HSBC initiates coverage of Knowledge Atlas / Zhipu (2513.HK) with a Hold rating and a HKD920 target price

Overview

This is HSBC's initiation coverage report on Knowledge Atlas / Zhipu (2513.HK). The core view is that global demand for AI coding, agents, and foundation-model APIs may grow exponentially. Zhipu has a solid growth base thanks to the GLM model family, talent and technology links to Tsinghua University, high value for money, and a MaaS monetization path; however, the share price has risen sharply since the IPO, and at roughly 126x 2026e PS, valuation is already fairly in line with historical P/ARR frameworks for OpenAI and Anthropic, so HSBC assigns a Hold rating.

Core views

HSBC believes Zhipu's strengths lie in model capability, coding use cases, price-performance ratio, and monetization flexibility. Open Platform and API are expected to be the main revenue drivers in the future, benefiting from AI coding, OpenClaw, AutoClaw, Kilo Code, and cloud-service-provider channels. On-premise deployment serves large enterprises, state-owned enterprises, government-related clients, and Belt and Road markets, providing more stable revenue and real-world feedback. The report also notes that compute bottlenecks, competing models, lock-up expiries, cash burn, and geopolitical factors will limit further upside in the valuation.

Analysis framework

The report combines industry demand estimates, company segment breakdowns, model capability comparisons, revenue forecasts, cost-efficiency assessment, DCF valuation, and comparable-company P/ARR frameworks. HSBC compares Zhipu with OpenAI, Anthropic, and MiniMax, focusing on ARR, token usage, API pricing, AI coding TAM, model iteration capability, changes in training and inference costs, and the sensitivity of valuation to market-share assumptions.

Methodology notes

• Valuation methodsDCF

  10-year discounted cash flow model

  The HKD920.00 target price comes from a 10-year DCF model, with key assumptions of a 12.2% WACC and a 3% perpetual growth rate; this method is used to reflect the company's long-term growth potential.

• Valuation methodsP/ARR comparable analysis

  Comparable valuation based on historical OpenAI and Anthropic valuations and ARR

  The report looks back at the valuation histories and ARR trajectories of OpenAI and Anthropic and applies implied P/ARR to assess Zhipu. This approach indicates that Zhipu's valuation is about USD36bn or HKD624 per share, but it has limitations, including the absence of a private-company liquidity discount and heavy reliance on out-of-sample extrapolation.

• industry_analysisAI coding TAM and token growth

  AI coding and agent growth drive token consumption growth

  The report cites IDC and other data, arguing that the number of global AI agents, action counts, and token consumption could grow rapidly through 2030e, with AI coding and agentic coding serving as important drivers of API usage.

• company_forecastRevenue segment CAGR forecast

  Revenue growth split by cloud deployment and on-premise deployment

  HSBC expects cloud deployment to be driven by API and coding demand, with revenue CAGR of about 301% in 2025-2028e; on-premise deployment, supported by security, compliance, and customization use cases, is expected to post about 70% revenue CAGR over the same period.

• technology_assessmentModel capability and cost-efficiency review

  Model architecture, training efficiency, and inference cost analysis

  The report discusses technologies such as MoE, DeepSeek Sparse Attention, asynchronous RL infrastructure, and multi-chip adaptation, and argues that these capabilities help reduce training and inference costs while supporting GLM-5.0's performance in coding and long-horizon agent tasks.

Asset mapping & comparison

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

• 2513.HK

  Covered company stock

  Strengths

  Strong AI coding capability, clear MaaS monetization path, simultaneous API price increases and token growth, high R&D headcount share, and a standout Tsinghua-related technology and talent background.

  Weaknesses

  Current valuation already reflects growth expectations fairly fully, while compute bottlenecks, cash burn, refinancing needs, and a later path to profitability remain concerns.

  Comparison

  The report believes Zhipu faces less valuation pressure than MiniMax and benefits more directly from coding-model demand and API pricing power; versus OpenAI and Anthropic, valuation is benchmarked against their P/ARR history, but Zhipu still differs in scale, liquidity, and competitive environment.

  Risks

  Competitive model shocks, rival IPOs, lock-up expiries, weaker-than-expected model iteration, price wars, heavy cash burn, geopolitical tensions, and changes in government AI policy.

• MiniMax (100 HK)

  Comparable company

  Strengths

  Also has global AI model commercialization potential.

  Weaknesses

  Part of its growth depends on video generation and companion-type applications, which may face IP, inappropriate content generation, and competition from ByteDance Seedance and Kuaishou Kling.

  Comparison

  HSBC is more positive on Zhipu because its valuation is relatively less demanding and API pricing plus token growth have validated model capability.

  Risks

  Lock-up expiry pressure, competition in video generation and companion businesses, and compliance risk.

• OpenAI / Anthropic

  Global comparable valuation and demand validation sample

  Strengths

  Anthropic's rapid growth in coding-related revenue validates the huge global demand for AI coding and agentic engineering.

  Weaknesses

  As private companies, their comparable valuations are subject to liquidity and sample-extrapolation limitations.

  Comparison

  The report uses OpenAI and Anthropic's P/ARR history to assess whether Zhipu's current valuation is reasonable.

  Risks

  If these companies IPO or reprice, Zhipu's scarcity premium could be compressed.

Key data

• RatingHoldHSBC's initiation coverage assigns a Hold rating.
• Target priceHKD920.00Based on a 10-year DCF, 12.2% WACC, and 3% perpetual growth.
• Current share priceHKD915.00The chart shows the current price at around HKD915.00.
• Implied upside0.5%Limited upside versus current price.
• Valuation multiple127x/50x 2026e/2027e PSThe price-to-sales assumptions underlying the target price.
• 2026e/2027e/2028e revenue forecastUSD413m / USD1.0bn / USD2.2bnDCF base revenue assumptions.
• Open Platform and API ARRUSD250m in Mar-26In Mar-26.
• API pricing change+83%The report says the company was able to raise prices without affecting daily paid token growth.
• Daily paid token growth4xOccurred alongside the API price increase, showing demand resilience.
• GLM Coding Plan users and usagetoken usage up 15x, paying users above 240kAs of 2026-03-31.
• Claw Plan users400k paying users within 20 days of launchLaunched in Mar-2026.
• MaaS registered users4mAs of Mar-2026.
• R&D team657 people, 74% of total employeesAs of 2025-06-30.
• Expected profitability timing2029eHSBC expects the company to turn profitable in 2029e.
• Training cost as % of revenue325% in 2025 to 37% in 2028eThe report expects a significant improvement in training-cost efficiency.
• Inference cost as % of revenue38% in 2025 to 57% in 2028eThe share of inference costs rises as API usage expands.

Impact & implications

For investors, Zhipu is a rare pure-play large-model and AI coding name in Hong Kong equities, with high growth and monetization flexibility, but the near-term investment case depends more on valuation and execution delivery. If the company resolves compute bottlenecks, continues to iterate GLM models, and expands API usage, the profitability path and revenue market share could be revised up. If competing models launch quickly, industry price competition intensifies, or rival IPOs reduce the scarcity premium, the valuation could rerate lower.

Risks

• Competing companies launch new models, diluting Zhipu's token usage growth.
• Anthropic, OpenAI, Stepfun, Moonshot, and others IPO, which could reduce Zhipu's scarcity premium and trigger a valuation multiple reset.
• Lock-up periods end on 2026-07-07 and 2027-01-07, which could create supply pressure.
• Model iteration performs worse than expected.
• The pace of reaching breakeven is slower than expected.
• Cash burn remains heavy and refinancing needs emerge.
• Price competition intensifies, squeezing API and MaaS monetization space.
• Geopolitical tensions may affect service sales, chip access, or equity holdings.
• Changes in government policy toward AI.
• Customer concentration risk.

What to watch

• Whether GLM-5.5 or later model iterations are launched on schedule and deliver capability improvements.
• Whether compute bottlenecks ease, especially supply of training and inference resources.
• ARR, daily paid tokens, paying users, and pricing changes for Open Platform and API.
• User retention and monetization for AI coding, OpenClaw, AutoClaw, and Claw Plan products.
• Order growth in on-premise deployment from large enterprises, state-owned enterprises, government clients, and Belt and Road markets.
• Whether training costs and inference costs as a share of revenue improve as expected.
• Whether the company conducts more financing in 2027e-2028e, and the terms of that financing.
• Progress on inclusion in Stock Connect.
• Share price and trading-volume reactions after the 2026-07 and 2027-01 lock-up expiries.
• Valuation, IPO, and product-competition dynamics among OpenAI, Anthropic, MiniMax, and other peers.