From Cost Advantage to Compute Autonomy: Deep Logic and Future Games Behind China's AI LLM Price War

Yesterday afternoon, one of my grad students excitedly shared a piece of news during our group meeting: "Professor, foreign media report that domestic large model API prices are 90% cheaper than in the US!" He then threw out a question: "Does this mean we've already overtaken on the curve in the AI race?" I didn't answer directly; instead, I asked everyone to recall a basic fact from our computer vision experiments—when your training data is insufficient, increasing the model scale can actually backfire. This analogy might be the best entry point for understanding the current price war.