Article illustration 1

In a geopolitical chess move echoing Cold War tech rivalries, OpenAI has quietly launched "sovereign AI" partnerships with foreign governments—including the United Arab Emirates—to build nationally controlled artificial intelligence infrastructures. These deals, partially coordinated with the US government, aim to prevent strategic rivals (read: China) from dominating the global AI landscape. But as OpenAI bets that engagement with monarchies will foster liberalization, China is racing ahead with an unexpected weapon: open-source AI models that are rapidly gaining worldwide adoption.

The Sovereignty Gambit

Sovereign AI has become a buzzword across Washington and Silicon Valley, representing a nation's ability to control AI systems within its borders. For OpenAI, this means collaborating on projects ranging from partial oversight to full-stack control—where governments manage everything from hardware to algorithms. Their UAE partnership includes plans for a massive 5-gigawatt data center cluster in Abu Dhabi, with 200 megawatts slated for 2026.

"The common thread is legality—tying infrastructure to geographical boundaries ensures development adheres to national laws," explains Trisha Ray of the Atlantic Council’s GeoTech Center. Yet sovereignty remains nebulous: While the UAE will deploy ChatGPT nationwide, OpenAI insists governments won't peer under the hood or alter core functionalities. Chief strategy officer Jason Kwon states unequivocally: "We're not going to suppress informational resources. We might add, but we're not going to eliminate."

The Open-Source Juggernaut

While U.S. firms navigate ethical quandaries, China is executing a parallel strategy through open-source dominance. Models from Alibaba (Qwen), Tencent, and startups like DeepSeek now rival U.S. offerings in capability—and are spreading faster. Alibaba's Qwen models have been downloaded over 300 million times globally, with 100,000 derivative models built atop them. Japanese startups increasingly adopt Qwen for its superior local-language performance, while UAE researchers recently used it to create a state-of-the-art system.

"China went from being very behind five years ago to now dominating open source," says Hugging Face CEO Clément Delangue. "One gigawatt in the U.S. means every lab redundantly trains similar models. In China, open-source sharing lets labs build on each other's work—effectively multiplying compute efficiency." This acceleration forced OpenAI's hand: Its recent open-weights model release was partly spurred by DeepSeek's explosive January debut.

Democracy’s Dilemma

OpenAI's engagement with autocratic regimes recalls Bill Clinton’s 2000 rationale for bringing China into the WTO: "We can pull China in the right direction." Yet critics note Beijing grew more authoritarian despite Western tech investments. Kwon acknowledges the gamble: "There’s a bet that engagement is better than containment. Sometimes that works, sometimes it hasn’t."

Meanwhile, Silicon Valley’s ethical guardrails appear looser than during 2019’s Google China search engine protests. "The notion that you must adhere to local laws has normalized," observes Ray. Delangue counters that true sovereignty requires inspectable systems: "There is no sovereignty without open source."

Two Paths Forward

The race reveals divergent philosophies: China’s open-source proliferation versus U.S.-led proprietary partnerships. Kwon suggests a hybrid future: "Countries want both the best closed models and open models for different use cases." But as Chinese models gain traction from Tokyo to Abu Dhabi—and Delangue predicts China could lead AI overall by 2025—the sovereign AI battlefield may ultimately be won through collaboration, not containment. In this high-stakes contest, infrastructure is just the opening move; the real victory lies in whose standards govern the algorithms reshaping global power.

Source: WIRED