China is giving AI away for free, and that's a problem for every data center being built in America.
China built hundreds of AI data centers during the boom and many are now sitting empty, according to MIT Technology Review. The US is currently building far more, but Edward Dowd argues that the comparison may miss the point.
China made a different strategic decision: open-source AI, decentralized architecture, and models like DeepSeek designed to be given away rather than monetized through centralized infrastructure.
That strategy poses a direct threat to the US data center buildout, and in more ways than geopolitics. Open-source models are currently three to four months behind the frontier models, but they are capable of handling roughly 90% of tasks at a fraction of the cost.
If switching costs between AI models are effectively zero, and a good-enough open-source alternative exists for most use cases, the financial case for the massive centralized infrastructure being built right now isn't sustainable.
China's model is the same one it has used in manufacturing: build at scale, drive down costs, undercut everyone else's price, and let adoption follow.
In AI, that means giving the technology away and hoping the world builds on Chinese models before the US can establish dominance.
Whether that works depends heavily on the restrictions the US imposes, but the question of who actually wins the AI race may not be decided by who builds the most data centers.