Soaring 10-fold with a market value exceeding 570 billion, is Zhipu AI a large model myth or a bubble?

Since artificial intelligence entered the era of generative AI, the market has shown great favor towards manufacturers of AI chips and large AI models. Among them, Cambricon and Zhipu are undoubtedly the most sought - after targets.

In the era of generative AI, the demand for computing power chips continues to rise. Coupled with multiple factors such as domestic substitution, Cambricon's performance has grown at a multiple rate and reaped huge profits. As a result, its market value has also soared, reaching as high as 880 billion yuan.

As a general large - model company, Zhipu has also shown amazing performance in the market. Its current market value is HK$571.6 billion, and the cumulative increase this year has reached an astonishing 1003%, achieving a ten - fold leap.

The picture is from Zhipu's official website

Zhipu is a prominent large - model enterprise in China, indicating that the market is full of confidence in the commercialization potential of large AI models. More importantly, the GLM - 5 base model it launched has achieved a significant improvement in agent and programming capabilities. Giants such as ByteDance, Alibaba, and Tencent have all connected to it and deeply integrated GLM. The number of registered users on the platform has exceeded 4 million.

Zhipu mainly relies on the Model - as - a - Service (MaaS) and commercialization platform to provide general large - model services for institutional customers and individual users, firmly ranking in the first camp of the market. According to IDC data, in the public cloud MaaS market, in terms of revenue, Volcengine occupies more than 40% of the market share. Alibaba Cloud, Baidu Smart Cloud, Zhipu, and China Mobile Cloud rank among the top five.

It is worth noting that the commercialization scale of Zhipu's MaaS platform has entered a new stage, and its revenue scale has increased by 12 times in three years.

In 2022, Zhipu's revenue was 57.4 million yuan. By 2025, it increased to 724 million yuan. The revenue of the API platform reached 190 million yuan, a year - on - year increase of about three times. The API platform achieved an ARR of 1.7 billion yuan (about 250 million US dollars), a year - on - year increase of 60 times.

In the past year, the coding ability of domestic large models has improved rapidly. However, a coding agent task often requires dozens of rounds of model calls. If the single - round response is just a few seconds slower, the overall time consumption may be extended by more than ten minutes.

The high - speed version of Zhipu's GLM - 5.1 brings both capabilities and extremely low latency into the production environment. The model output speed reaches 400 tokens/s, refreshing the current speed limit of API of global large - model manufacturers.

The high - speed version of Zhipu's GLM - 5.1 is suitable for scenarios with extremely high requirements for response latency, such as AI programming, real - time interaction, business decision - making, and real - time voice. It is now open to some enterprise customers on Zhipu's MaaS platform.

More importantly, the popularity of OpenClaw at the beginning of the year not only promoted the rapid development of agentic AI but also led to an explosion in token consumption, and the computing power is undergoing great changes.

In the past year, the global daily average token consumption has increased nearly 300 times. Behind this figure is the transformation of AI from an "auxiliary tool" to a "productive force subject", indicating that token consumption is redefining the way enterprises use artificial intelligence.

IDC believes that the competition focus of MaaS manufacturers is shifting from the past simple price competition to the comprehensive ability competition of "price, performance, and toolchain support".

It is worth mentioning that in the increasingly fierce competition environment of large models, Zhipu took the lead in significantly raising prices, which not only injected growth momentum into its performance but also broke the vicious competition of low - price involution and instead enhanced its competitiveness through differentiated services and technological upgrades.

Due to the continuous strong market demand for GLM, the user scale and call volume have increased rapidly. Zhipu's API call pricing has also been significantly increased, but the call volume has not decreased but increased, and the supply is in short supply.

Of course, it is still difficult to make profits from large models in the short term. The core lies in investing a large amount of computing resources in model training, which depends on AI infrastructure. In 2025, Zhipu's R & D expenditure was 3.18 billion yuan, and 70% of it was used for computing power. Zhipu believes that the scale of token consumption determines the volume of commercial value.

In terms of scale, with the deep penetration of GLM in scenarios such as coding and agents, the consumption of paid tokens has achieved an exponential leap. It claims that it is not for profit but to support the continuously rising exponential curve of high - quality token consumption.

However, due to multiple factors such as the high cost of AI computing power and industry competition, Zhipu's gross profit margin has shown a downward trend. The gross profit margin decreased from 56.3% in 2024 to 41% in 2025, resulting in an expansion of last year's loss to 4.718 billion yuan, compared with a loss of 2.958 billion yuan in the same period of the previous year, an expansion rate of up to 59.5%.

Finally, with the rapid iteration of large AI model technology, the commercialization process of large models is showing a growth trend. From the cloud to the edge, AI is everywhere.

For large - model manufacturers, promoting the development of large models and the transformation of commercialization results in a solid and steady manner will have a positive impact on performance. Against this trend background, the market has shown great favor towards large - model companies. However, compared with their revenue scale, the market value far exceeds the support of revenue scale and profitability. It is particularly important to rationally view the long - term development potential of large AI models.

This article is from the WeChat official account "Yang Jianyong" (ID: iotquanzi), author: Yang Jianyong, published by 36Kr with authorization.