Z.AI's first-half results were mixed: ARR ramped up rapidly, but the cloud transition and high R&D investment weighed on revenue and profit

Overview

This report reviews Z.AI's 1H26 results and analyzes its ARR growth, commercialization strategy, model roadmap, compute expansion, and path to profitability. Goldman Sachs recognizes the acceleration in MaaS demand, pricing power, and enterprise penetration, but believes differences in revenue recognition during the cloud transition, the decline in the on-premise deployment business, and persistently high R&D investment keep the risk-reward balanced.

Core views

1H26 results showed clear structural divergence. The company reported revenue of RMB954mn, up 400% YoY but 47% below Goldman Sachs' forecast. Open Platform/API revenue reached RMB825mn, up 2736% YoY and 412% HoH, 4% above consensus but 38% below Goldman Sachs' forecast. On-premise deployment revenue was only RMB129mn, down 20% YoY and 65% HoH, and 73% below Goldman Sachs' forecast. Goldman Sachs believes the pressure on overall revenue and profitability primarily resulted from the company's transition from on-premise deployment to cloud MaaS rather than weaker API demand itself. Earnings quality was also mixed. Open Platform/API gross margin reached 24.6%, above Goldman Sachs' 24.0% forecast and 2H25's 22.4%. Despite rising compute costs, improved pricing and inference efficiency continued to support the unit economics of this business. On the other hand, profitability from on-premise deployment was weaker than expected, reducing consolidated gross margin to 26%, below Goldman Sachs' previous estimate of 32% but still above MiniMax's 18%. 1H26 R&D expenses were RMB2.131bn, up 34% YoY and broadly in line with expectations; adjusted net loss was RMB1.964bn. The report believes gross profit generated by the inference business can already partially fund model R&D, but the company remains some distance from achieving clear overall profitability. ARR is the growth metric receiving the greatest attention in the report. ARR rose from US$250mn in March 2026 to US$1bn in July and then reached US$1.6bn in August; it was approximately US$2bn annualized based on the latest week's run rate, all on a gross basis including certain channel revenue shares. Growth was driven by both volume and pricing: year-to-date MaaS Token consumption increased 40-fold, Coding Plan usage increased 23-fold, and the average API selling price rose 101%. Enterprise adoption also deepened significantly, with average daily usage by the top ten customers increasing 98-fold from the beginning of the year. GLM has become a major model provider globally for four of China's leading internet platforms. Goldman Sachs notes that the gap between ARR and 1H26 reported revenue mainly stems from two factors. First, ARR includes certain channel revenue shares, while accounting revenue differs depending on whether it is recognized on a gross or net basis. Second, ARR is forward-looking, while the current growth cycle accelerated rapidly primarily after midyear and has not yet been fully reflected in first-half revenue. Based on continued increases in Token usage and pricing, Goldman Sachs forecasts 2H26 revenue of approximately RMB7.0bn and FY26 revenue of approximately RMB7.9bn, versus previous forecasts of RMB6.8bn and RMB8.6bn, respectively. It also raised its end-2026 ARR forecast from US$2.5bn to US$2.7bn, above its US$1.2bn forecast for MiniMax. Regarding commercialization, management emphasized that its objective is to capture higher-value Token demand rather than simply pursue total Token volume. Coding remains the primary entry point, but the product scope will expand into Agent, co-work, and autonomous AI workflows. As model capabilities improve, the charging model is expected to gradually shift from API consumption and subscriptions toward delivery based on task outcomes. The company plans to release a model approximately every 2 to 3 months to maintain its leading position in open-source models. Accordingly, Goldman Sachs raised its 2H26 and FY26 Open Platform gross-margin forecasts to 25.3% and 25.2%, respectively, from 23.4% and 23.5%. Supporting factors include pricing power, improved inference efficiency following model releases, and operating leverage from expanding Token scale. Regarding the model roadmap, the company believes scaling remains the core path to improving intelligence and requires simultaneous optimization of pre-training scale, effective depth, training depth, and task environments. Citing GLM-5.2 and GLM-5.3 as examples, management believes post-training currently offers relatively high returns as a scaling method. GLM-5.3 Flash further extends the frontier between intelligence and cost through a new architecture. The next stage will advance larger foundation models, longer native context windows, and native multimodal capabilities, followed by Fully Self Training, under which AI recursively helps generate and improve the data, environments, and infrastructure required for future training. Compute expansion and efficiency optimization are constraints on achieving the above roadmap. The company is expanding supply through self-operated clusters, leased capacity, and cloud-service procurement. A cluster containing more than 100,000 domestic chips already supports large-scale inference, while inference cost per Token has declined 80% year to date. In an environment where compute remains constrained, the company continues to optimize its software, networks, and inference stack. API revenue generated per RMB1 of compute cost, defined as the compute commercialization multiple, increased 14-fold YoY. Goldman Sachs raised its 2H26 and FY26 R&D expense forecasts to RMB4.3bn and RMB6.4bn, respectively, from RMB4.1bn and RMB6.2bn, with approximately two-thirds expected to relate to compute investment. The HK$31bn equity placement previously completed by the company will fund larger-scale model training and continuously growing compute demand. After incorporating the 1H26 results, Goldman Sachs lowered its 2026–2028 revenue forecasts by 8%, 4%, and 2%, respectively, to RMB7.921bn, RMB23.713bn, and RMB56.521bn. The adjustments came almost entirely from cuts to on-premise deployment revenue: the 2026–2028 forecasts were reduced to RMB456mn, RMB625mn, and RMB861mn, respectively, 59%, 66%, and 68% below the previous forecasts. Open Platform/API revenue forecasts remained broadly unchanged at RMB7.466bn, RMB23.088bn, and RMB55.660bn, respectively. Because first-half losses were higher than expected, the 2026–2028 earnings forecasts were lowered by 2%, 4%, and 2%, respectively, corresponding to adjusted net losses of approximately RMB4.618bn, RMB5.262bn, and RMB1.381bn. Regarding valuation, Goldman Sachs maintained its 12-month target price of HK$1,610 and its sector-relative Neutral rating. The target price is based on a DCF valuation assuming a 12% weighted average cost of capital, a 2% terminal growth rate, a 20% market share by 2030, and a 26% long-term adjusted EBIT margin by 2035. The implied bull- and bear-case valuations were slightly revised to HK$2,508 and HK$839, respectively. Goldman Sachs believes strong ARR and commercialization potential are balanced by foundation-model competition, high R&D investment, cash burn, and geopolitical risks.

Analysis framework

Goldman Sachs first compares the actual 1H26 results with its own forecasts and market consensus, separating the effects of Open Platform/API and on-premise deployment on revenue, gross profit, and losses. It then explains the growth mismatch through differences in accounting treatment and timing between ARR and reported revenue, and assesses commercialization trends based on Token usage, API pricing, customer adoption, and output per unit of compute. The report then evaluates the product and model roadmap, compute supply, and R&D investment, revises its 2026–2028 forecasts accordingly, and finally tests the target price and risk-reward through DCF and bull/bear scenarios.

Methodology notes

• Event-driven analysis and behavioral financeExpectation gap/expectation management

  Comparison of actual results with Goldman Sachs' forecasts and market consensus

  The report compares actual 1H26 revenue, gross margin, R&D spending, and losses with expectations on a line-by-line basis to identify areas that exceeded or fell short of expectations.

• Industry analysis frameworkVolume-price decomposition

  Decomposition of MaaS growth into Token usage and API pricing

  The report decomposes ARR and API revenue growth into expanding Token consumption and higher average API selling prices, indicating that growth came from both stronger demand and enhanced pricing power.

• Company fundamentals and financial framework

  ARR-to-reported-revenue bridge analysis

  The report explains why rapidly rising ARR has not yet been fully reflected in 1H26 reported revenue through gross-versus-net accounting differences, channel revenue sharing, and ARR's forward-looking nature.

• Valuation methodologyDCF discounted cash flow

  DCF valuation

  Goldman Sachs determines its 12-month target price by discounting future cash flows, using a 12% weighted average cost of capital and a 2% terminal growth rate while incorporating assumptions for market share in 2030 and margins in 2035.

• Valuation methodology

  Bull-, base-, and bear-case valuation

  The report applies different operating assumptions to derive implied bull-, base-, and bear-case valuations of HK$2,508, HK$1,610, and HK$839, presenting a valuation range under changes in key assumptions.

Asset mapping & comparison

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

• Z.AI Co. (02513.HK)

  The report's primary research subject, benefiting from stronger MaaS demand, API pricing, and enterprise adoption, but still undergoing a cloud transition and a period of high investment.

  Strengths

  Rapid ARR growth; simultaneous increases in Token usage and API pricing; improving Open Platform gross margin; deeper adoption by leading internet customers; lower inference costs and higher compute commercialization efficiency.

  Weaknesses

  On-premise deployment revenue and profitability were weaker than expected; consolidated gross margin was under pressure; R&D expenses and adjusted net losses remained high.

  Comparison

  Consolidated gross margin was 26%, above MiniMax's 18%; the end-2026 ARR forecast was US$2.7bn, above Goldman Sachs' US$1.2bn forecast for MiniMax; R&D expenses increased 34% YoY, below MiniMax's 139%.

  Risks

  Foundation-model competition, limited near-term earnings visibility due to high R&D investment, cash burn and self-funding capability, and geopolitical risks related to China-US technology competition.

• MiniMax Group

  A peer used in the report to compare ARR, gross margin, and R&D investment growth.

  Comparison

  The report states that MiniMax's consolidated gross margin was 18%, its R&D expenses increased 139% YoY, and Goldman Sachs' end-2026 ARR forecast for it was US$1.2bn, all of which were used for comparison with Z.AI.

Key data

• 1H26 revenueRMB954mnUp 400% YoY and 47% below Goldman Sachs' forecast
• 1H26 Open Platform/API revenueRMB825mnUp 2736% YoY and 412% HoH; 4% above consensus and 38% below Goldman Sachs' forecast
• 1H26 on-premise deployment revenueRMB129mnDown 20% YoY and 65% HoH, and 73% below Goldman Sachs' forecast
• August 2026 ARRUS$1.6bnApproximately US$2bn annualized based on the latest week's run rate, on a gross basis
• ARR ramp-up trajectoryUS$250mn／US$1bn／US$1.6bnCorresponding to March, July, and August 2026, respectively
• Token consumption growth+40X YTDYear-to-date usage growth in the MaaS business
• Coding Plan usage growth+23X YTDCoding remains the primary entry point for commercialization
• Average API selling price+101% YTDContributed to ARR growth together with usage growth
• Average daily usage by top ten customers+98XCompared with the beginning of 2026
• Open Platform/API gross margin24.6%Above Goldman Sachs' 24.0% forecast and 2H25's 22.4%
• Consolidated gross margin26%Below Goldman Sachs' previous estimate of 32% but above MiniMax's 18%
• 1H26 R&D expensesRMB2.131bnUp 34% YoY and broadly in line with Goldman Sachs' forecast
• 1H26 adjusted net lossRMB1.964bnGross profit from the inference business can partially support model investment
• End-2026 ARR forecastUS$2.7bnPreviously US$2.5bn; Goldman Sachs' forecast for MiniMax is US$1.2bn
• 2026–2028 revenue forecastsRMB7.921bn／23.713bn／56.521bnLowered by 8%, 4%, and 2%, respectively, from previous forecasts
• Inference cost per Token-80% YTDA cluster with more than 100,000 domestic chips already supports large-scale inference
• Compute commercialization multiple+14X YoYRefers to API revenue generated per RMB1 of compute expenditure
• 2H26／FY26 R&D expense forecastsRMB4.3bn／6.4bnPreviously RMB4.1bn／6.2bn, with approximately two-thirds related to compute investment
• 12-month target priceHK$1,610Unchanged; implied bull- and bear-case valuations are HK$2,508 and HK$839, respectively
• Key DCF parameters12% WACC／2% terminal growth rateAlso assumes a 20% market share in 2030 and a 26% long-term adjusted EBIT margin in 2035

Impact & implications

The report believes the simultaneous improvement in ARR, Token usage, API pricing, and enterprise customer adoption indicates that Z.AI's cloud MaaS commercialization is accelerating and is laying the foundation for an evolution toward Agent and task-outcome-based charging models. At the same time, contraction in the on-premise deployment business, the timing lag between revenue recognition and ARR, and higher compute and R&D investment mean near-term reported growth and profitability may continue to lag operating metrics. Therefore, while raising its ARR and Open Platform gross-margin forecasts, Goldman Sachs lowered its revenue and earnings forecasts and maintained its Neutral rating.

Risks

• Upside risk: Model intelligence exceeds expectations.
• Upside risk: A clear path to profitability is achieved faster than expected.
• Upside risk: Additional non-inference revenue-share sources emerge.
• Upside risk: Commercialization capabilities exceed expectations.
• Downside risk: Competition in the global foundation-model industry intensifies.
• Downside risk: High R&D expenses limit near-term earnings visibility.
• Downside risk: Cash burn and insufficient self-funding capability.
• Downside risk: Geopolitical uncertainty caused by intensifying China-US technology competition.

What to watch

• Track whether ARR can sustain its growth trajectory from US$250mn in March to US$1.6bn in August and reach the end-2026 forecast of US$2.7bn.
• Monitor whether Token usage, API pricing, and usage by leading enterprise customers can continue to grow together.
• Watch commercialization progress as the coding business expands into Agent, co-work, and autonomous AI workflows.
• Track model releases approximately every 2 to 3 months and progress in larger foundation models, long context, native multimodality, and Fully Self Training.
• Monitor whether inference cost per unit, the compute commercialization multiple, and Open Platform gross margin can continue to improve.
• Watch changes in losses, cash burn, and earnings visibility following increased R&D and compute investment.