America’s AI debate keeps returning to the same question: how do we keep China from catching us at the frontier?
It is the right question, but the answer that many consider is too narrow. Several policymakers point to semiconductor chips and their high computing power as the chokepoint limiting China’s AI development.
Yet chips are only one factor in the full technological stack, which includes memory, energy, cloud infrastructure, models, applications, developers, edge devices and much more.
Additionally, which model is adopted by developers or users will shape the future of the digital world, and which rules are emphasized or discarded. If the U.S. wishes to ensure users adopt its frontier models, it should ensure it remains dominant across the full stack and in ensuring open-source and open-weight AI are at the center of its strategy.
Throughout its history, the U.S. tech sector has been dominant because of its open-source approach to critical tools. This approach allowed students, startups, corporations and even governments to experiment broadly and iterate quickly on their ideas without waiting for permission from a few model providers.
Open-source helped build the internet and its app economy, from Linux to Apache. Open-source AI, like these precedent tools, will enable faster adoption by making it easier to test, customize, and commercialize, while making it harder for any firm to monopolize the market.
Open-source encourages competition by preventing AI from concentrating into a closed-lab oligopoly in which a handful of firms decide who gets access, at what price and on what terms. The LangChain State of Agent Engineering report shows open models becoming part of real agent-building workflows.
Nvidia has recognized the same reality: it released Nemotron open-source models amid the boom in Chinese offerings. That is the right instinct, but more U.S. firms should compete in open models rather than retreat from them.
Open source is also essential for edge AI. The future of AI will not live only in hyperscale data centers but also on your phone or in your car.
AI-powered consumer goods need models that can be compressed, locally hosted, audited, and adapted to tight power and latency constraints. The European startup Mistral released small open speech models designed to run locally on phones or laptops.
In the U.S., Nvidia is also pushing in this direction, working to bring AI closer to the edge and to the device. Lastly, Huawei’s new Kirin 9030 smartphone chip allocates more computing power to AI cores. Their 5G base stations now have AI inference hardware, and their autonomous vehicle unit already shipped over 420,000 AI chips for smart driving in the first half of 2026 alone.
China understands that open-source is the future; Alibaba’s Qwen, DeepSeek, Moonshot’s Kimi, and Zhipu’s GLM are open-source models that are increasingly customizable, boosting their adoption and encouraging the creation of an entirely Chinese AI stack as export controls force Chinese firms to work with domestic chip providers.
Wired reported that Qwen became one of the world’s most popular open-source model families, with broad use across the research community. Stanford HAI has also warned that Chinese open-weight models matter because permissive licenses let businesses modify them for specific use cases.
This makes sense as developers, to stretch their funding, will choose what is available, cheap, modifiable, and easy to deploy. If these open-source models perform best on Huawei Ascend or other Chinese silicon chips, or are effective with Nvidia or AMD, China or the U.S., respectively, will win adoption even when it does not control every app.
We are already seeing the effects of China’s open-source approach. It’s difficult to obtain reliable market share data on AI adoption, but recent figures show that Chinese open-source models have surpassed U.S. models in both monthly and total downloads.
Microsoft, which has developed its Copilot tool, is considering a Microsoft-hosted version of DeepSeek as a cheaper model option. This is because developers, who are looking for good models that can perform the work at a fraction of the cost, are now giving models from DeepSeek and Xiaomi a closer look. Closed-source models may be effective at advancing the frontier, but such advancement means little if China can commercialize through an open-source approach.
China’s domination of the open-source approach will have bigger ramifications for the broader ecosystem. It would help prioritize Chinese hardware, such as Huawei’s chips, thereby ensuring customers are dependent on the Chinese tech stack.
DeepSeek V4 variants have been adapted for Huawei chips, and day-zero compatibility between major Chinese models and domestic chips is improving amid restrictions. This would ensure customers remain within the full Chinese tech stack, building a commercial moat that leaves U.S. firms left out of competition. This loss would mean more than the loss of jobs and market share; it would mean that China would be writing the rules of the next tech boom.
The U.S. can ensure it outpaces China in four ways.
First, support open-source and open-weight AI where safety allows. This doesn’t mean allow every frontier model to be open immediately, but ensuring responsible openness will be a national advantage to developers looking to build with the latest models.
Second, make the best open models run best on American and allied infrastructure. That means optimized inference on NVIDIA, Google, Amazon, Cerebras, and AMD chips, strong support across U.S. cloud providers, and reference deployments for AI-powered products and systems.
Third, pair open models with open tools. Apart from models, developers also need compilers, runtimes, serving infrastructure, harnesses, and deployment recipes. China is trying to build that full-stack bridge around Huawei and domestic accelerators. America should encourage all model labs to work with inference providers such as SGLang and vLLM to optimize open models for American infrastructure.
Fourth, avoid diffusion rules that fence in American models while Chinese alternatives spread. Targeted controls are necessary for frontier capability, but overbroad restrictions backfire, pushing global developers toward Chinese models and hardware.
The open-source layer is where the stack becomes widely adopted across society. America should not choose between frontier leadership and open innovation; it needs both.
Closed models will push the frontier, but it is open models that will diffuse capability, create competition, and make the American stack indispensable.
Jason Hsu is a senior fellow at Hudson Institute, where he focuses on the United States’ technological cooperation with allies and partners.
The views and opinions expressed in this commentary are those of the author and do not reflect the official position of the Daily Caller News Foundation.
(DCNF)