Chinese internet companies don’t have enough computing to power their artificial intelligence ambitions. Also, they have no intention of acquiring Nvidia outright. Here are the key takeaways from the recent earnings season. Alibaba reported that cloud-related revenue reached the equivalent of $5.6 billion, up 34% year-on-year. Management said demand was outstripping supply and suggested it may need to spend more than the originally planned 380 billion yuan ($53.74 billion) to build out AI over the next three years. This marks a “turning point” for China’s domestic computing power industry, CITIC analysts said following Alibaba’s results. They predict that other cloud vendors in China will likely follow Alibaba’s lead and increase spending on AI computing. Chinese tech companies do not clearly disclose which chip suppliers they work with. Access to Nvidia’s cutting-edge semiconductors is officially banned by the US, while domestic alternatives such as Huawei and tech companies’ homegrown solutions are not listed. But analysts note that a small number of Chinese stocks traded on the mainland could potentially benefit. “Increased AI spending will initially benefit infrastructure companies in AI-related equipment,” said Brian Ticanko, an analyst at Stansbury Research. One chip beneficiary he expects is Shanghai-listed Cambricon, whose sales soared more than 4,000% in the first half of this year compared to a year ago. The stock’s price has more than doubled this year. Goldman Sachs also rates Cambricon as a “buy” and has a price target of 2,104 yuan. This is 55% above the stock’s closing price on Friday. “We expect large domestic semiconductor manufacturing equipment (SPE) companies and high-end semiconductor suppliers with large technological footprints to benefit as domestic demand and supply increase,” Goldman Sachs analysts said in a note Thursday. Bloomberg reported on Thursday, citing sources, that Cambricon plans to triple production next year to fill the demand gap. The Chinese company did not respond to requests for comment. According to Wind Information, the company denied the report. Along with Cambricon and Huawei, Shanghai-listed Hygon is on HSBC’s list of Chinese alternatives to Nvidia, AMD and Qualcomm in AI chip design. Graphics processing unit maker Moore Thread also recently entered the Chinese market, soaring more than 400% in its first trading session in Shanghai on Friday. In a show of support from the Chinese government, regulators approved the company’s IPO application just 88 days after receiving it. Tencent and ByteDance are among the many early-stage investors in Moore Threads, according to PitchBook. Without restrictions on AI chips, “our cloud revenue should grow more rapidly,” Tencent management said in its latest earnings call. The U.S. government remains divided over whether to allow Nvidia to sell advanced chips to China, according to media reports this week. “We are concerned about onshore GPU capacity as a constraint for (China) AI deployment in 2026. Issues such as latency make using offshore compute (often used for training) impractical,” Bernstein analysts said in a report Friday. “However, as fab capacity increases into 2027, both Bernstein’s discussions with semiconductor manufacturing colleagues and company executives indicate this is becoming a problem that is on the way to being solved,” the analysts said. Some Chinese companies are already making money from generative AI and are planning to increase spending on their businesses. Kuaishou said last month that it expects full-year sales of its video generation product, Kling AI, to reach $140 million, well above the $16 million goal it set for early 2025. “We recently began (scaling) up the power of Kling AI’s training computers to keep Kling AI at the forefront of technological advancement,” management said on the earnings call, according to a FactSet transcript. Kuaishou expects total capital spending this year to increase by mid-to-high double digits starting in 2024. Analysts at HSBC said Wednesday they still prefer Chinese AI components companies that also have exposure to overseas supply chains, such as Innolight. “As the world’s number one supplier of optical modules, Innolight is directly benefiting from the surging demand for 1.6 (trillion) optical modules,” the analysts said. They pointed out that both Google’s approach to data centers with tensor processing units and Nvidia’s approach with graphics processing units are creating high demand for optical modules. After the launch of Google’s Nano Banana Pro, HSBC announced that the consensus profit forecast for optical module stocks in mainland China had been revised upward. HSBC’s price target is 651.20 yuan, up more than 20% from Friday’s closing price. —CNBC’s Michael Bloom contributed to this report.
