Z.AI CO., LTD. (02513) Report Interpretation

Overview

This company update examines how Z.ai's raised ARR guidance could be supported by newly funded compute capacity, cloud-partner distribution and Cowork's expansion into enterprise workflows. Morgan Stanley reiterates Overweight and retains a HK$1,800 target price.

Core views

Z.ai reported approximately US$1.8bn of ARR by mid-September and raised its 2026 year-end ARR guidance from US$2.4bn to US$3.0bn. Morgan Stanley attributes the increase to three commercialization engines: coding remains the near-term driver as ZCode users grow after capacity restrictions were lifted; cloud revenue-sharing is expected to start contributing in 4Q26; and Cowork can access existing enterprise budget pools with shorter commercialization cycles. The central question, in the report's view, has shifted from whether demand exists to how rapidly additional compute, partner distribution and vertical workflows translate into recognized revenue and sustainable gross profit. The report frames the US$5bn, or roughly Rmb30bn, funding round as a visible bridge from capital to compute capacity and monetization. Management's framework indicates that the funding could support about 94,000 GPUs, with roughly 40% allocated to training and 60% to inference. The approximately 56,000-GPU inference pool implies a theoretical annual revenue ceiling of about Rmb41bn, or US$6bn, before allowing for actual utilization and pricing realization, while theoretical inference gross margins could reach up to about 80%. The training allocation would support about Rmb3bn of annualized training spending over four years, which management believes is enough for multiple GLM-5.3-class or larger training runs. Morgan Stanley notes that a compute bottleneck constrained prior demand conversion: when demand for GLM-5.2 surged in June, Z.ai's own infrastructure could serve only about 10% of token demand, while third parties deploying its open-weight models fulfilled the remaining approximately 90%. The new capacity is therefore expected to support both larger-model training and materially greater inference throughput. This link between incremental GPU supply and usable inference capacity is fundamental to the report's growth case. Cloud partnerships provide a second route to scale without Z.ai funding all inference capacity itself. Z.ai has reached revenue-sharing arrangements with leading domestic and overseas cloud vendors under which GLM open-weight models will be offered through partner infrastructure; management expects revenue contribution from October. Morgan Stanley argues that this structure can improve capital efficiency and extend enterprise distribution through hyperscalers, while demonstrating that open-weight distribution can coexist with monetization through managed APIs and revenue sharing. Cowork is the third driver. The product is expanding beyond coding, with cybersecurity identified as the first vertical to scale. More than 100 cybersecurity companies have integrated GLM models, and orders exceeded Rmb1bn within one month of GLM-5.3's launch. Management sees a similar commercialization pattern potentially extending to legal, financial, industrial and life-science workflows. The report also identifies a global state-of-the-art model launch, overseas cloud-service-provider cooperation and easing market competition as potential upside factors, while computing constraints, model performance lagging peers and geopolitical risk remain downside considerations. Morgan Stanley values Z.ai using DCF, assuming a 15% WACC and 3% terminal growth rate. Its HK$1,800 target price implies 35x 2027e price-to-sales, and the report reiterates Overweight.

Analysis framework

Morgan Stanley links management's ARR guidance to the operating drivers behind it: coding demand, available training and inference compute, cloud-partner distribution, and Cowork's vertical adoption. It then translates funding into an indicative GPU mix, theoretical inference capacity and potential revenue economics, while valuing the company with a DCF framework and a 2027e price-to-sales implication.

Methodology notes

• Valuation methodsDCF (Discounted Cash Flow)

  Discounted cash flow valuation

  Morgan Stanley values Z.ai using DCF with a 15% weighted average cost of capital and 3% terminal growth assumption.

• Valuation methodsPS valuation

  Price-to-sales valuation cross-check

  The report states that its HK$1,800 target price implies a 35x 2027e price-to-sales multiple.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  Funding-to-compute-to-monetization transmission

  The report traces how funding could finance GPUs, increase training and inference capacity, relieve supply constraints, and support revenue generation through direct and partner channels.

Asset mapping & comparison

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

• Z.AI CO., LTD. (02513.HK)

  Primary covered company; higher ARR guidance is linked to coding growth, cloud revenue sharing and Cowork commercialization.

  Strengths

  Rapid ZCode user growth after capacity restrictions eased, potential expansion of inference capacity, cloud-partner distribution, and early cybersecurity adoption.

  Weaknesses

  Its infrastructure served only about 10% of GLM-5.2 token demand during the June demand surge.

  Comparison

  The report flags the risk that model performance could lag peers but provides no detailed peer comparison.

  Risks

  Computing constraints, potential model-performance lag versus peers, and geopolitical risk.

Key data

• ARR as of mid-SeptemberUS$1.8bnReported monthly ARR level.
• 2026 year-end ARR guidanceUS$3.0bnRaised from US$2.4bn.
• Funding roundUS$5bn (~Rmb30bn)Management framework suggests this could support approximately 94,000 GPUs.
• Inference GPU allocation~56,000 GPUsAbout 60% of the potential GPU fleet; theoretical annual revenue ceiling of about Rmb41bn (US$6bn) before utilization and pricing adjustments.
• Theoretical inference gross marginUp to ~80%Theoretical level before actual utilization and pricing realization.
• Cybersecurity orders>Rmb1bnGenerated within one month of GLM-5.3's launch, with more than 100 cybersecurity companies integrating GLM models.
• Target priceHK$1,800.00Implies 35x 2027e price-to-sales under the report's valuation framework.

Impact & implications

The report argues that Z.ai's higher ARR target is increasingly supported by identifiable monetization channels rather than demand alone. Additional compute may raise inference throughput, cloud partners may broaden enterprise reach with less self-funded capacity, and Cowork's vertical deployments may convert existing enterprise budgets more quickly.

Risks

• Computing constraints could limit the conversion of demand into inference revenue.
• Model performance could lag peers.
• Geopolitical risk could affect the business.

What to watch

• Whether cloud-vendor revenue-sharing begins contributing from October and scales through 4Q26.
• The pace at which incremental compute converts into higher inference throughput, revenue and sustainable gross profit.
• Further Cowork adoption beyond cybersecurity in legal, financial, industrial and life-science workflows.
• Progress on a global state-of-the-art model launch, overseas cloud-service-provider cooperation and market competition.