At the recent World Artificial Intelligence Conference in Shanghai, President Xi Jinping made China’s AI strategy plain: Beijing is not treating open-source artificial intelligence as merely a product category. It is wielding it as an instrument of statecraft.
Xi called open-source AI a “historic opportunity,” promised training and technical assistance for developing countries, and presented Chinese models as a global public good.
China has even created a World AI Cooperation Organization that has already enlisted 29 countries.
This is a bid to shape not only which models the world uses, but whose standards, infrastructure, and influence will surround them.
Measuring the AI Race
I realize lawmakers have been understandably focused on what the economic possibilities of AI may ultimately be.
But in order to truly realize AI’s full potential, the White House, Congress, and regulators have to understand how different types of AI models either accelerate — or throttle — the speed at which AI brings its benefits to American society.
In short, a little Open Source 101 is needed in the Beltway.
The United States typically measures the AI race by computing power and the performance of the most advanced models. Those advantages are important. But national power also requires deploying technology broadly enough so that industries, allies, and partners build their futures around it.
That is where open-source AI can be a game changer.
Beating US Systems
In July, Beijing-based Moonshot AI unveiled Kimi K3, a general-purpose model for coding, analysis and writing. Its performance immediately challenged assumptions about the durability of America’s AI lead.
In independent rankings, Kimi beat leading systems from Anthropic and OpenAI on a front-end coding test and ranked above Anthropic’s flagship model in a broader text evaluation, all while costing 40 percent less to run.
Moonshot’s decision to make Kimi available for companies and governments to customize and operate on their own infrastructure places the United States at a critical inflection point.
America naively believes it has maintained a comfortable lead in the AI race because its closed-source models offer capabilities that open-source and foreign competitors cannot match. Customers accept higher prices and limited customization to use the world’s best technology.
A Chinese model offering comparable performance at lower cost, with greater flexibility and local control, weakens that value proposition. Even if America continues producing exceptional frontier models, it may not remain the country whose technology is most widely adopted.
Strategic Risk
Kimi is merely the latest evidence of this shift. DeepSeek and Alibaba’s Qwen family have already shown that Chinese open-source models can approach the technological frontier while remaining affordable, adaptable, and widely available.
Together, they give Beijing multiple opportunities to make Chinese technology the default foundation for global AI.
As developers learn these systems, companies build around them and governments incorporate them into public services, each adoption strengthens the ecosystem and raises the cost of switching.
The strategic risk is a global AI system increasingly organized around Chinese models, standards and infrastructure.
Bargaining Power
As that dependence deepens, it creates geopolitical leverage. Reuters reported that Beijing is considering restrictions on overseas access to its most advanced models, including more open versions.
China could encourage adoption while its models are inexpensive and accessible, then reserve later capabilities or updates for favored countries and companies.
Its rare-earth export controls have demonstrated how dominance over a critical input becomes bargaining power. AI access and advanced capabilities could eventually serve a similar function.
The Security Trade-Off
Reliance on Chinese models would also create an enduring security-assurance problem. Widespread use of an adversary’s technology expands opportunities for espionage, manipulation, and disruption.
Hosted services can collect prompts, proprietary information, and operational data. Downloadable models can be distributed with compromised code, libraries, or update mechanisms.
These risks multiply as adoption spreads. Every new deployment expands the number of systems exposed to Chinese technology while making the surrounding ecosystem more entrenched and harder to displace.
A Smarter Strategy
To be sure, Chinese models should not become embedded in American government or corporate networks, and I am confident in our superior cyber capabilities to limit the impact to our domestic national security and economic networks.
But keeping them out is a short-sighted, insular defensive measure, not a viable strategy for winning the global competition.
Unfortunately, Washington seems so focused on the down sides of open-source AI that it is in imminent danger of ceding global AI leadership.
Our leaders in Washington must realize that the open-source AI train has left the station, and if they persist in foolishly focusing only on the negatives, they will be ignoring the great opportunity for the US to become the global AI leader of choice. We will be left holding a ticket to second place.
Our strategy should be to embrace open-source AI and produce open models that are safer, more capable, and priced competitively.
Vice Admiral Lou Crenshaw (US Navy, Ret.) is the former Deputy Chief of Naval Operations for Resources, Requirements and Assessments (N8) and is a Fellow in the National Academy of Public Administration.
The views and opinions expressed here are those of the author and do not necessarily reflect the editorial position of Military AI.
Have a perspective to add? See our Write for Us page.
