China is deploying open-weight artificial intelligence models as a geopolitical tool to establish long-term technological dependency in developing nations, following a strategy that succeeded in telecommunications and cost Western countries billions to reverse. When Moonshot AI released its Kimi K3 model weights in July 2026, the move sparked alarm in Washington, but the real concern extends far beyond cybersecurity or misuse risks. The open-weight front represents a competition over which countries' foundational AI systems will become embedded in the rest of the world's infrastructure, a position that compounds power over decades.

The parallel to China's telecom expansion is striking. Two decades ago, China financed Huawei equipment for countries that couldn't afford Western alternatives, using state credit lines to offer prices no profit-seeking company could match. That strategy worked: Huawei remains the world's largest telecom equipment vendor despite U.S. efforts to undercut it. The dependency China purchased wasn't just market share; it was structural leverage, including leadership in 5G standard-essential patents and the standing to reshape internet architecture at international forums. When the United States finally moved to remove Chinese equipment from rural networks, the bill ballooned from $1.9 billion to $4.98 billion, with Congress needing an additional $3 billion. Six years later, the work remains unfinished.

Why Are Developing Nations Adopting Chinese AI Models?

The appeal of Chinese open-weight models mirrors the appeal of subsidized telecom equipment. Countries like Malaysia now describe Chinese AI models as "something you can control," offering both affordability and the promise of sovereignty without geopolitical strings. When the option promising the most control also comes at the lowest cost, developing nations face a rational choice. Ethiopia's experience with Chinese telecom financing illustrates the pattern: In 2006, ZTE offered Ethiopia $1.5 billion to build a national telecom network with no conditions attached about liberalizing its economy. Ethiopia took the deal precisely because it preserved state control and delivered results; mobile subscriptions grew more than tenfold in five years. But the fine print revealed the cost: enormous contracts awarded without competitive processes, repayment stretching over a decade, and enough leverage lost that Addis Ababa eventually had to split follow-on contracts between two Chinese vendors just to restore its own bargaining power.

The same dynamic is now playing out with AI. Open-weight models are the keystone to an indispensable new technology that one country can provide virtually for free to get a foot in the door. But they are useless without the intricate, interconnected stack of other components that are far more expensive. Weights are not the whole story; open-weight models are not inherently sovereignty-enhancing. The models are only one front in a wider competition clashing over compute, cloud, data, and applications.

How to Understand the Real Stakes in the AI Export Control Debate

• The Misreading of Motivations: Restriction hawks in Washington focus on banning the weights themselves, while openness advocates treat open-weight models as a point in favor of free competition. Both camps stop at the weight file and miss the larger picture of how dependency is built over time through infrastructure, not just software.
• The Infrastructure Trap: Countries adopting Chinese models align themselves with the entire architecture beneath those weights, including cloud services, compute resources, and data pipelines that become increasingly difficult to replace once embedded in national systems.
• The Sovereignty Paradox: Developing nations seeking autonomy from Western influence through cheaper Chinese alternatives may inadvertently trade short-term control for long-term dependency, just as Ethiopia did with telecom networks that became too expensive to remove.

The U.S. response has been predictable: The Trump administration is reportedly weighing Entity List designations for Moonshot AI, which would categorize the company as a national security risk and impose license restrictions on its imports. Treasury Secretary Scott Bessent indicated that sanctions are "on the table." Meanwhile, Nvidia, Microsoft, Meta, and over 230 other tech companies and organizations have issued an industry letter warning against broad restrictions on open models.

But this binary framing misses the central lesson from the telecom wars. The United States ran that experiment before, confronting subsidized Chinese technology in earnest only after much of the world had already adopted it. Washington feared espionage and wartime kill switches, yet there is no publicly confirmed case of a Huawei backdoor exploited for state espionage. The dependency itself was the power. The real question is not whether to ban or bless open-weight models, but how to help developing nations access indispensable technology without surrendering sovereignty or trading access today for dependency tomorrow.

The debate over Chinese AI models echoes concerns the telecom buildout raised two decades ago. China's playbook appears virtually identical to its broadband strategy, involving many of the same companies, the same countries, and the same vocabulary. Open weights play the role subsidized equipment once did: they are the keystone to an indispensable new technology that one country can provide to another virtually for free to get a foot in the door. But the promise of "no strings attached" has a shelf life. Once the welcome phase ends, the fine print emerges.

The stakes extend beyond any single model or country. No one model will win the AI competition outright; adopters hedge too deliberately for that. But position, held long enough, compounds. The question facing policymakers is whether to learn from the telecom experience or repeat it, this time in artificial intelligence.