Recently, domestic large model companies Zhipu Huazhang (hereinafter referred to as "Zhipu") and Xiyu Technology MINIMAX (hereinafter referred to as "MINIMAX") have successively disclosed their Hong Kong stock listing prospectuses, competing for the title of "first large model stock".

According to the prospectuses and Wind data statistics, Zhipu's revenue is slightly higher, while MINIMAX has a larger fundraising amount but also significantly larger losses.

On the other hand, MINIMAX's cash reserves are substantially greater than Zhipu's, which is crucial for large model enterprises as continued R&D requires significant "burn rate" and better models demand extensive training. Simultaneously, Zhipu's international revenue also lags behind MINIMAX, with the latter deriving approximately 70% of its revenue from outside China, whereas Zhipu's overseas revenue proportion is relatively low, with income基本上来自中国内地.

The most noteworthy issue for Zhipu is likely its major customer volatility. Its top four customers in 2023 (B, C, D, E) and top two customers in 2022 (F, G) largely disappeared after very short transaction periods within the reporting period. In the first half of 2025, the company's institutional customer count decreased by over 2,400, a reduction exceeding 43%.

Information shows that Zhipu and MINIMAX are both star enterprises among the "Chinese AI Six Little Dragons". The two companies have distinct positioning and business models: Zhipu is committed to developing advanced general-purpose large models, building a comprehensive MaaS platform, and operates with a strong B2B and G2G focus through an "open-source base + commercial customization" model. MINIMAX, conversely, adopts a "model as product" approach, acquiring customers from both B and C ends, establishing a high-quality, sustainable revenue matrix centered on subscription services and cloud APIs.

At the current industry development stage, high investment and substantial losses are the norm, and Zhipu and MINIMAX are no exception. Zhipu reported losses of RMB 6.238 billion over three and a half years, while MINIMAX incurred total losses of USD 1.32 billion over three years and three quarters.

According to Wind and prospectus data, MINIMAX's cumulative equity financing amounts to approximately RMB 10.9 billion, whereas Zhipu's equity financing totals around RMB 8.3 billion.

Although MINIMAX raised more capital and has higher losses, the company currently holds more ample funds on its books, which is more favorable for continued R&D spending and developing better large models.

As of the end of the first three quarters of 2025, MINIMAX's cash and cash equivalents stood at RMB 2.577 billion, with trading financial assets of RMB 4.577 billion, totaling RMB 7.154 billion, indicating cash reserves exceeding RMB 7 billion.

As of the end of the first half of 2025, Zhipu's cash and cash equivalents were RMB 2.552 billion, with no trading financial assets disclosed, resulting in significantly lower cash reserves compared to MINIMAX.

Zhipu's international revenue is also far behind MINIMAX's, with MINIMAX generating about 70% of its revenue from outside China. Zhipu's prospectus states that during the track record period, substantially all of its revenue was generated from mainland China. In the first half of 2025, 11.6% of the revenue from Zhipu's localized deployment segment business came from overseas.

Analysis suggests that training and inference costs constitute the primary cost expenditure for large model companies. Based on details disclosed in the prospectuses, the proportion of training and inference costs to revenue is relatively similar for both MINIMAX and Zhipu.

Despite a significant industry-wide decline in average inference costs, the proportion of training and inference costs relative to total costs has been increasing for both Zhipu and MINIMAX.

From 2022 to 2024, and the first half of 2025, Zhipu's computing power service fees as a percentage of R&D expenses were 17.3%, 58.9%, 70.7%, and 71.8% respectively, showing a year-on-year increase.

Interestingly, page 42 of Zhipu's Hong Kong prospectus states, "For the years ended December 31, 2022, 2023, and 2024, and for the six months ended June 30, 2024, and 2023, our computing power service fees were RMB 14.6 million, RMB 311.7 million, RMB 1,552.8 million, RMB 603.2 million, and RMB 1,145.1 million, respectively." In fact, the phrase "six months ended June 30, 2024, and 2023" should likely be "six months ended June 30, 2025, and 2024," as other sections of the prospectus, such as page 253, refer to the "six months ended June 30, 2025."

Consequently, Zhipu's Hong Kong prospectus contains a low-level information disclosure error, raising questions about the diligence of the sole sponsor, China International Capital Corporation Limited (CICC).

According to multiple media reports, many recent Hong Kong listing prospectuses have exhibited issues of "over-packaging," where originally straightforward business models and processes are often described using various sophisticated terminology, making related content obscure and difficult to understand. Other prospectuses have suffered from粗糙 file production and compromised quality. Consequently, Hong Kong's securities regulator issued a rare statement clarifying standards for the quality of listing documents.

Although large model companies are still in their early development stages, and the market shows considerable tolerance regarding their revenue and profitability, some fundamental operational and financial data warrant deeper consideration.

For instance, the issue of high major customer volatility at Zhipu. The prospectus shows that Zhipu's top four customers in 2023 (B, C, D, E) generated no revenue in 2024 or the first half of 2025.

Zhipu's top two customers in 2022 (F, G) generated no revenue for two and a half consecutive years from 2023 through the first half of 2025, with transactions lasting only one year. The third-largest customer in 2022 (H) had minimal transactions with Zhipu in 2023, followed by no transactions in 2024 and the first half of 2025.

Zhipu's largest customer in 2024 (A), after contributing RMB 59 million in revenue that year, recorded no transactions in the first half of 2025. Therefore, the volatility among Zhipu's top five customers is very high, with many customers transacting for only one year during the reporting period.

MINIMAX's customer volatility appears much better. For example, its top three customers in 2023 (A, B, C) remained within the top five customer list in 2024. MINIMAX's new top customer in 2024 (F) remained its largest customer in the first three quarters of 2025.

According to data disclosed in the prospectus, Zhipu's institutional customer count was 5,580 in 2024, dropping to 3,156 in the first half of 2025 – a decrease of over 2,400 customers, representing a change of 43%.

Zhipu's prospectus does not disclose the reasons for the significant customer volatility. Public information suggests that intensified market competition this year, with major players like Baidu and Alibaba capturing market share and other large model firms like deepseek entering the market aggressively, makes Zhipu's customer volatility somewhat understandable.

However, investors should note that it is currently impossible to predict which large model companies will ultimately succeed. Whether the surviving companies will find stable profit models also remains to be seen over time.