Z.AI Co. (02513) Report Interpretation

Overview

Goldman Sachs reviews Z.AI’s mixed 1H26 results: rapidly accelerating cloud-based MaaS and API monetization improved the ARR outlook, while the transition away from on-premise deployment, heavy R&D investment and continued losses constrained the near-term earnings outlook. The institution maintains Neutral and a HK$1,610 12-month target price.

Core views

Z.AI delivered a mixed 1H26 result. Open platform/API revenue was 4% above consensus, supported by stronger-than-expected monetization, while total revenue and profitability were held back by the continuing shift from on-premise deployment to cloud-based MaaS. First-half open platform/API revenue was Rmb825mn, up 2,736% year on year, whereas on-premise deployment revenue fell 20% year on year to Rmb129mn. Consolidated revenue reached Rmb954mn, up 400% year on year but 47% below Goldman Sachs estimates, in part because of gross-versus-net accounting differences and because ARR is a forward-looking measure during a steep ramp-up. Consolidated gross margin was 26%; API gross margin was 24.6%, above Goldman Sachs’ 24.0% forecast and 22.4% in 2H25, but weaker on-premise profitability offset part of that improvement. The central growth argument is the acceleration in recurring cloud demand. August ARR exceeded US$1.6bn, or US$2bn annualized from the latest weekly run-rate, after rising from US$250mn in March to US$1bn in July. Management attributed this momentum to MaaS volume growth, with token consumption up more than 40 times year to date and Coding Plan usage up 23 times, alongside API average selling prices up 101%. Enterprise engagement also deepened: daily usage among the top 10 customers increased 98 times from the start of the year, and GLM became the primary model supplier globally for four of China’s top internet platforms. Goldman Sachs therefore raises its 2026 year-end ARR estimate to US$2.7bn from US$2.5bn and forecasts 2H26/FY26E revenue of Rmb7.0bn/Rmb7.9bn, versus prior estimates of Rmb6.8bn/Rmb8.6bn. Management’s go-to-market strategy prioritizes higher-value token demand rather than maximizing aggregate token volume. Coding remains the entry point, while the company expands into Agent, Co-work and autonomous AI workflows. The report expects commercialization to progress from API consumption and subscriptions toward more outcome-oriented task delivery. Z.AI plans to sustain a roughly two-to-three-month model release cadence, and management sees larger foundation models, longer native context windows and native multimodality as the next development phase before Fully Self Training, a recursive framework in which AI helps improve future training data, environments and infrastructure. Goldman Sachs highlights GLM-5.2 and GLM-5.3 as evidence of post-training’s high return as a scaling lever, while GLM-5.3 Flash is presented as improving the intelligence-cost frontier through a new architecture and better efficiency. Compute scaling is both an operating requirement and a key profitability variable. Z.AI is expanding capacity through self-operated clusters, leased capacity and cloud-service procurement. Its 100k-plus domestic-chip cluster supports large-scale inference, and management cited an 80% year-to-date decline in unit token inference cost. The compute monetization multiplier—API revenue per Rmb1 of compute spending—improved 14 times year on year. Goldman Sachs expects open-platform gross margin of 25.3% in 2H26 and 25.2% in FY26E, up from prior forecasts of 23.4% and 23.5%, based on pricing power, inference-efficiency gains and operating leverage from rising token consumption. However, it raises R&D expense forecasts to Rmb4.3bn for 2H26 and Rmb6.4bn for FY26E, with about two-thirds related to compute investment; the earlier HK$31bn equity placement is expected to fund model training and expanding compute demand. Factoring in the first-half results, Goldman Sachs cuts 2026-28E revenue estimates by 8%/4%/2%, mainly due to lower on-premise deployment revenue while API revenue estimates remain broadly unchanged. Earnings estimates are reduced by 2%/4%/2% because first-half losses were larger than expected. First-half R&D expense was Rmb2.13bn, up 34% year on year, and adjusted net loss was Rmb1.96bn. The firm keeps its HK$1,610 12-month target price, based on a DCF assuming 20% market share by 2030E and a 26% long-term adjusted EBIT margin by 2035E. Its revised bull- and bear-case implied valuations are HK$2,508 and HK$839, respectively; Goldman Sachs maintains sector-relative Neutral on what it views as balanced risk-reward.

Analysis framework

Goldman Sachs compares 1H26 reported results with its estimates and consensus, then links ARR, token usage, pricing and enterprise adoption to its revenue outlook. It assesses margins through API economics, inference efficiency and compute spending, revises revenue and earnings forecasts, and values the company using a discounted cash flow framework with explicit long-term market-share and margin assumptions.

Methodology notes

• Valuation methodsDCF (Discounted Cash Flow)

  Discounted cash flow valuation

  Goldman Sachs derives its HK$1,610 target price using a DCF, discounting expected future economics under assumptions including 20% market share by 2030E, a 26% long-term adjusted EBIT margin by 2035E, a 12% WACC and 2% terminal growth rate.

• Industry AnalysisVolume-price decomposition

  Token-consumption volume and API pricing analysis

  The report explains the ARR and revenue outlook through both higher token usage and higher API average selling prices, separating demand volume from monetization per unit.

Asset mapping & comparison

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

• Z.AI Co. (2513.HK)

  Primary covered company; cloud-based MaaS, API monetization and enterprise adoption drive the growth thesis.

  Strengths

  Rapid ARR growth, higher API pricing, strong token consumption, deeper enterprise usage, improving inference efficiency and a growing role in Agent, Co-work and autonomous workflows.

  Weaknesses

  On-premise deployment revenue is declining, profitability remains limited and R&D and compute spending are high.

  Comparison

  API gross margin of 24.6% exceeded Goldman Sachs’ 24.0% estimate and was higher than MiniMax’s 18% consolidated gross margin cited in the report.

  Risks

  Global foundation-model competition, high R&D expense, cash burn, uncertain near-term profit visibility and geopolitical risk.

• MiniMax

  Comparable referenced in Goldman Sachs’ operating and ARR comparisons.

  Comparison

  Goldman Sachs cites MiniMax at 18% gross margin and compares its US$1.2bn ARR estimate with Z.AI’s US$2.7bn 2026 year-end ARR estimate.

Key data

• August ARROver US$1.6bnUS$2bn annualized based on the latest weekly run-rate, on a gross basis.
• ARR progressionUS$250mn in March; US$1bn in July; US$1.6bn in AugustIllustrates the accelerating ARR ramp-up.
• 2026 year-end ARR estimateUS$2.7bnRaised from US$2.5bn.
• 1H26 open platform/API revenueRmb825mnUp 2,736% year on year; 4% above consensus but 38% below Goldman Sachs estimates.
• 1H26 total revenueRmb954mnUp 400% year on year and 47% below Goldman Sachs estimates.
• 1H26 API gross margin24.6%Above Goldman Sachs’ 24.0% forecast and 22.4% in 2H25.
• 1H26 R&D expenseRmb2.13bnUp 34% year on year; adjusted net loss was Rmb1.96bn.
• Forecast revisions2026-28E revenue -8%/-4%/-2%; earnings -2%/-4%/-2%Primarily reflects lower on-premise deployment revenue and larger-than-expected first-half losses.
• Target priceHK$1,610Unchanged 12-month DCF-based target price.

Impact & implications

The report sees Z.AI’s cloud API and ARR momentum as validating a higher recurring-revenue outlook and supporting better open-platform margins. It also emphasizes that the transition comes with substantial compute and R&D requirements, lower near-term deployment revenue and continued uncertainty over the timing of sustainable profitability, leading Goldman Sachs to retain a Neutral stance.

Risks

• Model intelligence could be stronger than expected.
• The path to visible profitability could be faster than expected.
• Additional non-inference take-rate revenue streams or stronger commercialization could change the outlook.
• Competition in the global foundation-model industry could intensify.
• High R&D expense may limit near-term profit visibility.
• Cash burn and self-funding capability remain risks.
• Geopolitical risk could rise amid an intensified US-China technology race.