Chinese internet companies don’t have enough compute to power their artificial intelligence ambitions — and they aren’t about to buy Nvidia outright. That’s a key takeaway from the latest earnings season. Alibaba reported cloud-related revenue surged by 34% year-on-year to the equivalent of $5.6 billion. Management said demand was outstripping supply, and indicated they might need to spend more than the 380 billion yuan ($53.74 billion) initially earmarked for AI buildout over the next three years. That sets up China’s domestic computing power industry for a ” turning point ,” Citic analysts said following Alibaba’s earnings. They expect other cloud vendors in China will likely follow Alibaba in spending more on AI compute. Chinese tech companies don’t explicitly disclose which chip suppliers they are working with. They are officially barred by the U.S. from accessing Nvidia’s most advanced semiconductors, while homegrown alternatives such as Huawei or tech companies’ in-house solutions aren’t publicly traded. But analysts are watching a handful of mainland-traded Chinese stocks as potential beneficiaries. “Higher AI spend should initially benefit AI-related equipment infrastructure firms,” said Brian Tycangco, analyst at Stansberry Research. One of his expected chip beneficiaries is Shanghai-listed Cambricon, whose revenue surged by more than 4,000% in the first half of the year from a year ago. The stock has more than doubled in price this year. Goldman Sachs also rates Cambricon a buy, with a price target of 2,104 yuan. That’s 55% above where the stock closed Friday. “We expect leading domestic Semiconductor Production Equipment (SPE) players and high-end semiconductor suppliers with significant technological moats to benefit as domestic supply and demand increase,” the Goldman Sachs analysts said in a report Thursday. Bloomberg on Thursday reported, citing sources, that Cambricon plans to triple its output next year to fill the demand gap. The Chinese company did not immediately respond to a request for comment. It has denied the report, according to Wind Information. Other than Cambricon and Huawei, Shanghai-listed Hygon makes HSBC’s list as China’s alternatives to Nvidia, AMD and Qualcomm for AI chip design. Also fresh to the Chinese market is graphics processing unit maker Moore Threads, which surged more than 400% in its debut Friday in Shanghai. In a sign of Beijing’s support, regulators had approved the company’s IPO application just 88 days after accepting it. Tencent and ByteDance were one of the many early-stage investors in Moore Threads, according to PitchBook. If there weren’t restrictions on AI chips, “then our cloud revenue should be growing more quickly,” Tencent management said in their latest earnings call. Media reports this week indicate the U.S. government remains divided on whether to let Nvidia sell advanced chips to China. “We do worry about onshore GPU capacity as a constraint on [China] AI deployment in 2026 — issues like latency means using offshore compute (often utilised for training) is impractical for deployment at scale,” Bernstein analysts said in a report Friday. “But with fab capacity increasing into 2027,” the analysts said, “both our discussions with our Bernstein semis colleagues and those with execs at the companies point to this becoming a problem that’s on its way to being resolved.” Some Chinese companies are already making money from generative AI and plan to spend more on the business. Kuaishou said last month it expects full-year revenue for its video generation product Kling AI to reach $140 million, far exceeding the $16 million target set at the beginning of 2025. “We have recently started to [scale] up Kling AI’s training computer power to keep Kling AI at the forefront of technology advancement,” management said on an earnings call, according to a FactSet transcript. Kuaishou expects total capex for the year to increase by mid to high double-digits from 2024. The HSBC analysts on Wednesday said they still prefer Chinese AI components players that have exposure to overseas supply chains as well, such as Innolight. “As the global No 1 supplier of optical modules, Innolight directly benefits from the booming demand for 1.6 [trillion] optical modules,” the analysts said. They pointed out that whether Google’ s tensor processing unit approach to data centers or the graphics processing unit approach by Nvidia, both generate high demand for optical modules. After Google’s Nano Banana Pro launch HSBC said optical module stocks’ consensus earnings estimates in mainland China were revised up. HSBC has a price target of 651.20 yuan, or more than 20% upside from Friday’s close. —CNBC’s Michael Bloom contributed to this report.
Add A Comment