A Diverse World of Sovereign AI Zones

A decade ago, Benjamin Bratton published a groundbreaking book on planetary-scale computation titled “The Stack: On Software and Sovereignty.” The director of the Berggruen Institute’s Antikythera project argued that “geopolitical dynamics today revolve around computation. Data is now a sovereign substance, something over which and from which sovereignty is claimed. Cloud platforms take on roles traditionally performed by modern states, crossing national borders and oceans. Meanwhile, states morph into cloud platforms.”

Bratton saw that the segmentation of planetary computation into multipolar zones, or “hemispheres,” would shape geopolitics going forward. Each hemisphere would consist of vertical layers to form a single interlocking system, or “stack.” The raw materials and energy required for computation constitute the first layer, followed by cloud services for storing and processing data, the local experience interacting with the computational network, then the system of identification for users and, finally, the system interface with users.

His conceptual map of this new configuration of power is becoming manifest with the present competition between China and the U.S. over who will dominate AI. Conflicts between these gigantic “hemispherical stacks” revolve around data sovereignty, the “chip wars” over hardware and the foundational models that reflect different political, cultural and civilizational values.

The battle of the giants, however, is not the end of the story because computation is planetary with evolving stack architectures that will be as diverse as the social complexes that shape them.

Vietnam’s Third Stack

Writing in Noema from Hanoi, Dang Nguyen explains how Vietnam has refused to adopt either Chinese or American models of AI and instead is building its own “core tech stack” of language models, cloud infrastructure and even training data.

Harkening back to the wartime slogans of when Vietnam fought the U.S., and later China, in a lesser conflict, Nguyen quotes the head of one of the country’s leading IT companies saying “nothing is more precious than independence and freedom” when it comes to digital sovereignty, no less than with territorial integrity.

Describing the proud nation’s self-determined temper, she writes: “The familiar poles of AI politics — Silicon Valley’s proprietary platforms and Beijing’s centralized infrastructure — are never named, but everyone understands what is being contested: who gets to define the terms of intelligence itself. The stakes are stack-level choices — black-box dependence or modular improvisation; opacity or legibility; someone else’s roadmap or a sovereign design of your own … The decision is the difference between consuming intelligence as a service and composing it as an act of sovereignty. One rents a mind, the other trains its own in the wild.

“This is, in essence, a claim to AI sovereignty — the ability to build and govern infrastructures on Vietnam’s own terms while still enabling cross-border flows of data, talent and computation. AI sovereignty here does not mean isolation, but authorship — deciding which data, models and rules shape, and will shape, how machine intelligence is built and deployed.

In short, Vietnam is not picking sides. It is building a third stack.”

Infrastructural Non-Alignment

Again, echoing earlier geopolitical references to the post-colonial non-aligned movement of the 1950s-60s that sought to thread a neutral path between contending communist and capitalist powers, Nguyen sees a different map of sovereignty emerging today. “The sharper fault line now runs not between nations but infrastructures — between the guarded logic of proprietary systems and the unruly emergence of open-weight models; between centralized command and distributed improvisation; between the doctrine of safety and the discipline of scrutiny.”

To put it in practical terms: “If OpenAI, Anthropic, and Google DeepMind’s frontier models have largely represented the logic of enclosure, then more open-weight projects like DeepSeek and Meta’s LlaMA — not fully open-source but released in ways that allow retraining and scrutiny — gesture toward a counter-current that is partial, constrained, yet powerful in its transnational diffusion. Even as OpenAI has more recently released ‘open models,’ the broader movement of open-weight diffusion cuts across borders, destabilizing the notion that AI will crystallize into two superpower-led blocs.

“In other words, culture is not what is being exported; technology stacks are.”

Open Vs. Closed AI Models

When China unveiled its DeepSeek open-source model that proved on par with the mostly closed-source advanced AI models in the U.S., former Google CEO Eric Schmidt pointed out that open-source models promote innovative collaboration by allowing universities, researchers, smaller companies and countries to participate in AI development beyond the confines and without the expense of proprietary systems. He warned that the U.S. would fall behind if it did not move more toward open-source models. The trade-off, Schmidt argued, is that very openness carries the risk of misuse by malicious programmers. In the end, he believes, an equilibrium between closed and open systems will likely evolve.

I once asked Kai-Fu Lee, one of China’s leading AI entrepreneurs, whether state censorship there would distort the accurate training of large language models when compared to the West. His response basically recognized Bratton’s notion of hemispherical stacks.

LLMs will indeed carry the imprint of cultural-political values, he posited, not only in China, but everywhere. Different cultural zones with different values will censor different things. While the Chinese state might censor any criticism of the Party, in the West there is a kind of culturally driven “woke” or “anti-woke” censorship over sensitive speech on race and gender. In the Islamic world, there will be censorship over blasphemy against the Prophet Muhammad. Each “great space” will align what is acceptable or not in its LLM models according to their sensitivities.

While computational stacks are spinning a planetary web of communication, a singular monosystem is not what is emerging. Neither will the new configurations merely replicate historically defined territorial boundaries. Rather, the new map will blur into zones of influence where the weight of the major powers will be tempered by the diverse virtual territories of computational stacks adapted to the sovereignty of their own cultivated ways.