Barclays has released a comprehensive analysis following the landmark Hong Kong IPOs of two Chinese AI pure-plays—MiniMax (0100.HK) and Zhipu AI (2513.HK, also known as Knowledge Atlas)—marking a watershed moment for China's artificial intelligence sector. Published on January 27, 2026, the report dissects the business models, monetization challenges, and competitive dynamics that will shape the AI landscape across Barclays' China technology coverage universe, including Alibaba Group, Baidu, and Tencent.

Explosive Debuts, Uncertain Business Models

The IPOs kicked off 2026 with extraordinary investor enthusiasm. Zhipu raised approximately 558 million US dollars (RMB3.9 billion) on January 8, with shares surging 87% to a US$12 billion valuation by the same date. MiniMax followed on January 9, raising 619 million US dollars (RMB4.3 billion), with shares climbing 134% to a US$16 billion valuation by the same date. Yet beneath the froth lies a sobering reality: AI monetization in China remains highly uncertain, and the path to profitability is littered with obstacles.

According to Barclays' review of the prospectuses, MiniMax derived approximately 73% of its revenue in the first nine months of 2025 from outside China, with consumer-facing AI apps—such as virtual companion platform Talkie/Xingye and video generation tool Hailuo AI—accounting for the bulk of overseas income. This reliance on international markets underscores a critical challenge: Chinese consumers remain deeply reluctant to pay for AI applications. Similarly, Kuaishou Technology's Kling video app generated roughly 70% of its revenue from non-Chinese users, reinforcing the pattern that fee-paying consumer bases for Chinese AI products lie predominantly abroad.

Zhipu, by contrast, has taken a different route. Nearly 90% of its revenues in the first half of 2025 came from Chinese enterprises and government entities, with over 80% derived from on-premise deployments. This suggests that Chinese businesses, at least initially, prefer localized AI solutions to safeguard data privacy and meet specific operational needs. However, Barclays cautions that such on-premise arrangements raise questions about recurring revenue streams and the scalability of standardized APIs globally.

The MAU Gap and Monetization Struggles

The user engagement metrics paint a stark picture. MiniMax's AI-native products averaged 28 million monthly active users (MAUs) in the first nine months of 2025—a fraction of DeepSeek's approximately 250 million MAUs and ChatGPT's roughly 800 million MAUs as of September 2025. Even Alibaba's Qwen chatbot, which recently surpassed 100 million MAUs, dwarfs MiniMax's reach. Barclays emphasizes that in the technology and internet world, rapid user adoption typically trumps near-term monetization, making user base and user growth critical metrics for investors evaluating Chinese AI labs.

Monetization per user also lags. MiniMax's average revenue per paying user (ARPU) for AI-native products stood at US$34 in 2024—just 10% of US$346 ARPU. Meanwhile, Zhipu's API revenues represent only about 4% of OpenAI's $1 billion API revenue in 2024, with MiniMax's share even smaller at roughly 1%. These figures suggest that Chinese AI companies, despite claiming world-leading foundational models, face an uphill battle in converting technical prowess into commercial success.

Training Costs and Unit Economics

Barclays' analysis of R&D and cost structures offers a window into the capital intensity of AI development. Zhipu spent approximately 630 million on R&D (a proxy for model training and upgrading) from 2022 to mid-2025, while MiniMax allocated around 450 million over the same period. Extrapolating from these figures, Barclays estimates that Chinese tech giants may spend roughly 100millionto100millionto150 million per quarter—or potentially higher—on AI training, a sum that still falls well short of leading U.S. AI labs' expenditures.

Inference costs, reflected in cost of goods sold, reveal gross margins of approximately 60-70% for customized enterprise APIs and on-premise deployments. Consumer AI products, however, appear to operate near breakeven or at a loss on an inference level. Barclays suggests that platform companies like Alibaba, Baidu, and Tencent, which offer largely free-to-use AI chatbots, may be absorbing inference-level losses while monetizing AI indirectly through existing products, cloud services, and API offerings.

Implications for the Broader Ecosystem

The strong share performance of Zhipu and MiniMax has already reverberated across the sector, driving gains for Alibaba—which Barclays views as the leading full-stack AI player in China. Alibaba holds approximately 16.5% of MiniMax and 7.5% of Zhipu (through Ant Group), positioning it to benefit from the expansion of independent AI labs. Barclays also notes that Alibaba's Cloud business, along with the broader cloud infrastructure industry, stands to gain as these startups rely heavily on rented cloud GPUs for training and inference.

Yet the IPO frenzy raises a critical question: with Zhipu and MiniMax commanding market caps of 12 billion and 16 billion, respectively, on first-nine-months 2025 revenues of just 50 million to 60 million, every AI lab in China may now be eyeing a public listing. Consensus-implied multiples reveal a wide valuation range—Zhipu and MiniMax trade at 45-55x price-to-sales divided by growth, while OpenAI sits at 15-20x. This suggests Chinese AI companies are priced for aggressive growth and dramatic monetization improvements, reflecting robust investor enthusiasm but also heightened risk.

The Road Ahead

Barclays identifies several key takeaways for investors navigating this once-in-a-generation technological shift. First, AI business models in China are still in flux, with no clear consensus on whether consumer apps, enterprise APIs, or on-premise deployments will dominate. Second, Chinese consumers' unwillingness to pay for AI services means companies targeting fee-paying users must look abroad, where competition with established global players is fierce. Third, the rapid pace of model development—with major updates every one to three months—makes current benchmark rankings a fleeting snapshot rather than a definitive measure of capability.

Finally, Barclays highlights that platform tech companies in both China and the U.S. maintain "walled gardens" that could limit the utility of third-party AI agents, rendering them less valuable to users. For now, the market is betting that Chinese AI labs can overcome these hurdles. Whether that optimism is justified will depend on their ability to scale user bases, improve unit economics, and carve out defensible niches in an increasingly crowded and capital-intensive arena.