1. Compared to major full-stack cloud platform providers, what inherent disadvantages do Zhipu and MiniMax face in terms of underlying ecosystem and monetization? 2. Amid tight computing power supply and price wars among major tech firms, what severe challenges will large language models (LLMs) face regarding gross profit margins in the second half of the year?
Cailian Press, September 8 (Editor: Hu Jiarong) – Hong Kong-listed large-cap model stocks, the "duo," both declined today. As of press time, $Z.AI (02513.HK)$ declined nearly 9%, $MINIMAX-W (00100.HK)$ Fell by nearly 6%.
In terms of news flow, a recently released research report by Jefferies not only assessed Zhipu's individual financial status but also offered a pessimistic outlook on broader industry trends. The report stated: "The current Chinese large language model sector is overly crowded. Compared to pure independent AI laboratories, we favor major full-stack cloud platform providers that possess advantages in computing power and data, along with robust balance sheets."
In fact, both MiniMax and Zhipu are categorized as "independent AI laboratories." Analysts point out that such companies are inherently weaker than internet giants with self-sustaining business capabilities in terms of accessing underlying computing power, ecosystem integration, and commercial monetization channels. This fundamental shift in investment logic has directly shaken market confidence in MiniMax's long-term commercial moat.
To align with overseas peer valuation frameworks, Jefferies aggressively lowered the price-to-sales (P/S) multiple for Zhipu's cloud business from 50x to 30x. In the Hong Kong-listed LLM sector, the valuation logic for Zhipu and MiniMax is highly intertwined. As the industry benchmark Zhipu faces a downward shift in its valuation center and "de-bubbling," MiniMax's valuation model has swiftly encountered a "contagious" reassessment by market capital. With Zhipu's P/S ratio falling to 30x, MiniMax's high valuation naturally becomes unsustainable.
Furthermore, the research report warned of corporate pressure from computing costs. Due to the shortage of high-end NVIDIA chips and the ongoing ramp-up of domestic GPU clusters, computing power leasing costs continue to soar, driving up inference costs for large language models. Consequently, if Zhipu's gross profit margin comes under pressure in the second half of the year, MiniMax will likely struggle to escape this cost dilemma.
On another front, independent AI enterprises also face the commercial pain point of "high customer concentration and extremely low switching costs." The price war for domestic large model APIs has currently entered a heated phase, with most B-side clients and developers adhering to a pragmatic approach of "switching to whichever provider is cheaper and easier to use."