‘Sovereign AI’ Has Become a New Front in the US-China Tech War (www.wired.com)

OpenAI has been quietly building “sovereign AI” partnerships with foreign governments—some coordinated with the U.S. government—to let nations host and control AI infrastructure under their own laws. A high-profile example is a deal in the UAE that includes a planned 5‑gigawatt data‑center cluster (with ~200 MW coming online in 2026) and nationwide ChatGPT deployment; however, the UAE reportedly won’t get model weights or the ability to alter the system. OpenAI frames these deals as a way to spread democratic-aligned AI capabilities, while insisting it won’t remove informational content at the request of partner states. At the same time, OpenAI has released some open‑weight models, signaling flexibility between closed and open approaches. The development matters because sovereign AI has become a new front in U.S.–China tech competition: Chinese firms (Alibaba, Tencent, DeepSeek) are aggressively open‑sourcing foundation models—Alibaba’s Qwen family has 300M+ downloads and 100k derivative models—letting other labs adopt training tricks and stretch infrastructure investments more efficiently. Technically, “sovereignty” ranges from legal/geographic control of parts of the stack to full local custody of hardware, model weights, and tooling; proponents like Hugging Face argue true sovereignty requires open source, while others accept hybrid closed/open deployments. The result is bifurcating AI ecosystems, faster iteration in open communities, and new geopolitical skirmishes over where models run, who inspects them, and how national laws shape AI behavior.