Zhipu has fallen for six consecutive days and broken below the 1000 mark. What happened?

Hong Kong-listed AI leaders continued their pullback, prompting the market to re-evaluate the high-growth logic of large-scale model companies. As rapid expansion of ARR (Automatic Real-Time Ratio) becomes a market consensus, computing power supply, customer concentration, gross margin, and profitability are becoming key factors determining whether valuations can continue to rise.

On September 8, Zhipu closed down 10.02% at HK$916 per share, falling below the HK$1,000 mark, while MiniMax closed down 5.59%. It is worth noting that Zhipu has fallen for six consecutive trading days since September 1, with a cumulative decline of over 20% during this period.

A report released by Jefferies on September 6 pointed out that the rapid growth of Zhipu's ARR in August was due to the concentrated release of new products and the low base effect. Subsequent growth will also be constrained by the supply of computing power. At the same time, factors such as high customer concentration, limited API switching costs, and the utilization rate of new clusters may suppress the improvement of gross profit margin.

The market's concern is no longer whether Zhipu can grow, but whether the investment behind high growth can be recouped with corresponding profits. Against this backdrop, Jefferies maintains its "Hold" rating on Zhipu, with a target price of HK$1183.79.

ARR surged to $1.6 billion, with August growth having a "pulse" component.

According to Zhipu's management, its ARR reached $1.6 billion in August and is projected to rise to $2.4 billion by the end of the year. Calculated based on "revenue of the most recent month x 12", cloud revenue in August was approximately 893 million yuan, exceeding the total cloud revenue for the first half of the year.

However, Jefferies believes that the rapid growth in August's ARR was not entirely due to sustained demand. The Coding Plan resumed at the end of July after a hiatus of approximately six months, coupled with the concentrated release of GLM-5.3 and 5.3-Flash, significantly boosting revenue that month. Therefore, whether this high growth can be sustained remains to be seen.

It is particularly noteworthy that management previously described ARR growth as “pulsating” growth related to the pace of computing power availability.

Jefferies concludes that Zhipu's current growth constraints are gradually shifting from the demand side to the computing power side. Next-generation basic model training resources are prioritized over inference resources. If computing power expansion cannot keep pace, even strong demand may not be fully converted into revenue.

Customer concentration is not low, and the competitive advantage of the API business still needs to be proven.

Customer structure is another risk that Jefferies is concerned about.

As of August, Zhipu had two clients each contributing over $250 million in ARR, accounting for at least 31% of the total ARR combined; the top ten clients contributed approximately 40% of the daily token usage . Large clients can rapidly drive up ARR, but this also means that revenue is more sensitive to changes in demand from a smaller number of clients.

More importantly, as large model APIs become increasingly standardized, the cost for enterprise customers to migrate models is relatively limited, and switching between model vendors is not difficult. With large technology companies like Alibaba and ByteDance continuously increasing their investment in large models, Zhipu needs to rely on model performance, price, and service capabilities to retain customers.

Therefore, Jefferies believes that in the future, judging the quality of Zhipu's growth should not only look at the scale of ARR, but also observe the retention of major customers and whether the customer structure can be further diversified.

Gross margin has just improved, but Jefferies predicts it will decline in the second half of the year.

Gross margin is another reason why Jefferies remains cautious about Zhipu's profit prospects.

In the first half of 2026, Zhipu Cloud's gross profit margin rose to 24.6% from 18.9% in 2025, with the overall gross profit margin reaching 26.4%. However, Jefferies predicts that the cloud gross profit margin will fall back to 20.2% in the second half of the year, and the overall gross profit margin will drop to 20.9%.

The core pressure comes from the addition of computing power. After the new cluster is put into operation, depreciation and operating costs will occur first, while token load needs time to ramp up, and insufficient utilization in the short term may drag down profit margins.

In addition, Jefferies believes that long text services have higher requirements for computing resources such as inference efficiency, storage capacity and bandwidth, and the related computing power configuration and utilization efficiency may also affect unit cost and gross profit margin.

Therefore, whether Zhipu's gross profit margin can continue to improve depends not only on revenue growth, but also on computing power utilization, chip efficiency, and product structure.

Earnings forecasts have been revised upwards, but the sustainability of growth remains a concern.

Zhipu achieved revenue of 954 million yuan in the first half of 2026, representing a year-on-year increase of approximately 400% and a quarter-on-quarter increase of 79%. However, this high growth rate still suffers from a significant low base effect, with revenue in the same period of 2025 being only 191 million yuan. Previously, institutions such as JPMorgan Chase, UBS, and Goldman Sachs generally predicted that Zhipu's full-year revenue would exceed 5 billion yuan.

On the profit side, the company is still in a high-investment phase. In the first half of the year, Zhipu reported a net loss attributable to shareholders of 2.071 billion yuan and R&D expenses of 2.131 billion yuan, with R&D investment exceeding revenue during the same period.

Jefferies has raised its revenue forecasts for Zhipu by 37% to 119% for 2026-2029, primarily reflecting higher expectations for cloud business growth. Specifically, the 2026 revenue forecast has been revised upwards to RMB 6.906 billion, and further to RMB 14.848 billion in 2027. Meanwhile, the net loss forecast has been lowered by 14% to 21%, indicating improved expectations for both revenue growth and loss narrowing for Zhipu.

However, Jefferies also lowered its valuation multiple for the cloud business from 50x ARR to 30x, while maintaining a "hold" rating. The bank believes that Zhipu's ARR growth still faces challenges from factors such as computing power supply, customer concentration, and gross margin fluctuations. Therefore, while raising its earnings expectations, it remains cautious about valuation and the sustainability of growth.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.