A Diverse World of Sovereign AI Zones (www.noemamag.com)

Benjamin Bratton’s “The Stack” thesis — that planetary computation is layered into interlocking infrastructures that increasingly shape sovereignty — is playing out in real time: competition between U.S. and Chinese “hemispherical stacks” centers on data sovereignty, chip supply, cloud services and foundational models, but a third path is emerging. Vietnam, for example, is deliberately building its own core tech stack (language models, cloud infrastructure and domestic training data) to assert “AI sovereignty” — not isolation, but authorship over which data, models and rules govern intelligence on its terms. Bratton’s stack layers (energy/raw materials, cloud compute, local UX, identity systems and user interfaces) are useful for mapping these choices and their geopolitical consequences. Technically and politically, this signals a shift from bipolar domination to infrastructural non‑alignment: a spectrum from closed proprietary systems (OpenAI, Anthropic, Google DeepMind) to “open‑weight” models that enable retraining and scrutiny (DeepSeek, Meta’s LlaMA). Open models accelerate diffusion and innovation across universities, startups and smaller states but raise misuse risks; closed models centralize control and safety governance. LLMs will also encode cultural‑political values (different censorship norms, biases and safety priorities), so the future will be a mosaic of interoperable yet divergent stacks. The implication for AI/ML: expect more diverse architectures, regulatory fragmentation, supply‑chain politics (chips, data flows) and novel opportunities — and risks — for localized model development and governance.