中国AI实验室IPO已然启动：智谱与MiniMax案例；对行业的启示

Chinese AI lab IPOs havebegun...with Zhipu and MiniMax;takeaways for our space China TechnologyPOSITIVE The first two Hong Kong IPOs of AI lab pure-plays kickedoffthe race to grab the land and capital. We summarize ourviews and the implications of increased competition,expanded marketofferings,and more for our space. China TechnologyJiong Shao, CFA+1 212 526 5562jiong.shao@barclays.comBCI, US Lian Xiu (Roger) Duan+1 212 526 4633lianxiu.duan@barclays.comBCI, US The following listed companies are not under coverage by Barclays Research: MiniMax and Zhipu.Information about these companies is being provided for information purposes only and is not aninvestment recommendation by Barclays Research. Xinyao Song+1 212 526 6972xinyao.song@barclays.comBCI, US In early 2026, Hong Kong stock exchange had the first two IPOs of Chinese AI lab pure-plays -MiniMax (0100.HK; not covered) and Zhipu (2513.HK; not covered), a.k.a. Knowledge Atlas, in adirect English translation.These two companies are among the leading Chineseindependent AI labs, as opposed to AI labs inside mega tech platforms such as BABA (OW)and ByteDance (private, not covered).Both China and the US have placed strategicimportance on AI capabilities across multiple segments – foundational models, GPUs, andinfrastructure for their respective national champions.We have reviewed the IPOprospectuses for MiniMax and Zhipu.We share our observations from these documents,which could provide investors with some food for thought as we all are trying to find a pathforward, as well as insights into potential winners and losers in this once-in-a-generationtechnological breakthrough in our coverage universe. Firstoff,AI business model and monetization in China remain highly uncertain.Accordingto its prospectus, around 73% of MiniMax's revenues in the first nine months of 2025 came fromoutside of China, with revenues from its consumer facing AI apps (e.g. virtual companion appand video generation app) making up the bulk of that overseas revenues. For Zhipu, which weconsider to be an early leader in AI agents in China, nearly 90% of its revenues in 1H25 camefrom Chinese enterprises and government entities with more than 80% from on-premisedeployments. Generally, despite LLM companies in the industry claiming they have some of theworld's leading foundational models, their product focus and commercial monetization arequitedifferent.We highlight that Chinese consumers are very reluctant to pay for any apps.Other AI products from China, such as the Kuaishou's (EW) Kling video app, are seeing the samephenomenon and generate the bulk of their revenues from fees paid by consumers from outsideof China. As a result, we believe that if any Chinese AI company aims to build up a fee-payingconsumer user base, they are betterofflooking outside of China. Barclays Capital Inc. and/or one of itsaffiliatesdoes and seeks to do business with companiescovered in its research reports. As a result, investors should be aware that the firm may have aconflict of interest that couldaffectthe objectivity of this report. Investors should consider thisreport as only a single factor in making their investment decision. On the other hand, Chinese businesses may be feeling more comfortable when initial AIdeployments are on-premise as opposed tooffthe cloud.We believe any potential cloud-based deployment by businesses in China could serve as an indicator that enterprises may bemoving beyond the initial experimentation phase for their AI initiatives. In addition, the factthat Zhipu, as an early leader in AI agents in China, has substantially lower monetization fromits agent product, may show that we are still quite early in making AI agents truly useful sousers, enterprises or consumers might be willing to pay for them. Furthermore,we maintainour belief that platform tech companies in both China and the US have the virtual wallsaround their priced "gardens" that would keep "other people's" AI agents outside ofthose walls, rendering them less useful. Another observation is that Chinese consumer AI chatbot's MAUs are much lower thanthose of their US peers.The MAUs of these two Chinese AI labs (mostly MiniMax in this case,since it is the more consumer facing of the two) are much lower (28m across multiple AI apps inthe first nine months of 2025), compared to DeepSeek (private, not covered; ~250mn MAUs as ofSep 25) and ChatGPT (private, not covered; ~800m MAUs as of Sep 25). Also,we don't thinkMiniMax or Zhipu will pose significant competition to BABA's Qwen, which exceeded over100mn MAUs recently. In the tech and internet world, rapid user adoption isoftenviewed as farmore important than near-term monetization, anduser base and user growth of Chinese AIlabs will likely be a key focus for investors. In terms of model capabilities, developers in China are dedicated to advancing modelcapabilities, and each new version would rapidly refresh benchmark scores. To highlight thecosts of training their mode