Zhipu AI’s shares tumble nearly 23% on computing woes, user complaints

Chinese artificial intelligence company Zhipu AI’s shares fell nearly 23 per cent on Monday, wiping out more than HK$70 billion (US$9 billion) in market capitalisation as concerns mount over the Hong Kong-listed firm’s computing resources constraints. The stock slump came as Zhipu issued public calls last week to partner with computing resources providers globally, as user complaints about service quality continue to dog the firm despite frenzied investor interest in recent weeks. The shares closed at HK$560, though they are still up more than 380 per cent since the company’s listing on January 8. The rare public move underscores the commercial challenges for Chinese AI companies to monetise the increasingly powerful capabilities of their underlying models, as computing constraints limit the number of users their models can serve. Known as Z.ai internationally, Zhipu last week issued appeals to both domestic and international inference-compute providers to offer support for its GLM model, offering to share revenue. It was the first time a Chinese AI firm issued a public call seeking computing resources. Inference refers to the deployment of trained AI models. In contrast to the computational resources needed to train AI models, inference-compute costs are usually recurring and more closely tied to revenue generation. According to McKinsey & Co, inference is projected to account for more than half of total AI workloads by 2030. Globally, Zhipu is looking for “high-performance compute providers” to partner with, including those operating “advanced GPU [graphics processing unit] clusters”, said Carol Lin, CEO of Z.ai International. The development comes as Zhipu apologised on Saturday after users in China and overseas reported issues with the company’s flagship “GLM coding plan” product, including response delays and rate limits. Zhipu issued its first public apology in late January after users began demanding refunds when service quality noticeably degraded following the company’s initial public offering (IPO) in early January. The company said at the time that it would begin restricting new user sign-ups in an attempt to shore up service quality for existing users. However, those efforts have failed to assuage user frustration, with many continuing to post complaints on community forums including Reddit and Discord. “We feel very ashamed after reading all your feedback messages on various platforms,” said Zhipu in its most recent apology. “Please give us a little more time, and we will take real action to prove ourselves.” Zhipu operates separate inference computing clusters for international and Chinese users. Amid surging international demand, the company has struggled to secure the additional computing resources needed to serve its growing user base as it pursues international revenue growth beyond its domestic business, particularly in the lucrative market of AI-assisted coding. Since being blacklisted by Washington in early 2025, which has limited its access to US chips, Zhipu has emerged as one of the biggest collaborators with the domestic chip industry as Beijing seeks to replace advanced US semiconductors, notably those from Nvidia. According to a technical report of its latest GLM-5 model, Zhipu already works in “close collaboration” with seven mainstream Chinese chipmakers to ensure inference adaptation with their chips from “day one”, including Huawei Technologies, Moore Threads Technology, Hygon Information Technology, Cambricon Technologies, Kunlunxin (Beijing) Technology, MetaX Integrated Circuits and Enflame. However, in a WeChat article on February 16 announcing its latest “computing power partnership” programme domestically, Zhipu said that “repeated efforts to expand domestic chip supply … [were] unable to completely resolve the current supply shortage”, adding that it was looking for more chipmakers to partner with. Despite these struggles, Zhipu has emerged as an investor darling over the past month as global investors look to tap China’s burgeoning AI industry. Zhipu’s IPO prospectus cited reliance on third parties to provide computing resources and its limited track record in commercialisation as major risk factors for its business.