Goldman Sachs Initiates Coverage on Knowledge Atlas Technology: Neutral, target price HK$1,880

Overview

This report is Goldman Sachs’ initiation coverage on Knowledge Atlas Technology (2513.HK), also known as Zhipu/Z.ai. The report views the company as one of China’s largest AI model companies by market value, with leading model capabilities in enterprise, coding, and agentic scenarios. Although GLM-5.2’s model performance, enterprise adoption, and global SME penetration are all improving, the stock has risen sharply since IPO and valuation is already at a relatively high level, so the report maintains a Neutral rating.

Core views

Core views include: first, Zhipu has a unique data flywheel in coding scenarios, where success/failure feedback from real engineering tasks is used for continuous model iteration; second, current disclosed ARR likely understates true demand because compute constraints create official API supply shortages, and underlying demand is viewed as 2-3x supply; third, if Zhipu eventually shifts from open source to an open-weight model with commercial terms, it may unlock more complete revenue and profit potential through third-party deployment sharing; fourth, short-term valuation already reflects leadership fairly, so the rating remains Neutral.

Analysis framework

The report uses a combined framework of company fundamentals, model capability benchmarks, API pricing, token consumption, cost efficiency, financing capacity, organizational and R&D efficiency, near-term agentic commercialization prospects, and DCF valuation, and compares against MiniMax, DeepSeek, Alibaba, ByteDance, and global AI model peers.

Methodology notes

• competitive positioningCompetitive Positioning framework

  pricing power, cost advantage, and financing strength

  Goldman Sachs measures Chinese foundation model players’ competitive position by pricing power, cost advantage, and financing strength, and rates Zhipu as one of the strongest-positioned Chinese players.

• Valuation methodsDCF

  12-month target price and scenario valuation

  The report derives a target price of HK$1,880 through DCF, corresponding to a base-case US$110bn valuation, and provides US$200bn bullish and US$54bn bearish cases.

• model capabilityArena AI / Code Arena / GDPval-AA

  coding and agentic task ranking

  The report cites GLM-5.2’s leading performance in coding and agentic benchmarks and emphasizes its relative strength among Chinese and open-source models.

Asset mapping & comparison

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

• Knowledge Atlas Technology (2513.HK)

  Core covered name

  Strengths

  Leading Chinese enterprise/coding AI model; GLM-5.2 near global frontier; coding data flywheel; improving API pricing power; relatively strong cash and financing capacity.

  Weaknesses

  Valuation is high; short-term earnings visibility is limited; R&D and training costs remain high; inference compute supply is still constrained.

  Comparison

  The report compares it with MiniMax, DeepSeek, Alibaba Qwen, ByteDance Doubao, and global AI model firms, and believes Zhipu is in the leading tier in enterprise coding scenarios and competitive positioning.

  Risks

  Global foundation model competition, compute and chip constraints, cash burn, self-developed/domestic chip adaptation, geopolitical and U.S.-China tech competition.

• MiniMax

  Peer listed AI model company comparison

  Strengths

  Provides a valuation reference point as a pure-play AI model company.

  Weaknesses

  The report estimates its ARR multiple is below Zhipu.

  Comparison

  Zhipu’s ARR multiple is higher than MiniMax in 2026E/2027E, reflecting a valuation premium for top-tier model companies.

  Risks

  Likewise faces foundation model competition and pressure to validate commercialization.

• Alibaba / ByteDance / US hyperscalers

  Potential distribution platforms and competitors/collaborators

  Strengths

  Have cloud infrastructure, enterprise customer base, and model deployment capabilities.

  Weaknesses

  Under an open-source model, Zhipu may not receive revenue sharing from third-party on-premise deployments.

  Comparison

  The report believes that if open-weight with commercial terms is adopted, third-party deployment could become a long-term asset-light revenue source for Zhipu.

  Risks

  Platform bargaining power, model substitution, and uncertainty around revenue-sharing mechanics.

Key data

• RatingNeutralInitiation coverage rating.
• 12-month target priceHK$1,880Based on DCF valuation.
• Current priceHK$2,032Price disclosed on report cover page.
• Implied upside/downside-7.5%Downside versus current price for the target price.
• Base valuationUS$110bnBase-case valuation in the report.
• Bullish/bearish valuationUS$200bn / US$54bnCorresponding to higher or lower long-term revenue share assumptions.
• Market capitalizationHK$946.1bn / US$120.7bnMarket data disclosed in the report.
• Enterprise valueHK$912.0bn / US$116.3bnEnterprise value disclosed in the report.
• API ARR at end-2026EUS$1.5bnAbove the company’s disclosed target of US$1bn.
• 2027E open-platform API revenueUS$3bnAbout double the estimated API ARR by end-2026E.
• 2026E/2027E/2028E revenueRMB6.3bn / RMB23.7bn / RMB58.8bnGoldman Sachs forecast.
• 2026E net casharound RMB32bnThe report believes it is sufficient to support ongoing R&D and operations.
• R&D headcount share70%+R&D staff comprise a majority of the approximately 1,000+ employees.

Impact & implications

The investment implication is that Zhipu has become an important valuation anchor among pure-play AI model stocks in the Hong Kong market; if GLM model capabilities continue to improve, API demand is released, domestic inference-chip supply improves, and open-weight commercialization is gradually implemented, long-term revenue optionality is substantial. In the near term, however, the stock has already largely reflected leadership in the valuation, and earnings visibility, R&D spending, and geopolitical technology competition remain valuation constraints.

Risks

• Upside risks include stronger-than-expected model intelligence, a faster and clearer path to profitability, non-inference revenue sharing income materializing, and commercialization strength exceeding expectations.
• Downside risks include intensified global foundation model competition, insufficient short-term earnings visibility due to high R&D spend, and pressure on cash burn and self-funding capability.
• Domestic and global chip and compute supply constraints could affect API revenue release and model training cadence.
• U.S.-China tech rivalry and geopolitical risks could affect models, chips, customers, and overseas expansion.
• The stock has risen sharply since IPO and has a relatively low free float, so changes in supply-demand structure could cause valuation volatility.

What to watch

• New model launches from Chinese AI model players in the second half of 2026, especially models with 2T+ parameter scales.
• The release cadence and performance of Zhipu GLM6 and subsequent models.
• Commercialization progress of coding and agent products such as ZCode, Claw Plan, AutoGLM, and AutoClaw.
• API token consumption, platform user count, pricing increases, and gross margin improvement.
• Domestic inference chip adaptation, ASIC/inference chip capacity ramp, and proprietary stack capabilities.
• Whether open-weight with commercial terms is implemented, and whether a third-party deployment revenue-sharing model takes shape.
• Progress of A-share STAR Market listing, use of H-share private placement funds, and changes in net cash.