Z.ai raises ARR guidance; MaaS scaling and improved compute efficiency lift the price target to HK$1,800

Overview

The report reviews Z.ai's 1H26 results and management conference call, with the key change being that ARR, user activity, and customer usage were significantly above expectations. Morgan Stanley consequently raised its medium- to long-term revenue forecasts and price target, while noting that lower gross margins, higher R&D investment, and compute constraints remain key variables for near-term earnings and valuation realization.

Core views

The report believes the most important incremental takeaway from the results briefing was that commercialization is progressing significantly faster than previously expected. Management disclosed that ARR reached US$1.6bn on a monthly run-rate basis in August 2026 and exceeded US$2bn on a weekly run-rate basis, above Morgan Stanley's US$1.2bn expectation. The company raised its year-end ARR guidance to US$2.4bn, while Morgan Stanley also raised its own forecast from US$2bn to US$2.4bn. The report believes year-end ARR could rise further if compute supply is better than expected. Model capabilities remain the foundation for commercial expansion. Management positions GLM-5 as a model evolving from coding capabilities toward broader agent engineering, stating that GLM-5.1 and 5.2 lead domestic models in long-horizon tasks. GLM-5.3 further strengthens post-training capabilities, while GLM-5.3-Flash combines frontier capabilities, low cost, and support for domestic chips. The future roadmap will simultaneously advance larger foundation models, reasoning capabilities, Token-loop architectures, and intensive post-training. The planned GLM-6.0 is described as a “fully self-trained” model capable of self-evaluation and self-correction, and management regards it as an important milestone toward recursive self-improvement. Management expects potentially meaningful progress in model competitiveness over the next one to two months. Z.ai does not view coding merely as a commercial vertical, but defines it as a “meta-capability” for building professional agents. The company believes coding capabilities can be transferred to professional workflows such as cybersecurity, legal services, and finance. GLM-5.3's capabilities have already begun entering cybersecurity use cases and could further evolve into “cowork” digital colleagues that replicate complex workflows. The company estimates the total coding market at US$500bn and the total cowork market at US$5tn. Its market strategy is therefore expanding from “coding” to “cowork” and from a D2B model targeting developers and enterprises to B2C. Demand and monetization metrics support this thesis. From January to August 2026, Token consumption increased by more than 40x, including growth of more than 23x in coding-plan Tokens, while the average API price rose by 101% over the same period. The open platform had more than 7.4mn users, MaaS platform users increased by 144% year-to-date, and paying daily active users increased by 603%. Average daily Token usage among the top ten revenue-contributing customers grew by approximately 98x, or 9,798%. The company became the highest-priority global supplier to four of China's top ten internet companies, with average daily revenue increasing 234x over the past 12 months. After sales restrictions on APIs and coding plans were removed, commercial reach expanded further, and ZCode exceeded 1mn monthly active users in its first month after launch. The report also states that, as of the end of August, the numbers of customers with ARR above US$100k, US$500k, US$1mn, US$10mn, US$25mn, and US$250mn were approximately 115, 25, 37, 8, 2, and 2, respectively, increases of approximately 94, 20, 35, 7, 2, and 2 from the beginning of the year. Improved compute investment efficiency is the key transmission mechanism from high revenue growth to margin expansion. The compute multiplier in 2Q26, defined as API revenue divided by compute cost, increased 14x YoY to 0.46x, while unit inference cost declined by 80% in 1H26. Over the past six months, management has maintained a strategy of “comprehensive infrastructure investment,” generating scale benefits through multi-source compute procurement, increased deployment of domestic GPUs, model and network optimization, and potential self-built data centers, while also improving cache hit rates and Token throughput and reducing service costs. Management expects the open platform's full-year gross margin to have further upside and targets more than 50% over the next 12—18 months. Actual 1H26 revenue was Rmb954mn, up 400% YoY and 1% below Morgan Stanley's forecast. Gross profit was Rmb252mn, representing a gross margin of 26%, below its 32% forecast. Operating loss was Rmb2.161bn, 16% narrower than Morgan Stanley's forecast loss, while net loss attributable to shareholders was Rmb2.071bn, 20% narrower than forecast. After incorporating the results and higher ARR, Morgan Stanley raised its 2026, 2027, and 2028 revenue forecasts by 20.4%, 25.8%, and 29.3%, respectively; the revision table shows 25.6% for 2027. The new revenue forecasts are Rmb6.329bn, Rmb19.313bn, and Rmb47.641bn, respectively, implying growth of 773.8%, 205.2%, and 146.7%. Growth is driven almost entirely by cloud deployment: cloud revenue forecasts for 2026—2028 are Rmb5.902bn, Rmb18.886bn, and Rmb47.214bn, respectively, while on-premises deployment revenue remains Rmb427mn in each year. Higher revenue does not immediately translate into better near-term earnings forecasts because the report simultaneously lowers its gross-margin assumptions and raises its R&D expense assumptions. The forecast operating loss for 2026 widened by 4.5%, the forecast operating loss for 2027 narrowed by 35.2%, and the 2028 operating profit forecast increased by 489.1% to Rmb5.018bn. The non-IFRS loss-per-share forecast widened by 14.4% in 2026 and narrowed by 28.0% in 2027, while the 2028 non-IFRS EPS forecast increased by 122.6% to Rmb12.02. The report expects gross margin to rise from 31.3% in 2026 to 35.3% in 2027 and 40.1% in 2028 and expects the company to break even in 2028, reflecting revenue scale ultimately outpacing growth in R&D costs. For valuation, Morgan Stanley uses a DCF, with the 15% WACC and 3% perpetual growth rate unchanged. Upward revisions to long-term revenue offset wider near-term losses, driving a 5.9% increase in the price target from HK$1,700 to HK$1,800. Relative to the HK$1,195 closing price, this implies 51% upside. The price target corresponds to 35x 2027E P/S, below the historical one-year forward P/S of 176x. The report states that the company was then trading at 25x 2027E P/S and therefore considers the valuation undemanding. The bull-case scenario assumes the company launches a global SOTA model, continues raising API prices, and expands its cloud customer base, generating 2027 revenue of Rmb33.8bn and a valuation of HK$3,340 at 40x P/S. The base case assumes Rmb19.3bn of revenue and a valuation of HK$1,800 at 35x P/S. The bear case assumes that model performance lags and revenue grows only linearly, resulting in 2027 revenue of Rmb13.5bn and a valuation of HK$670 at 20x P/S. The main downside transmission channels are compute constraints, model performance lagging peers, and geopolitical risks.

Analysis framework

Morgan Stanley first uses actual 1H26 results and information from the management conference call to assess ARR, usage, customer growth, and compute efficiency, and then translates these operating metrics into forecasts for cloud revenue, gross margin, R&D expenditure, and break-even timing. The report subsequently updates its 2026—2028 financial forecasts through Morgan Stanley ModelWare, uses Refinitiv data as a reference for consensus and market valuation, and combines DCF, 2027 P/S, and bull—base—bear scenarios to derive its price target and risk-reward assessment.

Methodology notes

• Valuation MethodDCF Discounted Cash Flow

  DCF Valuation

  The report discounts forecast-period free cash flow and terminal value to present value using a 15% weighted average cost of capital and a 3% perpetual growth rate. After upward revisions to long-term revenue offset wider near-term losses, it derives a price target of HK$1,800.

• Valuation MethodPS valuation

  2027E P/S Cross-Check

  The report uses revenue multiples to validate the DCF result: the price target corresponds to 35x 2027E P/S, compared with the current trading level of approximately 25x, a historical one-year forward P/S of 176x, and global peer valuations.

• (Out-of-Vocabulary Method)

  Bull—Base—Bear Scenario Analysis

  Based on whether the model can reach global SOTA levels and whether cloud customer expansion and API price increases can generate nonlinear growth, the report sets 2027 revenue at Rmb33.8bn, Rmb19.3bn, and Rmb13.5bn under the respective scenarios and applies P/S multiples of 40x, 35x, and 20x.

Asset mapping & comparison

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

• Z.AI CO., LTD. (02513.HK)

  The report's sole primary research and rating subject, benefiting from foundation-model iteration, MaaS scaling, API price increases, and improved compute efficiency.

  Strengths

  ARR and Token usage significantly exceeded expectations, open-platform users exceeded 7.4mn, the GLM model series continues to iterate, unit inference cost declined by 80%, and the company has become the highest-priority global supplier to four of China's top ten internet companies.

  Weaknesses

  The 1H26 gross margin was only 26%, the company remains in a phase of substantial operating losses, and continued R&D and infrastructure investment limits the near-term conversion of revenue growth into earnings.

  Comparison

  Management states that GLM-5.1 and 5.2 are among the leading domestic models. The price target corresponds to 35x 2027E P/S, which the report considers consistent with global peer valuations and below the company's historical one-year forward P/S of 176x.

  Risks

  Constraints on compute supply, model performance lagging peers, and geopolitical risks could weaken cloud expansion, revenue growth, and valuation realization.

Key data

• August 2026 ARRUS$1.6bn on a monthly run-rate basis; more than US$2bn on a weekly run-rate basisAbove Morgan Stanley's US$1.2bn expectation
• Year-End ARR Guidance and ForecastUS$2.4bnThe company raised its guidance, while Morgan Stanley raised its forecast from US$2bn to US$2.4bn accordingly
• Tokens and API PricingToken growth of more than 40x; coding-plan Token growth of more than 23x; average API price up 101%From January to August 2026
• Platform UsersMore than 7.4mn open-platform users; MaaS users up 144%; paying daily active users up 603%The latter two figures are year-to-date growth rates
• Key Customer UsageApproximately 98x, or 9,798%Year-to-date growth in average daily Token usage among the top ten revenue-contributing customers
• Compute EconomicsCompute multiplier of 0.46x, up 14x YoY; unit inference cost down 80%The compute multiplier is a 2Q26 figure, while unit inference cost reflects the change in 1H26
• 1H26 RevenueRmb954mnUp 400% YoY and 1% below Morgan Stanley's forecast
• 1H26 Gross Margin26%Gross profit of Rmb252mn, below Morgan Stanley's 32% gross-margin forecast
• 1H26 LossesOperating loss of Rmb2.161bn; net loss attributable to shareholders of Rmb2.071bnThe losses were 16% and 20% narrower than Morgan Stanley's forecasts, respectively
• Revenue Forecast Upgrades2026E +20.4%; 2027E +25.8%; 2028E +29.3%The 2027E revision table shows +25.6%
• 2026—2028 Revenue ForecastsRmb6.329bn, Rmb19.313bn, Rmb47.641bnExpected growth rates are 773.8%, 205.2%, and 146.7%, respectively
• 2028 Earnings ForecastOperating profit of Rmb5.018bn; non-IFRS EPS of Rmb12.02The forecasts were raised by 489.1% and 122.6%, respectively, and the report expects the company to break even in 2028
• Price Target and Current PricePrice target HK$1,800; current price HK$1,195The price target was raised by 5.9%, implying 51% upside
• Risk-Reward ScenariosBull case HK$3,340; base case HK$1,800; bear case HK$670Corresponding to 40x, 35x, and 20x 2027E P/S, respectively

Impact & implications

The report believes that simultaneous growth in ARR, Token usage, paying users, and large-customer penetration indicates that Z.ai's MaaS commercialization is accelerating, with cloud revenue potentially accounting for nearly all incremental revenue over the next three years. Improved compute efficiency provides a path toward medium-term gross-margin expansion and break-even in 2028, but lower near-term gross margins and higher R&D investment mean that upward revenue revisions cannot immediately eliminate losses. The higher price target primarily depends on improvements in long-term revenue and cash flow.

Risks

• Compute supply constraints could limit Token delivery, ARR growth, and revenue upside.
• If model performance fails to keep pace with global SOTA and peer levels, 2027 revenue growth could deteriorate from nonlinear to linear growth.
• Geopolitical risks could affect the company's global expansion and access to infrastructure.

What to watch

• Whether year-end ARR can reach US$2.4bn and whether more ample compute supply can drive outperformance.
• Progress in model competitiveness over the next one to two months and execution of the GLM-5.3 and GLM-6.0 roadmaps.
• Whether the open platform's gross margin can improve toward the target of more than 50% over the next 12—18 months.
• Whether the compute multiplier, unit inference cost, cache hit rate, and Token throughput can continue to improve.
• Progress in user conversion from coding to cowork, from D2B to B2C, and for LLM+Harness products such as ZCode.
• Whether the company's cooperation with overseas cloud service providers and global expansion progress, and whether market competition eases.
• Whether cloud revenue growth can drive a continued narrowing of operating losses and enable break-even in 2028.