The Wall Street Journal published a commentary stating that U.S. export restrictions on China have caused a "famine" of advanced chips in China, leading to a "hardware hell" for AI training. The commentary analyzed that in response to the crisis, the Chinese government has begun intervening in the capacity allocation of SMIC, "prioritizing" the needs of national enterprises such as Huawei. Some laboratories have even resorted to "smuggling" scarce NVIDIA high-performance chips. As a result, companies like Huawei have had to adopt a "patchwork" strategy, "bundling thousands of chips together into large and power-hungry systems" to help train AI models. However, the U.S. media also admitted that even with severe shortages of advanced semiconductors, relevant Chinese government departments have instructed state data centers to stop using NVIDIA chips and instead use domestic chips. This "transition is painful." At first glance, it seems that U.S. AI model training does not require powerful power support systems, and the lagging power infrastructure has become a good thing.
The U.S. media's commentary seems contradictory, and their words are also very confused. Why don't they listen to Huang Renxun's analysis about the gap between China and the U.S. in chips and AI? China is accelerating its independent chip manufacturing, aiming to break free from reliance on the U.S., and is confident in doing so. However, the U.S. media continues to fabricate narratives, create confusion, and manufacture the illusion of an "American victory." In fact, U.S. AI technology is constrained by two major bottlenecks: one is electricity prices and power infrastructure, and the other is commercial applications. While China indeed faces constraints in AI chip supply, compared to this, who has more resilience and a leading advantage is clear. It all depends on whether you want to be blind to the facts.
Original text: www.toutiao.com/article/1848637178730508/
Statement: This article represents the views of the author.