Deconstructing Annual Reports of ZHIPU AI, MiniMax, and Mininglamp: A Comparative Analysis of Financial Profiles Amid Divergent AI Commercialization Paths

In the first half of 2026, the Hong Kong stock market's AI sector experienced a significant expansion as Knowledge Atlas（ZHIPU AI） (02513.HK), MINIMAX-W (00100.HK), and MININGLAMP-W (02718.HK) consecutively went public on the HKEX. These three companies represent three distinct business paradigms: foundation models, consumer-facing (To-C) applications, and enterprise-grade Agent services, respectively. With the full disclosure of their 2025 annual reports, their differentiated financial profiles and valuation logics have become increasingly distinct.

From a business model perspective, ZHIPU AI centers on its GLM series of large language models, delivering foundational model capabilities to the market via API interfaces, characterizing a "Model-as-a-Product" technology platform route. MiniMax relies on applications like Xingye and Talkie to reach a global user base exceeding 200 million, pursuing a consumer-grade route driven by LLMs. Conversely, Mininglamp Technology does not build its own foundation models; instead, it orchestrates the capabilities of various third-party models into Agentic Services to deliver business outcomes directly to corporate clients, following a service-layer path.

In terms of financial metrics, the three companies exhibit striking differences. ZHIPU AI reported a full-year revenue of RMB 724 million for 2025, a year-on-year increase of 132%, with a gross margin of 41%. MiniMax recorded full-year revenue of approximately USD 79.038 million (equivalent to roughly RMB 560 million), up 159% year-on-year, with a gross margin of 25%. Mininglamp Technology led the trio in revenue scale, posting full-year revenue of RMB 1.426 billion with a gross margin of 55%, while its operating profit approached breakeven. Notably, its Agentic Service business segment, which surpassed RMB 100 million, maintained a high growth rate.

Regarding valuations, the market currently favors the Price-to-Sales (P/S) ratio to price AI enterprises. As of the market close on June 5, 2026, the total market capitalizations of ZHIPU AI, MiniMax, and Mininglamp Technology stood at approximately HKD 578.3 billion, HKD 173.4 billion, and HKD 33.0 billion, respectively. Converting their annual revenues based on the exchange rate of that day, their respective P/S ratios translate to roughly 733x, 281x, and 21x. Analysts point out that different business models command distinct valuation frameworks: foundation models and consumer apps capture high premiums due to their platform characteristics, whereas enterprise service providers are temporarily priced similarly to traditional software companies due to the presence of mature business segments within their revenue structures.

Figure: Comparison of Core 2025 Financial Metrics Among the HK AI Trio (Knowledge Atlas（ZHIPU AI）, MiniMax, Mininglamp)

Industry insiders believe that these three paths are not mutually exclusive but rather anchor the model layer, application layer, and service layer, respectively. Enhancing model capabilities expands the orchestration space for Agents; the widespread adoption of applications helps materialize the commercial value of models; and the deep penetration of enterprise services drives the prosperity of the entire ecosystem. As AI commercialization remains in its infancy, the coexistence of multiple business models is likely to persist as a long-term structural characteristic of the Hong Kong stock market's AI sector.

NewTimeSpace Disclaimer: All content herein is the original work of NewTimeSpace. Any reproduction, reprinting, or use of this content in any other manner must clearly indicate the source as "NewTimeSpace". NewTimeSpace and its authorized third-party information providers strive to ensure the accuracy and reliability of the data, but do not guarantee the absolute correctness thereof. This content is for reference only and does not constitute any investment advice. All transaction risks shall be borne by the user.