DeepSeek Impact One Year Later in China and the United States AI Five Layers Reconciliation: China Leads in Energy, Computing Power Only 1% to 4% of the United States

According to 1M AI News monitoring, The Wire China has published a lengthy analysis comparing the AI competition between China and the U.S. based on NVIDIA CEO Jensen Huang's AI "Five-Layer Cake" framework (Energy, Infrastructure, Chips and Computing Power, Models, Applications).

In the Energy layer, China holds the advantage. Last year, China invested over $500 billion in energy projects, adding 543 gigawatts of new power generation capacity, equivalent to twice Germany's total capacity. In contrast, the U.S.'s total electricity generation has remained relatively flat over the past 20 years, with Morgan Stanley estimating that U.S. data centers may face a power shortfall of about 49 gigawatts by 2028. However, Oxford Institute for Energy Studies senior researcher Anders Hove pointed out that China's systemic bottlenecks in power trading and usage have weakened its capacity advantage, stating that "even with lower capital costs and interest rates, China's electricity prices are not much cheaper."

In terms of Infrastructure and Chip layers, the U.S. is significantly ahead. The U.S. has around 5500 data centers, which is 10 times more than China, hosting approximately 75% of global computing power as of last May. The four U.S. tech giants (Microsoft, Amazon, Alphabet, Meta) are projected to have a combined capital spending of $650 billion this year, while ByteDance plans to invest $23 billion. A report from the Council on Foreign Relations indicated that even if China could produce millions of chips, due to quality differences, its computing power would only be 1% to 4% of the U.S.'s. Researcher Ryan Fedasiuk from the U.S. Chamber of Commerce estimated that China is unlikely to come close to the U.S.'s dominant system in terms of computing power scale until around 2028. However, China's chip industry is progressing towards self-reliance: following SMIC, Hua Hong Semiconductor has become the second Chinese company capable of a 7nm process, and Morgan Stanley estimates that by 2030, China's domestic supply chain could meet 76% of the domestic demand for AI chips.

The gap is narrowing at the Model layer. In February, Chinese companies' token consumption on the OpenRouter platform has surpassed that of their U.S. peers. Rand Corporation researcher Konstantin Pilz believes Chinese models are roughly 3 to 6 months behind the cutting edge, partly due to the use of techniques such as distillation, with accusations from OpenAI and Anthropic that DeepSeek is "free-riding" through distillation. Chinese companies are also continuing to escalate in terms of pricing and open-source strategies.

At the Application layer, Chinese users show a higher acceptance of AI, with the OpenClaw frenzy being one example. However, Assistant Professor Jeffrey Ding from George Washington University pointed out that China's cloud computing and industrial software adoption rates are much lower than those in the U.S., and the path of AI spreading from Beijing, Shenzhen, Shanghai to provinces like Qinghai "will be more challenging than in the U.S."