Table of Contents
ToggleThis comparison is particularly jarring to readers in the crypto community. For a long time, the energy consumption of Bitcoin mining has been a common weapon used by mainstream media to attack the blockchain industry; now, AI data centers are rapidly approaching the same level.
Data Center: The Power Black Hole
The numbers speak louder than any description. Global data center power demand has reached approximately 47,000 MW (excluding cryptocurrency mining); of which the total power capacity of AI data centers will reach 29.6 GW by the end of 2025—a figure equivalent to the peak power demand of New York State.
The United States is far ahead in this data center arms race: it has 5,427 data centers, more than 10 times that of the second-place country. The proportion of AI hardware in total data center investment also continues to climb, indicating that the end of computing power expansion is yet to come.
It is worth noting that the above data comes from statistics cited in the report up to the end of 2025, while the scale of large-scale model training continues to expand, and energy demand will only continue to rise.
The Gap Between China and the US: From an Insurmountable Chasm to a Closer Gap
Energy consumption represents the cost of the AI arms race, while model capabilities are the outcome. The report offers a warning to the US camp in this regard: China's top-tier large-scale models have essentially caught up with the US.
Specific figures are as follows: As of March 2026, the lead of Anthropic's top models has shrunk to only 2.7%; the gap between Claude Opus 4.6 and the byte-skipping Dola-Seed-2.0 Preview has narrowed to a mere 39 Elo points. In February 2025, when DeepSeek-R1 was released, it briefly caught up with the top US level, shocking the entire tech world; today, that is no longer an isolated incident, but the norm.
Anthropic, xAI, Google, OpenAI, Alibaba, and DeepSeek—these six companies' flagship models all fall into the same top tier in the AI capability assessment and grading system. In terms of model capability, there is no longer a generational gap between China and the US; only a marginal difference remains.
In terms of the number of models, the United States will produce 50 representative cutting-edge models in 2025, while China will produce 30. Among the top ten representative models, Alibaba, DeepSeek, Tsinghua University, and ByteDance are all on the list.
Investment: A staggering 23:1 ratio
However, in terms of capital investment, the United States and China still live in two different universes.
In 2025, private AI investment in the United States reached $285.9 billion, while China's was only $12.4 billion—a difference of approximately 23 times. This contrast, coupled with the near-identical capabilities of their models, highlights China's efficiency advantage in its "high-efficiency, high-return" approach, and has also raised questions within the US policy community about whether it can maintain its leading position through massive spending.
The landscape of patents and academic achievements is quite different: China accounts for 74.2% of global AI patents, while the United States accounts for only 12.1%; in 2024, China contributed 20.6% of AI academic paper citations, Europe 19.5%, and the United States 12.6%—the United States has ranked third in paper citations.
Acceleration has revealed hidden concerns.
Competition in model capabilities is fierce, and the expansion speed of applications is equally impressive. The report shows that organizational AI adoption has reached 88%, and generative AI will reach 53% of the population within three years, surpassing the speed of any previous wave of technological advancement.
However, along with the rapid expansion, the number of recorded AI incidents also climbed from 233 in 2024 to 362, with reports of issues including model illusions, biases, and cybersecurity vulnerabilities continuing to increase.
The talent pool is also showing cracks: the number of AI talents flowing into the United States has dropped by 89% since 2017, and the double whammy of tightening immigration policies and geopolitical tensions is eroding the United States' long-standing attractiveness to talent.
In the chip supply chain, TSMC still manufactures almost all of the top AI chips, and geopolitical risks are highly concentrated in the Taiwan Strait.
Energy shortages, narrowing model gaps, and stark capital disparities—the 2026 AI Index Report uses numbers to depict a dual competition of resources and capabilities, and the electricity bills in this competition are climbing at a rate that even Bitcoin miners can feel.
