HSBC maintains Hold on Z.AI, cuts target price but raises ARR forecast

Overview

This is an HSBC company research report on Z.AI (2513.HK). The core of the report is an upward revision to Z.AI's ARR growth assumptions, while the target price is cut due to long-term margin, competition, and share count factors. HSBC maintains a Hold rating, believing the company has strong growth momentum in China's LLM and MaaS market, but industry competition and a declining scarcity premium limit valuation upside.

Core views

HSBC believes Z.AI's commercialization progress is faster than previously expected, with ARR already reaching USD1bn in July 2026 and potentially reaching USD2bn in December 2026 and USD6.5bn in December 2027. The launch of Kimi K3 reinforces the trend of continued increases in parameter scale and also indicates that the lead window in frontier model capability is narrowing. With its 1GW data center, breakthroughs in inference on domestic chips, and XCore Sigma's software capabilities, Z.AI has an opportunity to ease compute bottlenecks and capture strong user demand; however, a higher share of API revenue will pressure margins, and competitor model iterations, IPOs, or fundraising could also weaken Z.AI's scarcity premium.

Analysis framework

The report analyzes Z.AI using DCF valuation, ARR forecast revisions, peer P/ARR comparisons, model parameter scale, OpenRouter token usage trends, and competitive events. On valuation, it uses a 10-year DCF model and references ARR and valuation transactions of frontier AI companies such as Anthropic; on fundamentals, it focuses on GLM iteration, parameter and throughput performance after the Kimi K3 launch, compute supply, and how user demand translates into ARR.

Methodology notes

• Valuation methodsDCF

  10-year DCF valuation

  HSBC's target price is based on a 10-year DCF model, assuming a WACC of 11.8% and a terminal growth rate of 3%, while incorporating assumptions for the risk-free rate, market risk premium, regulatory risk premium, beta, and capital structure.

• Valuation methodsP/ARR

  ARR multiple comparison

  The target price implies roughly 14x P/ARR for December 2027, higher than Anthropic's roughly 8x; the report believes Z.AI's faster ARR growth can partly justify this premium.

• industry_analysisScaling law

  Parameter scale and model capability iteration

  The report attributes Kimi K3's capability improvement to an increase in parameter scale from 1trn in K2.7 to 2.8trn, and on this basis judges that Chinese open-weight models may enter competition around the 3trn parameter scale.

• operating_indicatorToken consumption trend

  OpenRouter token usage trend

  The report uses OpenRouter daily token trends to observe demand changes after the GLM-5.2 launch, and views token usage growth as a supplementary indicator of MaaS and API commercialization momentum.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Z.AI 2513.HK

  Covered company

  Strengths

  ARR growth is faster than expected, GLM models continue catching up with leading Chinese and U.S. models, it is already operating a 1GW data center, and it improves heterogeneous AI compute utilization through XCore Sigma.

  Weaknesses

  A higher API revenue mix may reduce gross margin, long-term FCF assumptions have been lowered, and the July 2026 H-share placement brings equity dilution.

  Comparison

  The target price implies 14x P/ARR for December 2027, higher than Anthropic's roughly 8x, but the report believes Z.AI's ARR growth of roughly 220% is faster than Anthropic's roughly 50%.

  Risks

  Competitor model iterations, lock-up expiry, geopolitical restrictions, cash burn, and refinancing needs.

• Anthropic

  Valuation and ARR comparison peer

  Strengths

  Used as a valuation reference for global frontier AI platforms, with secondary market trading valuation around USD1.2trn.

  Weaknesses

  The report estimates its ARR growth is lower than Z.AI's.

  Comparison

  Anthropic is at roughly 8x P/ARR, below the multiple implied by Z.AI's target price.

  Risks

  Its fundraising or IPO could reduce Z.AI's scarcity premium.

• Moonshot / Kimi K3

  Competitive and technology signal

  Strengths

  Kimi K3's parameter scale increased to 2.8trn, showing rapid capability improvement in Chinese open-weight models.

  Weaknesses

  K3 token throughput is slower than peers, and it paused new to-C subscriptions two days after launch, indicating compute constraints.

  Comparison

  The launch of Kimi K3 increases pressure on subsequent GLM model iterations at Z.AI.

  Risks

  If Kimi later raises funding or IPOs, it could weaken Z.AI's scarcity valuation.

• MiniMax、DeepSeek、OpenAI、Google、Anthropic、SpaceXAI、Meta

  Competitors and industry references

  Strengths

  Frontier lab model update frequency is accelerating, pushing overall industry capability higher.

  Weaknesses

  Intensifying competition makes leadership advantages more short-lived and may trigger pricing pressure.

  Comparison

  Multiple leading Chinese and U.S. players released model updates in July 2026, and the industry has entered a phase of high-frequency iteration.

  Risks

  New model launches may dilute Z.AI token usage growth and compress valuation multiples.

Key data

• RatingHoldHSBC maintains a Hold rating.
• Target priceHKD1,500.00Cut from HKD1,900.00.
• Current share priceHKD1,281.00As of the close on 2026-07-27.
• Implied upside+17.1%Upside from the target price relative to the current share price.
• December 2026 ARR forecastUSD2bnRaised from USD1bn.
• December 2027 ARR forecastUSD6.5bnRaised from USD2.4bn.
• 2026e revenueCNY7,206mNew forecast implies 895% YoY growth.
• 2027e revenueCNY29,176mNew forecast implies 305% YoY growth.
• 2028e revenueCNY66,237mNew forecast implies 127% YoY growth.
• 2027e non-IFRS EPSRMB3.72Previous forecast was RMB-4.46.
• Market capitalizationHKD586,584m / USD74,796mReport market data.
• 3-month average daily turnoverUSD648mReport market data.

Impact & implications

The implication of the report for Z.AI is a steeper growth curve and an earlier profitability inflection point, but also clearer valuation constraints. The ARR upgrade shows strong demand and commercialization, while compute investment and domestic-chip inference capability may strengthen long-term competitiveness; however, rapid model iteration, price competition, declining financing scarcity, and lock-up expiry may pressure valuation multiples, so HSBC chooses to maintain Hold rather than upgrade the rating.

Risks

• Competitors launch new models, diluting Z.AI's token usage growth.
• Fundraising or IPOs by competitors such as Anthropic, OpenAI, and Kimi reduce Z.AI's scarcity premium.
• About 38% of shares may come off lock-up in early January 2027, bringing supply pressure.
• Model performance after iteration is weaker than expected.
• Progress toward breakeven is slower than expected.
• High cash burn and refinancing needs.
• Chinese frontier models may face geopolitical risks such as bans in the U.S. market.
• A higher API revenue mix dilutes margins.

What to watch

• The launch and capability performance of the new GLM model in 3Q26.
• Whether Z.AI's ARR reaches USD2bn in December 2026 as expected.
• Whether token usage trends on channels such as OpenRouter continue improving after the release of GLM-5.2 or subsequent models.
• The actual impact of the 1GW data center and domestic-chip inference capability on throughput, cost, and the ability to absorb user demand.
• The iteration speed of Kimi, MiniMax, DeepSeek, and leading U.S. model companies.
• The impact of API price competition on gross margin and FCF.
• Share supply pressure around the lock-up expiry on or around 2027-01-07.
• The impact of frontier AI company fundraising or IPOs on the scarcity premium.