The ongoing correction in Hong Kong-listed AI leaders has prompted investors to reassess the high-growth narrative once attached to large language model companies. As rapid ARR expansion becomes a widely accepted consensus, the key determinants of whether valuations can continue climbing now hinge on compute supply, customer concentration, gross margins, and overall profitability.

On September 8, Z.AI closed down 10.02% at HK$916 per share, breaching the HK$1,000 threshold, while MiniMax fell 5.59%. Notably, Z.AI has now declined for six consecutive trading sessions since September 1, with cumulative losses exceeding 20% during that period.

A September 6 report from Jefferies pointed out that the rapid ARR growth in August was partly driven by concentrated new product launches and a low comparison base, while future expansion could be constrained by compute supply. Additionally, high customer concentration, limited API switching costs, and utilization rates of new clusters may all pressure margin improvements.

Market concerns have shifted from whether Z.AI can grow to whether the heavy investment behind that growth can translate into corresponding profits. Against this backdrop, Jefferies maintains a "Hold" rating on Z.AI with a target price of HK$1,183.79.

ARR Hits US$1.6 Billion, August Growth Shows "Spike" Characteristics

Management at Z.AI disclosed that August ARR reached US$1.6 billion, with expectations to rise to US$2.4 billion by year-end. Based on the "most recent month's revenue multiplied by 12" calculation, August cloud revenue was approximately RMB 893 million, already exceeding the total cloud revenue from the first half of the year.

However, Jefferies believes the rapid ARR growth in August does not fully stem from sustainable demand. The Coding Plan resumed at the end of July after a roughly six-month suspension, combined with the concentrated release of GLM-5.3 and 5.3-Flash, drove significant revenue release during the month. Whether this high growth can be sustained therefore remains to be seen.

Particularly noteworthy is that management previously described ARR growth as "pulse-like," linked to the pace of compute capacity deployment.

Based on this, Jefferies concludes that Z.AI's current growth constraints are shifting from the demand side to the compute side. With next-generation foundation model training resources prioritized over inference resource allocation, even strong demand may not fully convert into revenue if compute expansion fails to keep pace.

Customer Concentration Remains High, API Business Moat Yet to Be Proven

Customer structure is another risk highlighted by Jefferies.

As of August, two customers of Z.AI each contributed over US$250 million in ARR, together accounting for at least 31% of total ARR; the top ten customers accounted for approximately 40% of daily average Token usage. Large customers can quickly boost ARR, but this also makes revenue more sensitive to demand fluctuations from a small number of clients.

More importantly, as large model APIs become increasingly standardized, the cost for enterprise customers to switch models is relatively limited, making transitions between model providers less difficult. With major tech companies like Alibaba and ByteDance continuously increasing their investments in large models, Z.AI must rely on sustained model performance, pricing, and service capabilities to retain customers.

Jefferies therefore argues that future assessments of Z.AI's growth quality should not focus solely on ARR scale, but also on large customer retention and whether the customer base can become more diversified.

Gross Margin Just Improved, Yet Jefferies Expects H2 Decline

Gross margin is another reason Jefferies remains cautious on Z.AI's profitability outlook.

In the first half of 2026, Z.AI's cloud gross margin rose to 24.6% from 18.9% in fiscal 2025, with overall gross margin reaching 26.4%. However, Jefferies projects cloud gross margin will retreat to 20.2% in the second half, bringing overall gross margin down to 20.9%.

The core pressure comes from new compute capacity. After new clusters are deployed, depreciation and operating costs are incurred first, while Token loads take time to ramp up. Short-term utilization shortfalls may therefore weigh on profit margins.

Additionally, Jefferies believes long-text services demand higher requirements for inference efficiency, storage capacity, bandwidth, and other compute resources. The allocation and utilization efficiency of related compute resources may also impact unit costs and gross margins.

Consequently, whether Z.AI's gross margin can continue improving depends not only on revenue growth but also on compute utilization rates, chip efficiency, and product mix.

Earnings Forecasts Upgraded, Growth Sustainability Still Under Scrutiny

Z.AI reported first-half 2026 revenue of RMB 954 million, up approximately 400% year-over-year and 79% quarter-over-quarter. However, this high growth rate still reflects a notable low-base effect, as revenue in the same period of 2025 was only RMB 191 million. Previously, institutions including JPMorgan, UBS, and Goldman Sachs generally projected full-year revenue exceeding RMB 5 billion.

The profitability side remains in a high-investment phase. Z.AI recorded a net loss attributable to shareholders of RMB 2.071 billion in the first half, with R&D expenses of RMB 2.131 billion—R&D spending already exceeding revenue for the same period.

Jefferies raised its 2026 to 2029 revenue forecasts for Z.AI by 37% to 119%, primarily reflecting higher expectations for cloud business growth. The 2026 revenue forecast was raised to RMB 6.906 billion, with 2027 climbing further to RMB 14.848 billion. During the same period, net loss forecasts were reduced by 14% to 21%, indicating improved expectations for both revenue growth and narrowing losses.

However, Jefferies simultaneously lowered the cloud business valuation multiple from 50x ARR to 30x while maintaining a "Hold" rating. The firm believes Z.AI's ARR growth still faces headwinds from compute supply, customer concentration, and gross margin volatility. Therefore, despite upgrading earnings expectations, it remains cautious on valuation and growth sustainability.