HSBC is positive on AI-driven monetization in Chinese software and believes concerns that AI will replace software are overstated

Overview

This is a thematic report on China’s software industry. The core view is that rapid AI upgrades will not simply replace software companies, but rather enable them to create new monetization through AI features, multi-agent systems, AI hardware integration, and overseas expansion. HSBC believes that geopolitical and 'AI eats software' concerns are already largely reflected in valuations. The SWS software index is up only 2% year-to-date, below 5% for CSI 300, and trades at 3.6x TTM PS, below the 3.8x historical average.

Core views

The key points include: first, the performance gap between leading domestic and global large models has narrowed significantly, but falling model prices and limited differentiation make competition in the base model layer intense; second, application software and embedded AI software are better placed to benefit from lower-cost inference and enterprise demand; third, enterprise customers value security, compliance, business processes, and industry knowledge, where legacy software companies outperform younger AI startups; fourth, multi-agent systems, natural-language interaction, and capability modularization will shift enterprise software from seat-based charging to charging by workflow, task volume, or outcomes; fifth, hardware formats such as AI glasses and AI smartphones increase opportunities for embedded software and edge AI deployment; sixth, Chinese software overseas expansion remains early stage, but AI can serve as a product hook for software internationalization.

Analysis framework

The report uses a value-chain layered analysis, splitting opportunities into model layer, application software, enterprise services, multi-agent systems, AI consumer hardware, software infrastructure, and overseas applications, and supports the investment case with valuation metrics, AI order share of revenue, model API prices, user and token usage data, AI hardware shipment forecasts, and company case studies.

Methodology notes

• Industry chain analysisAI model layer and application layer segmentation

  Distinguishes competition and value capture between the foundational large model layer and downstream AI application layer.

  The report argues that large model performance is converging and prices are falling rapidly, intensifying competition at the model layer, while downstream software applications can benefit from lower inference costs and broader use cases.

• Business model analysisWaaS/RaaS

  Workflow-as-a-Service and Result-as-a-Service.

  Software vendors can shift from per-seat user pricing to charging by workflow, task volume, or business outcome, thereby reducing the negative impact of AI-driven efficiency gains on seat counts and opening new revenue opportunities.

• Market sizingTAM and CAGR forecasting

  Forecasts long-term market size using shipment volume, ASP, active agent counts, and task execution volumes.

  The report cites or estimates long-term growth in AI glasses, AI agents, and agent task execution as the basis for expansion in AI software and hardware ecosystems.

• Valuation comparisonTTM PS versus historical averages

  Uses SWS software index TTM PS versus its historical average to assess valuation level.

  The report notes the SWS software index is at 3.6x TTM PS, below the 3.8x historical average, and suggests relevant risks are partially priced.

Asset mapping & comparison

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

• Intsig

  An intelligent text recognition and AI application software candidate; the report prefers it and notes a concurrent Buy rating in the same report.

  Strengths

  It has multimodal recognition capabilities, has added AI features to products such as CamScanner, AI Translate, AI Chat, and CamExam/Quiz AI, and may develop new token or performance-based monetization through MCP or multi-agent system calls.

  Weaknesses

  Consumer willingness to pay may be affected by free product competition and macro conditions.

  Comparison

  Compared with model providers, application software firms are closer to user scenarios and vertical workflows.

  Risks

  AI function commercialization of user payment conversion, MAU growth, ARPPU improvement, and monetization may fall short of expectations.

• ArcSoft

  A target for imaging algorithms and AI hardware-embedded software; the report prefers it and notes a concurrent Buy rating in the same report.

  Strengths

  Benefits from higher penetration of AI features in smartphones, AI glasses, and other hardware.

  Weaknesses

  Dependent on downstream hardware shipment cycles and customer adoption pace.

  Comparison

  Compared with pure software subscriptions, AI features embedded in hardware are easier for consumers to accept as paid offerings.

  Risks

  AI hardware demand, ASP expansion, or algorithmic value capture may not materialize as expected.

• Kingsoft Office

  An office productivity software and AI application candidate with a Buy rating.

  Strengths

  WPS 365 collaboration has integrated the OpenClaw digital employee; WPS AI domestic MAU reached 80m, and token usage has grown rapidly.

  Weaknesses

  AI user engagement and monetization through paid conversion still require continued validation.

  Comparison

  Office software has high-frequency usage scenarios and a strong enterprise customer base, making it suitable for multi-agent system rollout.

  Risks

  Competition from free AI tools, enterprise budget pressure, and slower-than-expected AI commercialization.

• Kingdee

  An ERP and enterprise service AI suite candidate with a Buy rating.

  Strengths

  AI contract value reached RMB356m in 2025; management guided to RMB1bn AI suite revenue in 2026e, with a projected RMB2.2bn by 2028e.

  Weaknesses

  AI suite revenue growth depends on continued enterprise subscriptions and product maturity.

  Comparison

  Relative to model vendors, ERP providers have advantages in industry workflows, proprietary data, and customer stickiness.

  Risks

  Enterprise software spending slowdown, AI contract execution, and renewal performance may disappoint.

• Yonyou

  An ERP and enterprise service AI order growth candidate with a Buy rating.

  Strengths

  AI orders reached RMB1,670m in 2025, representing 18% of revenue, indicating strong enterprise demand; ARR growth accelerated.

  Weaknesses

  Conversion of AI orders into revenue and profit still needs to be monitored.

  Comparison

  Like other enterprise software providers, its advantage lies in complex enterprise workflows and customer trust.

  Risks

  AI feature maturity, subscription rollout, and enterprise budget support may fall short of expectations.

• Sangfor

  An AI infrastructure software candidate with a Buy rating.

  Strengths

  Benefits from growth in AI infrastructure and token demand.

  Weaknesses

  Sensitive to AI infrastructure investment cycles.

  Comparison

  Compared with application-layer companies, infrastructure software is more directly exposed to compute and token demand but may also face cost and competitive pressure.

  Risks

  Token costs, data center capex, and enterprise IT budget volatility.

• Baosight

  A data center operations related candidate with a Buy rating.

  Strengths

  Benefits from AI data center demand and China’s lower-cost power advantage in the western region.

  Weaknesses

  Capex and power resource constraints may affect expansion pace.

  Comparison

  Compared with pure software applications, data center operations are more infrastructure-oriented.

  Risks

  Power costs, utilization rate, policy, and capex risks.

• Thundersoft

  A candidate for edge AI deployment and AI hardware software with a Buy rating.

  Strengths

  Benefits from growth of edge AI devices such as AI smartphones and AI glasses.

  Weaknesses

  Dependent on terminal maker product cycles and shipment performance.

  Comparison

  Edge AI deployment sits at the intersection of hardware and software and is driven by rising AI hardware penetration.

  Risks

  AI hardware scale-up, customer concentration, and intensified competition may fall short of expectations.

Key data

• SWS software index year-to-date performance2%Below the CSI 300's 5% year-to-date gain.
• SWS software index valuation3.6x TTM PSBelow the 3.8x historical average.
• OpenAI GPT-4 series API price declineabout 90%Prices have declined sharply since launch, reflecting downward movement in inference costs.
• Leading China model API costsabout 20-30% of US modelsThe report believes domestic model prices are lower but relatively close to each other, making competition in the model layer intense.
• AI contract value as a share of revenue for China software companiesabout 7% in 2025The report says around 6% in 1H25 and around 7% in 2025.
• Yonyou 2025 AI ordersRMB1,670mRepresented 18% of total revenue, a clear increase from RMB320m in 1H25.
• Kingdee 2025 AI contract valueRMB356mApproximately 4% of total revenue, above a prior expectation of RMB310m.
• Kingdee 2028 AI suite revenue forecastRMB2.2bnAbout 23% of total revenue; AI contract value/AI suite revenue CAGR from 2025-28e is expected to be 85%.
• WPS AI domestic MAU80mAs of end-2025, up 307% year-on-year.
• WPS AI average daily token usageover 200bnMore than 12x year-over-year growth.
• Global active agent count forecastto reach 2.2bn by 2030eIDC estimates 2025-30e CAGR at 139%.
• Global AI agent task execution forecast415 tera by 2030eIDC and HSBC estimate 2025-30e CAGR at 524%.
• Global AI glasses shipment forecastto reach 90m units by 2030eWellsenn estimates 2025-30e CAGR at 64%.
• Global AI glasses TAM forecastRMB135bnCalculated based on RMB1,500 ASP and 90m units shipped.
• Doubao AI Phone priceRMB3,499Sold out on the first day; secondary-market premium of RMB300-3,500 suggests strong demand for AI-native smartphone experiences.

Impact & implications

The investment implication is that the market may be underestimating AI’s positive impact on software companies. If AI orders continue to grow, enterprises adopt multi-agent systems, WaaS/RaaS charging models are implemented, AI hardware shipments rise, and overseas expansion broadens, Chinese software companies could see revenue growth, margin improvement, and valuation rerating. The report is most constructive on software companies with industry process know-how, an enterprise customer base, proprietary data, security and compliance capabilities, and AI software that can be embedded into AI hardware.

Risks

• High token costs could suppress growth in agent demand.
• Current multi-agent systems are still immature, with potential errors and poor user experience.
• Cybersecurity risks such as data leaks could affect enterprise adoption of AI software.
• Competition among large models and AI applications may compress pricing and margins.
• Consumer willingness to pay for AI software may be weakened by free products from internet giants and a relatively weak macro environment.
• The pace at which AI orders convert into revenue, profit, and cash flow may be slower than expected.
• AI hardware shipments, ASP upgrades, or global expansion progress may underperform expectations.

What to watch

• Changes in Chinese software company AI orders, AI revenue mix, and renewal rates.
• The rollout speed, accuracy, stability, and security-compliance performance of enterprise-grade multi-agent systems.
• Changes in domestic and global large model API pricing and performance gaps.
• Actual commercialization progress of WaaS/RaaS pricing by workflow or outcome.
• Shipment volume, ASP, and user acceptance of AI glasses, AI smartphones, and other AI hardware.
• AI product usage, paid conversion, and revenue contribution of key companies such as Intsig, ArcSoft, Kingdee, Yonyou, and Kingsoft Office.
• Growth in overseas software application revenue for Chinese software firms and their ability to acquire international customers.