Model capabilities and compute investment are driving Z.AI's ARR expansion, but near-term earnings visibility remains limited

Overview

Goldman Sachs hosted the investor relations team of Zhipu/Z.AI at its 2026 Asia Leaders Conference in Hong Kong. The report centers on four themes: model capabilities, ARR trends, GLM-5.3 commercial licensing, and compute investment. It concludes that the foundation for growth remains strong, but high R&D expenses, cash burn, and limited near-term earnings visibility constrain a more positive rating.

Core views

The report first focuses on frontier model capabilities. The company plans to continue enhancing its foundation model capabilities, with an emphasis on post-training and long-horizon tasks. Coding remains the largest and most comprehensively covered enterprise use case, but the company believes its model capabilities can extend further into professional workflows such as cybersecurity, finance, and law. The company is developing workflow-oriented products such as Z.work and introducing domain data supplied by enterprise partners through industry collaborations to improve applicability in vertical scenarios. The second theme is rapid ARR growth and its drivers. The company stated that ARR reached US$1.6bn in August 2026 and US$2bn in the final week of August; management's year-end target is US$2.4bn, compared with Goldman Sachs' estimate of US$2.7bn. Demand comes primarily from China's internet industry, particularly large internet companies. One important reason these customers adopt GLM is that its capabilities outperform their internally developed models. Management believes that revenue can maintain strong growth as long as the company's model intelligence remains market-leading. However, current demand significantly exceeds available compute supply, meaning model capabilities determine the upper bound of demand, while compute supply determines the pace of near-term revenue realization. The third theme is the commercialization of GLM-5.3's open weights. The company will open-source certain models or model weights while retaining important technical details related to deployment and infrastructure optimization. GLM-5.3 departs from the company's previous approach by introducing licensing and revenue-sharing arrangements for commercial use, thereby creating a potential new revenue stream. Management believes the company understands its own model architecture better than hyperscale cloud service providers and may therefore achieve better margins by operating the models itself. The company is also negotiating commercial terms with major global cloud service providers, which could generate additional revenue if agreements are reached. The fourth theme is expanding compute capacity through capital investment. The company believes its self-developed compute clusters can deliver cost advantages through architectural optimization, infrastructure expertise, and more cost-efficient inference. Meanwhile, models with smaller parameter counts can be scaled through post-training to achieve the performance of larger models from competitors. The company completed a US$4bn follow-on financing in July 2026 and is using the proceeds to secure additional compute capacity in an effort to alleviate the bottleneck caused by demand far exceeding supply. The financial forecasts reflect a trajectory of high growth alongside continued losses. Goldman Sachs expects revenue to increase from Rmb724.3mn in 2025 to Rmb7,921.5mn in 2026, Rmb23,712.9mn in 2027, and Rmb56,520.8mn in 2028. EBITDA over the same period is projected at -Rmb3,505.7mn, -Rmb5,188.3mn, -Rmb5,824.7mn, and -Rmb2,158.8mn, respectively, while EPS is projected at -Rmb19.95, -Rmb10.29, -Rmb11.19, and -Rmb2.91. Revenue is expected to scale rapidly, but the company is still forecast to remain loss-making through at least 2028, consistent with the report's emphasis on limited near-term earnings visibility. On valuation, Goldman Sachs assigns a Neutral rating and derives a 12-month target price of HK$1,610.00 using DCF. Based on the September 1, 2026 closing price of HK$1,179.00, the report indicates upside of 36.6%. The target price history table also shows a target price of HK$1,610.00 on August 3, 2026, versus that day's closing price of HK$940.00, and a target price of HK$1,880.00 on July 9, 2026, versus that day's closing price of HK$2,032.00. The table states that none of the target prices were adjusted for corporate actions. The company's M&A Rank is 3, representing a relatively low 0%-15% probability of becoming an acquisition target under Goldman Sachs' framework; therefore, no material M&A component is incorporated into the target price.

Analysis framework

The report begins with management discussions at the conference and sequentially analyzes model capabilities, ARR and customer demand, GLM-5.3 commercial licensing, and compute investment, before using Goldman Sachs' forecasts to assess the trajectory of revenue growth and narrowing losses. Its core logic treats model competitiveness as the driver of demand and available compute capacity as the constraint on revenue realization. It compares management's US$2.4bn year-end ARR target with Goldman Sachs' US$2.7bn estimate and ultimately derives a 12-month target price through DCF. Financial and pricing data are sourced from company information, Goldman Sachs Research estimates, and FactSet.

Methodology notes

• Valuation MethodologyDiscounted Cash Flow (DCF)

  DCF Valuation

  The report uses a discounted cash flow methodology to derive a 12-month target price of HK$1,610.00 by discounting the expected value of the company's future cash flows to the present. The excerpt does not disclose the specific discount rate or terminal value assumptions.

• Industry Analysis FrameworkSupply-demand framework

  AI Demand and Compute Supply Constraints

  The report analyzes customer demand generated by model intelligence and limited compute supply within the same framework: model performance drives ARR, while insufficient compute capacity may constrain the pace at which demand is converted into revenue.

• Event Strategy and Behavioral FinanceEvent-driven analysis

  Management Discussions at the Asia Leaders Conference

  Using information disclosed by the company's investor relations team at the conference as the event basis, the report identifies the latest developments in model strategy, ARR, commercial licensing, and compute investment.

• Event Strategy and Behavioral Finance

  M&A Rank

  Goldman Sachs uses a scale of 1 to 3 to assess the likelihood that a company becomes an acquisition target. Z.AI is ranked 3, corresponding to a relatively low probability of 0%-15%, and no M&A component is included in the target price under this framework.

Asset mapping & comparison

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

• Z.AI Co (02513.HK)

  The sole primary company covered in the report; its model capabilities, commercialization model, and compute investment collectively affect ARR growth and DCF valuation.

  Strengths

  Its model capabilities outperform some customers' internally developed models; enterprise coding and professional workflows offer room for expansion; self-developed clusters, architectural optimization, and efficient inference may deliver cost advantages; GLM-5.3 commercial licensing may add revenue sources.

  Weaknesses

  Available compute capacity falls short of demand, while high R&D expenses limit near-term earnings visibility and create pressure from cash burn and constrained self-financing capacity.

  Comparison

  Management stated that smaller-scale models can be scaled through post-training to achieve the performance of competitors' larger models. One reason large Chinese internet customers adopt GLM is that its capabilities outperform their internally developed models.

  Risks

  Upside risks include stronger-than-expected model intelligence, earnings trajectory, non-inference revenue, and commercialization capabilities. Downside risks include competition in the global foundation model industry, high R&D expenses, cash burn, and geopolitical risks arising from intensified US-China technology competition.

Key data

• August 2026 ARRUS$1.6bnThe August ARR level disclosed by the company.
• ARR in the Final Week of AugustUS$2bnThe latest month-end run rate disclosed by the company.
• Year-End ARR TargetManagement: US$2.4bn; Goldman Sachs estimate: US$2.7bnManagement's target is below Goldman Sachs' estimate.
• Follow-on FinancingUS$4bnCompleted by the company in July 2026, with the proceeds being used to secure additional compute capacity.
• 12-Month Target PriceHK$1,610.00Based on DCF valuation.
• Reference Share PriceHK$1,179.00As of the September 1, 2026 close.
• Upside to Target Price36.6%The upside indicated in the report based on the reference share price.
• Market CapitalizationHK$549.0bn / US$70.0bnAs listed in the report's company data table.
• Enterprise ValueHK$513.8bn / US$65.5bnAs listed in the report's company data table.
• Three-Month Average Daily Trading ValueHK$7.3bn / US$929.6mnAs listed in the report's company data table.
• Revenue Forecast2025: Rmb724.3mn; 2026: Rmb7,921.5mn; 2027: Rmb23,712.9mn; 2028: Rmb56,520.8mnFigures for 2026 through 2028 are Goldman Sachs forecasts.
• EBITDA Forecast2025: -Rmb3,505.7mn; 2026: -Rmb5,188.3mn; 2027: -Rmb5,824.7mn; 2028: -Rmb2,158.8mnFigures for 2026 through 2028 are Goldman Sachs forecasts, with losses projected for all periods.
• EPS Forecast2025: -Rmb19.95; 2026: -Rmb10.29; 2027: -Rmb11.19; 2028: -Rmb2.91Figures for 2026 through 2028 are Goldman Sachs forecasts, with negative EPS projected for all periods.
• Target Price History2026-08-03: HK$1,610.00; 2026-07-09: HK$1,880.00The corresponding closing prices were HK$940.00 and HK$2,032.00, respectively; the target prices were not adjusted for corporate actions.
• M&A Rank3Corresponds to a relatively low 0%-15% probability of becoming an acquisition target under Goldman Sachs' framework, with no material M&A component included in the target price.

Impact & implications

The report believes Z.AI's growth depends on whether two capabilities can advance in tandem: model intelligence must remain market-leading to sustain demand from large internet customers, while compute expansion determines how quickly that demand can be converted into ARR. GLM-5.3 commercial licensing, revenue sharing, and negotiations with global cloud service providers could broaden revenue sources, but the financial forecasts still indicate that the company will remain loss-making through at least 2028. Therefore, despite the upside implied by the DCF-based target price, Goldman Sachs maintains a Neutral rating.

Risks

• Upside risk: Model intelligence is stronger than expected.
• Upside risk: A clear path to profitability is achieved faster than expected.
• Upside risk: More non-inference, revenue-sharing income sources emerge.
• Upside risk: Commercialization capabilities are stronger than expected.
• Downside risk: Competition in the global foundation model industry intensifies.
• Downside risk: High R&D expenses result in limited near-term earnings visibility.
• Downside risk: Cash burn and insufficient self-financing capacity.
• Downside risk: Geopolitical uncertainty arising from intensified US-China technology competition.