Introduction#

China is positioning itself to potentially rival the United States in artificial intelligence (AI) over the next decade. However, its success may hinge more on energy resources than on semiconductor technology.

Current AI Compute Landscape#

A recent analysis by Bernstein highlights that AI leadership is primarily determined by compute power, which relies on energy availability, data centers, and semiconductor efficiency. Currently, the U.S. leads with approximately 35 zettaFLOPS of AI compute power, while China has around 5 zettaFLOPS, representing only 15% of the U.S. capacity.

China's Energy Advantage#

China's significant advantage lies in its energy production. The country generates more than twice the electricity of the U.S. and is rapidly increasing its capacity by over 500 gigawatts each year, surpassing the rest of the world combined. This energy surplus allows China to expand its data centers aggressively, even if its semiconductor technology is not as advanced.

Future Projections#

Bernstein estimates that if China can overcome its semiconductor limitations by leveraging its energy resources, it could match U.S. compute capacity by 2035. This would require substantial investment, nearly $1 trillion, in AI data centers and a swift enhancement of power infrastructure and battery storage. In a more optimistic scenario, if energy remains the primary constraint, China could exceed U.S. compute capacity, potentially reaching three times the U.S. level by 2035.

Challenges Ahead#

Despite these advantages, China faces significant challenges. It currently lags in advanced semiconductor technology, with its AI chips operating at about one-quarter the efficiency of those produced in the U.S. However, this gap could narrow to over 50% by 2035. Additionally, export controls and limited access to cutting-edge manufacturing tools pose ongoing risks.

Conclusion#

The competition for AI supremacy is complex, with the U.S. leading in chip and software development while China excels in energy production and manufacturing scale. The ultimate outcome may depend as much on energy resources as on technological advancements.