What is the most popular artificial intelligence model these days? Ask anyone in the US or Europe, and you’ll probably hear about the relative merits of OpenAI’s ChatGPT, Anthropic’s Claude, or Google’s Gemini. All wrong. Over the past two weeks, the world’s most used AI was one that few Westerners had ever heard of: Kimi K2.6, a Chinese open-source model that topped the pack. OpenRouter leader board.
This space looks at how Western policymakers and CEOs can prepare for the wrong race when considering semiconductor benchmarks and the question of which AI design is best. China, meanwhile, is quietly building something different: an ecosystem of open source prototypes that are cheap and good enough for most use cases. By the time Western capitals find out, China’s AI designs may become global standards and be difficult to replace—even with the most advanced technology.
What is the most popular artificial intelligence model these days? Ask anyone in the US or Europe, and you’ll probably hear about the relative merits of OpenAI’s ChatGPT, Anthropic’s Claude, or Google’s Gemini. All wrong. Over the past two weeks, the world’s most used AI was one that few Westerners had ever heard of: Kimi K2.6, a Chinese open-source model that topped the pack. OpenRouter leader board.
This rank highlights how policy makers and CEOs in Western countries can focus on the wrong race when considering semiconductor criteria and the question of which AI model is the highest. China, meanwhile, is quietly building something different: an ecosystem of open source prototypes that are cheap and good enough for most use cases. By the time Western capitals find out, China’s AI designs may become global standards and be difficult to replace—even with the most advanced technology.
The dominant narrative in the West is that US export controls are blocking China’s access to sophisticated processing chips, slowing the development of Chinese AI companies. This is true but it obscures an important part of the story: If Beijing’s AI focuses on universal access and cost-effectiveness, then Chinese AI companies don’t need the latest chips to win the global AI race.
Take Kimi K2.6, an excellent open-source prototype by Beijing-based Moonshot AI. Turn on industry benchmarksthe model sits inside touch distance of the edge Western ones like Anthropic’s Claude Opus 4.7 and OpenAI’s GPT-5.5. For the price, it is in a different world: the cost of Kimi about $4 per million output tokens, approx six for eight times cheaper than Opus 4.7 and GPT-5.5. (A token is a unit that measures the input and output of AI models.) For the average user, this difference in cost may not make much sense. For a company running hundreds of AI agents, it can be decisive.
Kimi is not an outsider. While the Moonshot pattern leads the chart for use through interfaceanother Chinese model, Alibaba’s Qwen, is fast becoming the default ecosystem for personally– hosted AI models. Two data points help illustrate this point. First, by March, Qwen had caught on more than 50 percent of downloads of open source models in the world, have to be overtaken its biggest competitor in the West, Meta’s Llama, at the end of 2025. Second, Qwen has been downloaded all over. 1 billion times. Interest in Qwen is not limited to cost-conscious companies. Last November, the Singapore government he announced that it would get rid of Llama and build its main AI model with Qwen instead.
With its open-source AI strategy, Beijing is revising the logic of its Belt and Road Initiative (BRI)—but with a twist. The BRI involved Chinese companies submitting fully financed infrastructure projects in an effort to close third countries to China’s orbit. The release of Chinese open source AI models follows the same logic, but this time the infrastructure is invisible and free. The low cost of AI deployment is close to zero (the main costs—the servers and the electricity to power them—are provided by the host country), which makes AI deployment a cheaper investment for Beijing than building ports, railways, or power plants. Additionally, the BRI created a highly visible, Chinese-owned infrastructure that sometimes seemed unpopular. In contrast, the reliance on AI is invisible to policy makers and the population, preventing regression.
China’s AI stake is a long-term one with a clear end goal: to ensure that China’s AI designs become (and thereby create) global standards. The underlying rationale is that the default AI tool could quickly become the de facto industry standard. Once manufacturers and companies build Chinese-made architecture, they use Chinese ideas and technical standards—thus giving Beijing long-term influence. It is part of a larger strategy; in his Chart of 2035 standardsBeijing is aiming to see Chinese products become the global default in an effort to create next-generation technology standards. As an example, China has been handing over its access Sign in free shipping program in recent years. The platform quickly became a household name in the transportation industry after being to be deployed in at least 86 ports in 24 countries.
The battle to set international AI standards will be waged mostly in the global south. Western economies are tied to American models, while China runs the Chinese economy. Swing states are everywhere, and three things make Chinese models a great place to move forward. First, the American-made AI is very expensive for wider distribution in cost-sensitive developing countries. Second, US models are typically trained with Western data, making them unsuitable for understanding local contexts in the global south. In contrast, open source Chinese formats can be downloaded and synced using country-specific data. Exhibit A is AfricaQwen-14BQwen-based AI adapted to 20 African languages through training on African data. Western open source architecture, Llamaoffers only patchy vaccine of international languages of the south.
Third, the growing hatred in the world against the United States strengthens China’s strategy. To make the most of this, Beijing does not even lift a finger. Washington frames the global AI competition as a national security race to be won against China, with no offers for third countries that need cheap and reliable tools.
The results of two major AI conferences for the global economy of the south illustrate this point. Last April in Kigali, Rwanda, the International Conference on AI on Africa and the outcome of the African Declaration on AI called for AI. administration built on values, sustainability, and responsibility—something that would probably seem foreign to Silicon Valley tech executives.
At the Indian AI Impact Conference in February, global south policymakers doubled down. New Delhi suspended the meeting agenda in a “People, Planet, and Development” framework that emphasizes social empowerment, development, and inclusion. As Amitabh Kant, India’s former G-20 sherpa, to put it on the sidelines of the conference: “We provide more data to OpenAI than the US Data from (the) global south helps improve models. These will sell you high-priced products. So India needs to build models on its own data.”
The obvious objection to this narrative—that countries in the global south might reject Chinese AI on security grounds—seems shaky. The international governments of the south did not reject China’s ports, railways or power plants on security grounds under the BRI. With no cheap Western AI offerings on the horizon (US AI companies are cash-strapped and thus unlikely to cut prices), there’s even less reason to expect a different design this time around. Besides, the Chinese version of AI is hard to deny. It helps that open-source models don’t have a single vendor; is transmitted silently through public websitesnot through state-to-state contracts that can grab headlines.
Critics may also report that Beijing’s models come with the control of the Chinese Communist Party. Even if it is a private host, China’s DeepSeek sometimes. he refuses answer sensitive questions. But such censorship has been focused on China’s most sensitive topics such as Taiwan, Tiananmen, Tibet and Xinjiang, making it irrelevant to domestic needs in the global south. Yes, Beijing could threaten to pull the plug on its AI designs, for example by restricting access to updates in the country. follow up relations with Taiwan. But as Washington has become chaotic and unpredictable under US President Donald Trump, global leaders in the global south have every reason to think that relying on Chinese AI may be safer. for a long time rather than relying on American technology.
The real AI race may not be a hardware arms race won by access to the most sophisticated chips. Instead, the global race for AI could be a competition to decide which models and standards will become the default infrastructure in countries that remain up for grabs. China does not need to dominate the most advanced models to win the AI race. If Chinese designs become the cheap and adequate default in emerging markets, Beijing will have built lasting influence for decades.
Instead of fixing the AI horse race between ChatGPT, Claude, and Gemini, Western policymakers might want to consider how they can offer their own version of AI to the global south before the default is set. Meanwhile, developers across Asia, Africa, and Latin America will continue to sync their Kimi, Qwen, and DeepSeek agents.
