Summary of Key Points
This article fundamentally challenges the common perception of the AI competition between China and the United States: in the past, everyone focused on chip production capacity and GPU performance, but the real battleground has now shifted to the power infrastructure. Faced with the soaring electricity demands generated by AI, the two countries have adopted completely opposite approaches. The United States, due to its power grid’s inability to keep up with the rapid expansion of AI computing power, has fallen into a vicious cycle of “lack of power for computing → reliance on natural gas as a backup → soaring electricity prices → public opposition → approval freezes.” This has even led to a surge in speculative applications for “phantom power” licenses. In contrast, China, leveraging its unified power grid and leading renewable energy capacity, has adopted a proactive strategy of “coordinating power supply with computing needs,” using AI to absorb the excess green energy produced in the western regions. The cost differences between these two approaches will ultimately determine the long-term competitiveness of their AI industries.
---
Detailed and Easy-to-Understand Explanation
1. The Competition Behind Your 3-Sec AI Responses
The speed with which AI generates responses—just 3 seconds— hides a much more substantial barrier than simply chip performance. A high-end AI training GPU consumes as much power as dozens of ordinary home computers, and the electricity usage of a medium-sized AI data center in one hour is equivalent to what an average household uses in a whole year. The growth in AI power demand is staggering: the total electricity consumption of data centers in the United States is expected to double in just two years, with 8.5% of the country’s peak electricity demand coming from data centers by 2027. In China, the annual increase in power demand from the AI industry will be equivalent to more than half of the annual output of the Three Gorges Dam. While chips were once seen as the key to AI competition, it’s now clear that even with the most advanced GPUs, without a stable power supply, all that hardware will be useless.
2. The 474 Gigawatt Waiting List in Texas: A “Speculation Game” for Power Licenses
The governor of Texas has halted all new data center approvals, as the local power grid’s waiting list has reached 474 gigawatts—more than five times the state’s highest-ever electricity demand, with 90% of the applications coming from data centers. However, over 90% of these applications are for “phantom power” projects that will never be realized. In Texas, the rule is “first come, first served.” Regardless of whether the project has funding or customers, or when it will start, submitting an application reserves a spot on the grid. This is similar to how people queue up to buy housing rights before a new development is even planned. After the audit, less than 5% of the hundreds of gigawatt applications will actually be approved, meaning many speculators will be disqualified, and only the truly capable developers will get the chance to proceed.
3. The US Relying on Natural Gas to Sustain AI Power: A Dead End
The US is not against using green energy for AI; companies like Google and Meta have signed long-term contracts for wind and solar power. However, the fragmented power grid makes it difficult to transport green energy from the western regions to eastern areas where computing power is concentrated. As a result, 80% of US data centers are now relying on natural gas generators. This has led to a soaring price of natural gas, with the cost per kilowatt increasing by more than three times. The situation has created a vicious cycle: more AI power is built, but the grid can’t handle it, leading to increased demand for fossil fuels, higher electricity prices, public opposition to data centers, and policies to freeze new data center construction in several states. Thus, the expansion of AI in the US is severely constrained by energy issues.
4. China’s Innovative Approach: Using AI as a “Sponge” for Green Energy
Unlike the US, which waits for power shortages to arise and then tries to make up for them, China has planned ahead with a strategy of “coordinating power supply with computing needs.” Previously, much of the renewable energy in the west went unused due to excess production. Now, AI centers are built in areas with abundant renewable resources, and the electricity is directly delivered to data centers. Tasks that don’t require immediate results are scheduled during peak solar and wind production times. China’s unique ultra-high-voltage power grid allows green energy to be transported with almost no loss over long distances. Policies require that green energy account for more than 80% of the power used in new national-level AI centers, effectively linking AI with green energy consumption.
5. No Absolute Advantage: The Cost Difference Determines Long-Term Competitiveness
While China’s approach isn’t perfect (renewable energy is volatile and costly, and non-national-level centers are not required to use green energy), the long-term cost of running AI on green energy is significantly lower than using natural gas. This means that AI services in the US will be cheaper and more reliable for end-users. In the ultimate competition between China and the US, it’s about who can provide cheaper and more stable power to support AI development.