As China and the U.S. compete for domination in artificial intelligence, compute power will be the key factor in determining the winner, Bernstein found. The firm also named a few stocks to play on the trend. “China will need to make huge investments in semis and power infrastructure to catch up with the U.S., creating clear upside for related equities,” the analysts said It’s a step back from the market’s current focus on semiconductor access. The U.S. has restricted China’s ability to buy advanced chips for training AI models. Chinese companies have countered by using more, albeit less powerful chips, while launching AI models that are far cheaper to use than those of U.S. rivals. Just as chips are essential to the AI race, so is the electricity needed to power them. “Compute power depends not only on advanced semiconductors, but also on power supply to run hyperscalers, and here China is leading,” Bernstein’s Hong Kong-based analysts said in a March 17 report. By 2035, they predict the U.S. will use 511 zetta floating point operations per second, or ZFLOPS, of AI compute. That compares to the current level of 35 ZFLOPs, the report said, referring to a measure of computing speed. China only has 5 ZFLOPs of AI compute right now, but last year the country added more than 500 gigawatts of power capacity. Maintaining that pace would give China 1,936 ZFLOPs by 2035, the analysts said. That would be more than three times that of the U.S. Fueling power needs Energy security has long been a focus for Beijing, as evidenced by its efforts to stockpile crude oil and diversify into renewables. The country’s push into electric vehicles has also reduced demand for oil, while supporting the development of batteries. “To support rapid growth in power generation and renewables, China will require significantly more battery and grid infrastructure,” the Bernstein analysts said, “which is why CATL and Sungrow remain among our top picks.” The analysts have an outperform rating on the mainland China-listed shares of both companies. Their price target is 530 yuan (roughly $76.96) for CATL, and 260 yuan for Sungrow. CATL, short for Contemporary Amperex Technology, is a major Chinese battery supplier for the country’s fast-growing electric vehicle manufacturers. Sungrow is a large solar and energy storage company. Electricity accounts for about 30% of China’s total energy mix, above the 21% global average, the Bernstein analysts said, noting that the cost of renewables in China can be one-third of those in the U.S. They pointed out that solar has enabled China to rapidly increase its power production, but it’s Beijing’s ongoing push into nuclear as well as expansion of battery-based energy storage that can help improve power supply stability. The Bernstein analysts predict that China’s domestic AI chips will also become more powerful, climbing in efficiency to more than 50% of U.S. chips in 2035 from around 25% currently. Their outperform-rated China semiconductor picks include Cambricon and Hygon, with price targets of 2,000 yuan and 280 yuan, respectively. Both stocks are listed on Shanghai’s tech-focused STAR board. Overall, Bernstein estimates China will ramp up AI-related spending, primarily in data centers, by 32% a year through 2035, while U.S. capex in the category will only grow by 8% a year, albeit from a higher base. —CNBC’s Michael Bloom contributed to this report
