaboration?
In the twentieth century, international cooperation became practically synonymous with the rule-based multilateral system supported by treaty-based institutions such as the United Nations, the World Bank, and the World Trade Organization. However, rivalry among major powers and structural inequalities have eroded the performance of these institutions, leading to a state of paralysis and facilitating the coercion of the weak by the strong. Development funding and humanitarian aid continue to decline, raising doubts about foundational principles like compromise, reciprocity, and efforts to achieve mutually beneficial outcomes.
The diminishing cooperation from national governments has opened more space for other actors—including cities, businesses, philanthropic organizations, and standard-setting bodies—to shape outcomes. In the field of artificial intelligence, a handful of private companies in Shenzhen and Silicon Valley are racing to enhance their dominance over the infrastructure and operating systems that will underpin tomorrow’s economy.
If these companies are allowed to succeed unchecked, almost everyone else will be left with a choice between dependency and irrelevance. Governments and other actors working for the public good will not only be highly exposed to geopolitical coercion and excessive reliance on specific vendors, but they will also have limited options when it comes to benefiting from and redistributing the advantages of artificial intelligence, or managing the negative environmental and societal externalities associated with the technology.
However, as demonstrated by the coalition behind the massive language model Apertus, a new type of international cooperation has become possible. This collaboration does not rely on painstaking negotiations and complicated treaties, but instead on a shared infrastructure for problem-solving. Regardless of the AI scenario that may unfold in the coming years—be it technological plateauing, slow proliferation, general artificial intelligence, or a collapsed bubble—the best opportunity for middle powers to keep pace with the United States and China, and to enhance their autonomy and resilience, lies in cooperation.
Improving the distribution of artificial intelligence products is a fundamental necessity. To achieve this, middle powers and their associated AI labs and companies should work to expand initiatives such as the “Joint AI Inference Tool,” a nonprofit responsible for providing global online access to the Apertus model and other open-source models. However, it is crucial for these countries to also close the capability gap with leading models such as GPT-5 or DeepSeek-V3.1, which requires bolder action. Middle powers can only participate in the development of a package of world-class AI through coordination across energy, computing, data pipelines, and talent.
This pattern of cooperation is not without precedent. In the 1970s, European governments worked to pool their capital and talent, and coordinate their industrial policies to create an aircraft manufacturer capable of competing with American Boeing. An “AI Strategy Similar to the Airbus Project” might require the establishment of an international public-private lab dedicated to pre-training a range of open-source foundational models and providing them freely as a public infrastructure. The result would not be another unified giant among AI titans, but an open infrastructure that numerous actors could build upon.
This approach could propel innovation by allowing national labs, universities, and participating companies close to the frontiers of creativity (such as Mistral and Cohere) to reallocate up to 70% of their pre-training model funding for use in post-training (specialized or inferential models), distribution, and demand-driven applications. Furthermore, this would help empower governments and companies to control the ecosystems governing the operation of artificial intelligence on which they increasingly rely, rather than remain at the mercy of geopolitical uncertainties and business decisions, including those that lead to “gradual degradation and decline.”
However, the potential benefits extend even further. This open infrastructure—and the data pipelines built upon it—could be reused to tackle other common challenges, such as reducing transaction costs in global green energy trade or developing an international framework for collective bargaining benefiting gig economy workers. To fully demonstrate the potential of this new cooperative framework, middle powers should target issues that already have environmental data systems and mature technologies to address; they should prioritize the self-interest of participants over the transaction costs associated with cooperation; and they should make the value of joint work clear to citizens and political leaders.
In a few years, when the current cycle of artificial intelligence innovation and capital reaches its conclusion, middle powers will either lament the demise of the rule-based system while watching the giant AI entities entrench geopolitical fault lines, or they will reap the benefits of new frameworks for innovation that foster cooperation. The case for collaborative artificial intelligence is abundantly clear.
