中国AI月度报告：软件领域AI颠覆担忧过重，尤其中国

China (PRC) | Technology China AI Monthly - Excess Fear of AIDisruption in Software, Esp for China We view AI's impact on SW as a "natural selection" event, rather than SaaS"apocalypse". Our extensive talks with 7 China SW players on AI reveallikelihood of more resilience to AI disruption than US peers, given China'sproject-based or module-based SaaS pricing. But IT budget pressure isthe key headwind. China SW U/P US peers (7%) and CSI 300 (18%) YTD,driven by earnings D/G and multiple compression. Our top pick in China SWremains Kingdee (KD). China Software underperforms (U/P) US on earnings downgrade and multiple compression.Since Oct 2025, the US software (SW) sector has sold off sharply on fears that AI will disruptSaaS. Anthropic's release of Claude Cowork on Jan 12, 2026 intensified the downturn throughMar 31 (~23% decline), before the market realized that disruption fears were overdone, driving~21% recovery since then.OurUS SW basket (14 stocks):has fallen 6.4% YTD purely due tomultiple contraction. Cons 2026 rev profit has risen 2% since Jan 12, while EV/S is down 6.5%YTD from 6.1x to 5.7x.OurChina (CN) SW basket (19 stocks):has fallen 13% YTD. Cons 2026rev has fallen 5% since Jan 12, and EV/S is down 15% YTD from 5.0x to 4.3x, suggesting thesell-off was driven by earnings downgrade (D/G) and multiple contraction. The rise of LLMs and AI agents is driving a structural shift in SW architecture andeconomics.We believe AI is NOT a SaaS "apocalypse", but a "natural selection mechanism", asplayers with defensible moats that proactively embrace AI can evolve into AI-native platforms,while those with weak moats and are slow to adopt AI risk displacement. Three key impacts:1) Architectural shift.SW will transition from UI-centric, human-operated systems to API-first, agent-executable workflow platforms. Agents can automate tasks autonomously butstill rely on SW APIs for workflows, business logic, and data.2) Pressure on seat-basedpricing.As agents replace human operators, per-seat pricing models face structural pressure,pushing vendors toward consumption- and outcome-based pricing.3) Margin pressure.Unliketraditional SW's near-zero marginal cost, LLM inference is a big usage-based variable cost,which means SW players will have to pass such costs on to customers or be able to chargea healthy margin. CN SW is less prone to AI disruption, but frail IT budgets remain the headwind.Many CNSW players operate on project-based models, serving large SOEs requiring heavy systemsintegration and customization. Data security concerns and asset-based mentality have limitedpublic cloud SaaS penetration with SOEs. Moreover, Chinese SaaS players rarely use seat-based pricing; instead, they charge by module. This makes CN SW generally more resilient toAI disruption than its US peers. However, the primary headwind remains IT budget pressurefor large enterprises in a weak macro environment. CN= China, HW= Hardware, KD=Kingdee,SW= Software, UI= User Interface KD still our top pick in China SW.We believe industrial SW, financial IT, and ERP are the mostresilient to AI disruption given deep workflow integration, proprietary data, zero tolerance forhallucination, and strict vendor qualifications. We view KD's recent sell-off as a good entrypoint: 1) KD's ERP is deeply embedded in clients' workflows as a system of record, and iscostly and risky to replace; 2) KD's proprietary data and industry know-how is difficult for AIvendors to replicate; 3) module-based pricing insulates rev from workforce reductions; 4) KDhas launched its own AI-native ERP suite and Lingee AIOS, guiding Rmb1.0bn in AI rev for 2026;5) SaaS now >50% of rev, drives teen rev growth, GPM expansion, and strong op leverage aidedby internal AI-driven cost efficiencies; 6) valuation at 0.7x 2026E PEG is highly attractive.Conton next page... Edison Lee, CFA * | Equity Analyst852 3743 8009 | edison.lee@jefferies.com Matt Ma * | Equity Analyst852 3767 1109 | matt.ma@jefferies.com Annie Ping, CFA, FRM * | Equity Associate+852 3767 1273 | annie.ping@jefferies.com Nick Cheng * | Equity Analyst+852 3743 8750 | nick.cheng@jefferies.com Jacky He * | Equity Analyst+852 3743 8084 | jacky.he@jefferies.com China's AI models narrow the performance gap with the US, while aggressive price cuts intensifycompetition and could pressure margins.In May, Anthropic's Claude Opus 4.8, OpenAI's GPT 5.5,and Google's Gemini 3.1 Pro remain the top three globally on AA's intelligence ranking. In China,leadership keeps shifting, Qwen 3.7 Max surpassed Kimi K2.6 as the best Chinese model. TheUS-China performance gap has slightly narrowed, with Qwen 3.7 Max now 8% below the top USmodel (vs. 10% in April; 18% in Dec 2025). Moreover, Chinese models continue to offer superiorvalue for performance, their avg price as a percentage of US model price fell to 21% in May (31%in Apr), driven by Qwen/Xiaomi price cuts on their flagship models by 46%/86%. DeepSeek alsoannounced the 75% promotional di