According to a recent analysis by Bernstein, AI leadership is ultimately a function of computing power, which scales with energy, data centers, and semiconductor capabilities. The United States currently holds a dominant position with approximately 35 zettaflops of AI computing power versus China's 5 zettaflops, or about 15% of the US level.

However, China's structural advantage lies in energy. The country already produces more than twice as much electricity as the US and is expanding capacity at an unprecedented rate, over 500 gigawatts annually, more than the rest of the world combined. This provides the opportunity to aggressively scale data centers, even if chips remain less efficient.

Bernstein estimates that China could match US computing power by 2035 if it compensates for its weaker semiconductors through scale. This will require massive investments, nearly $1 trillion in capital expenditures on AI data centers, along with rapid expansion of energy infrastructure and battery storage.

In a more aggressive scenario, where energy remains the only constraint, China could even surpass US computing power, potentially reaching more than three times US levels by 2035.

However, key bottlenecks remain. China lags in advanced semiconductor technology, with domestic AI chips currently operating at about a quarter the efficiency of their US counterparts, although this gap could narrow to more than 50% by 2035. Export controls and limited access to advanced manufacturing tools continue to pose risks.

Thus, the race is asymmetrical: the US leads in chips and software, while China dominates in energy, production scale, and cost efficiency. If energy proves to be the ultimate limiting factor for AI growth, China's advantage could be decisive.

The outcome remains uncertain, but analysis suggests that AI supremacy will be determined as much by megawatts as by microchips.