Zhipu Slides for Sixth Straight Session Below HK$1,000 as Jefferies Maintains Hold Rating
nashnova research
Zhipu has fallen six straight sessions, losing over 20% and dropping below HK$1,000; Jefferies raised revenue forecasts but cut the valuation multiple to 30× ARR and held its Hold rating — the market's question has shifted from "can it grow" to whether growth can convert into profit.
Six days of losses — what is the market worried about?
Zhipu has dropped for six consecutive sessions since Sept 1, closing down 10.02% at HK$916 on Sept 8. MiniMax fell 5.59% the same day.
This means → the focus has shifted from "can Zhipu grow" to whether heavy spending on growth can produce matching profits.
Three variables are capping the valuation: compute-supply constraints, customer-concentration risk, and gross-margin pressure.
ARR hit US$1.6 billion — how much is a one-off pulse?
Management disclosed August ARR — annualized recurring revenue, the latest month's revenue multiplied by 12 — reached US$1.6 billion, with a year-end target of US$2.4 billion. August cloud revenue alone was roughly RMB 893 million, exceeding the entire first-half total.
Jefferies noted that August's ARR surge contains a "pulse" component: the Coding Plan resumed in late July after a ~six-month pause, and the concentrated launch of GLM-5.3 and 5.3-Flash drove a one-time revenue release.
In plain terms = the month looked unusually strong partly because pent-up demand flushed through all at once — not every month will repeat it.
This reflects a shift: Zhipu's growth constraint is moving from the demand side to the compute side. Next-generation model training claims resources ahead of inference; if compute expansion cannot keep pace, strong demand may not fully convert into revenue.
Two clients account for 31% of ARR — is the moat deep enough?
As of August, two clients each contributed over US$250 million in ARR, together at least 31% of the total. The top ten clients drove roughly 40% of daily token volume.
This means → large clients can push ARR up fast, but revenue becomes highly sensitive to a handful of accounts.
Jefferies also flagged that large-model APIs are becoming standardized; switching costs for enterprise clients are relatively low. In plain terms = customers can change providers without much friction — Alibaba and ByteDance are both investing heavily, so Zhipu must keep winning on model performance, pricing, and service.
Gross margin just improved — why will it slide back?
First-half cloud gross margin rose from 18.9% (full-year 2025) to 24.6%; overall gross margin reached 26.4% — an apparent improvement.
Yet Jefferies projects second-half cloud gross margin will fall back to 20.2%, with overall margin dropping to 20.9%.
In plain terms = when new compute clusters come online, depreciation and operating costs start immediately, but token workloads take time to ramp — costs arrive first, revenue follows later, dragging margins in the near term.
Revenue up 400% but still deep in the red — where is the money going?
First-half revenue was RMB 954 million, up roughly 400% year-on-year; yet net loss attributable to shareholders was RMB 2.07 billion, with R&D spending of RMB 2.13 billion — R&D alone exceeded total revenue.
This means → Zhipu is still firmly in the spend-to-scale phase; a profit inflection point is not yet visible.
Jefferies' verdict — raise revenue, cut the multiple, hold the rating?
Jefferies raised its 2026–2029 revenue forecasts by 37% to 119%, lifting 2026 to RMB 6.91 billion and 2027 to RMB 14.85 billion; net-loss estimates were trimmed by 14% to 21%.
At the same time, the firm cut the cloud-business valuation multiple from 50× ARR to 30×, maintained a Hold rating, and set a target price of HK$1,183.79.
In plain terms = Jefferies acknowledges Zhipu's revenue is running ahead of expectations, but considers the market's valuation too rich. The pace of compute-utilization ramp-up and large-client retention are the key observation windows for verifying growth quality in the second half.
市场有风险，内容仅供研究参考，不构成投资建议。