Why Compute Power is the New Battleground in AI
The race for artificial intelligence supremacy between the United States and China is shifting focus. While much attention has been on semiconductor access following U.S. export restrictions, a new analysis from Bernstein argues that the ultimate decider will be sheer compute power—and on that front, China may be building a surprising lead.
In a March 17 report, Bernstein’s Hong Kong-based analysts contend that China’s massive investments in power generation and grid infrastructure position it to potentially outpace the U.S. in total AI computing capacity by 2035. This perspective represents a step back from the immediate market fixation on chip availability, highlighting a longer-term, foundational competition.
China’s Energy Advantage Could Outpace U.S. Chip Dominance
The analysts project that by 2035, the U.S. will have 511 ZFLOPS (zetta floating point operations per second) of AI compute, up from a current baseline of 35 ZFLOPS. However, China’s trajectory is markedly steeper. Despite starting from a much smaller base of just 5 ZFLOPS today, China added over 500 gigawatts of new power capacity in 2023 alone. If it maintains this rate of power expansion, Bernstein forecasts China could achieve 1,936 ZFLOPS of AI compute by 2035—more than triple the projected U.S. level.
“Compute power depends not only on advanced semiconductors, but also on power supply to run hyperscalers, and here China is leading,” the report states. This advantage stems from a long-standing national priority on energy security, seen in strategic oil stockpiling and a world-leading push into renewables. China’s electricity mix is already around 30% renewables, above the global average of 21%, and the cost of its renewable energy can be one-third of that in the U.S., according to the analysts.
Battery and Grid Infrastructure: The Unsung Heroes
Harnessing this power for AI requires more than just generation; it demands stability and storage. The Bernstein report identifies this as a critical growth vector, directly naming two Chinese leaders in the space as top investment picks.
Contemporary Amperex Technology (CATL), the world’s largest EV battery supplier, and Sungrow, a major solar and energy storage inverter provider, are both rated “outperform” by Bernstein. The firm sets price targets of 530 yuan (~$76.96) for CATL and 260 yuan for Sungrow. The analysts explain that China’s rapid scale-up in power generation and renewables “will require significantly more battery and grid infrastructure,” creating a sustained demand pipeline for these companies. They note that while solar has fueled China’s power production boom, ongoing investments in nuclear power and grid-scale battery storage are essential for providing the consistent, high-capacity power needed by AI data centers.
Semiconductor Stocks Riding the Wave
The report also highlights Chinese semiconductor companies that stand to benefit from the domestic AI build-out, even as they operate with less advanced chips than their U.S. counterparts. Bernstein predicts that the efficiency of China’s homegrown AI chips will climb to over 50% of U.S. chip efficiency by 2035, up from roughly 25% today.
Top picks in this segment include Cambricon and Hygon, both listed on Shanghai’s STAR board. Bernstein rates both stocks “outperform” with price targets of 2,000 yuan for Cambricon and 280 yuan for Hygon. These companies are part of China’s strategy to counter U.S. restrictions by deploying greater volumes of domestically produced, albeit less powerful, chips to train and run increasingly cost-competitive AI models.
