Views: 0
The global race for Artificial Intelligence is no longer confined to Silicon Valley. China has entered the game with a very particular strategy: "open" models, aggressive pricing, and a release pace that is unsettling the major North American tech companies. But what's behind this move?
When we talk about open-source software, we think of programs like Linux, where thousands of developers can view, modify, and improve the code. The business isn't in selling the software (which is free), but in providing support to the companies that use it on a large scale. It's estimated that Linux runs, in one form or another, on around 200 million businesses worldwide.
With AI, the term "open source" is used more loosely. Chinese companies like Moonshot AI present their models, such as the Kimi family, as open source: their weights (the trained parameters) are available on platforms like Hugging Face for anyone to download and run on their own servers. That's what "open weighting" is, strictly speaking: the final training result is shared, but not necessarily the entire process, the data, or the tools used to create it.
The underlying problem is that every response costs money. This is the key difference with Linux. Every time someone asks a question to an AI model, a process called inference is triggered, which consumes expensive, specialized chips, electricity, and data center capacity. The more users a model has, the more expensive it becomes to operate. If the subscription price is low, the numbers don't add up.
This is exactly what happened with Kimi. Unlike most proprietary US models, Kimi K3 was released as open source, allowing developers to download and run it for free on their own servers. Demand was so high that computing power couldn't keep up with all the new users.
Due to limited access to AI hardware and investment compared to US companies, China has opted for open models, while its US rivals operate mostly with closed, paid subscriptions. It's a way to compete without having the same access to the most advanced chips.
Is losing money a strategy? According to various industry analyses, many Chinese AI companies are operating at a considerable loss. And yet they continue to invest, likely with the backing of the Chinese government. The logical question is: why continue if it's not profitable?
The answer seems to lie in geopolitics. The United States and China are competing for technological leadership in the 21st century, just as they did previously with semiconductors and telecommunications. Offering AI at very low prices, even if it means losses, forces US companies to also reduce their prices, making it harder for them to achieve the profitability their investors seek.
The result is that the real money in generative AI might not lie in the models themselves, but in those who manufacture the chips and build the data centers that everyone needs, regardless of which model wins.
This could be a battle with no clear ending. We don't yet have all the answers to know how this struggle will end. But the history of technology teaches us that sometimes putting obstacles in the rival's path—even at the expense of one's own bottom line—is a valid strategy to gain time, market share, and, with it, future influence.
Meanwhile, the ones who benefit are the users: access to increasingly powerful models at increasingly lower prices. Time will tell if this pace is sustainable.
EN 
ES