Erosion risks

• Shift of discovery/search to apps, social, and short video
• AI assistants reducing classic search queries
• Regulatory constraints on online marketing formats and targeting

Leading indicators

• Search engine share in China (mobile/desktop)
• Baidu Core online marketing revenue trend
• User engagement on Baidu App / search-plus-newsfeed

Counterarguments

• Advertisers can reallocate budgets to short-video and social platforms with superior engagement
• Cross-platform search (in-app, e-commerce) reduces dependence on standalone web search

Erosion risks

• Negative publicity or trust incidents (content/ads/safety)
• Perceived product quality decline vs competitors
• Nationalism/geopolitics affecting consumer sentiment

Leading indicators

• Net promoter score / brand surveys (where available)
• Regulatory actions and fines related to advertising/search
• User retention / DAU-MAU trends

Counterarguments

• Brand alone may not prevent budget shifts if advertisers see better performance elsewhere
• Users may multi-home across platforms for discovery

Erosion risks

• Developers/advertisers prioritizing ecosystems elsewhere (e.g., super-apps, short-video)
• Regulatory requirements increasing interoperability and limiting platform control

Leading indicators

• Ad landing-page mix shifting to Baidu-native properties
• Advertiser retention and spend concentration
• Developer adoption of Baidu mini-program/agent tooling

Counterarguments

• Most advertisers can still use off-platform landing pages, limiting lock-in
• Cross-platform tooling reduces friction to port experiences elsewhere

Erosion risks

• Price competition and commoditization in cloud/IaaS
• Export controls or supply constraints on advanced compute
• Fast-following competitors matching foundation-model capability

Leading indicators

• AI Cloud revenue growth and gross margin
• Compute utilization and capex intensity
• Adoption of ERNIE API and foundation-model workloads

Counterarguments

• Enterprises increasingly adopt multi-cloud strategies, limiting vendor lock-in
• Open-source models reduce differentiation and shift value to services/integration

Erosion risks

• Standardized APIs and open-source tooling lower switching costs
• Customers internalize AI stacks or shift to system integrators
• Regulatory constraints on data movement and model usage

Leading indicators

• Number of ERNIE API customers / usage growth
• Net revenue retention in AI Cloud
• Share of revenue from subscription/recurring contracts

Counterarguments

• APIs and model endpoints can be swapped relatively quickly for many use cases
• Multi-cloud and abstraction layers reduce practical lock-in

Erosion risks

• Project-heavy mix prevents true productized scaling
• Competitors replicate reference architectures quickly
• Public sector procurement favors incumbents or state-linked vendors

Leading indicators

• Solution gross margin and delivery cycle time
• Share of standardized products vs bespoke projects
• Partner ecosystem growth (SIs/ISVs)

Counterarguments

• Cloud markets often compete on price; standardization benefits may be competed away
• Switching costs are limited when workloads are containerized/portable

Erosion risks

• Regulators expanding licenses to more competitors
• Safety incidents causing stricter rules or pauses
• Local protectionism in procurement and pilots

Leading indicators

• Number of cities with fully driverless permits
• Regulatory approvals and operational design domain expansions
• Safety metrics and incident rates

Counterarguments

• Licenses may not translate to durable economics or market leadership
• OEMs and well-funded startups can secure comparable permits over time

Erosion risks

• Rapid tech progress from competitors narrows capability gaps
• High capex/opex prevents scaling despite technical progress
• Consumer adoption remains limited without subsidies

Leading indicators

• Apollo Go rides completed and paid-ride mix
• Unit economics per ride (cost per km, utilization)
• OEM partnership wins for self-driving/ADAS

Counterarguments

• Data/ride volume advantages may be less valuable if simulation and foundation models dominate learning
• Commercialization could lag even with technical leadership

Erosion risks

• Content regulation and censorship affecting slate
• Rising content costs with uncertain ROI
• User time shifting to short/mini-form video platforms

Leading indicators

• Membership services revenue trend
• Content spend vs engagement / hit rate
• Monthly active users and watch time

Counterarguments

• Streaming services compete heavily on price and hit content; advantages are often transient
• Short-video platforms can capture discretionary viewing time without comparable content spend

Erosion risks

• Rising talent costs and competition for IP
• Audience fragmentation reducing scale benefits
• Platform shift toward user-generated content

Leading indicators

• Content cost as % of revenue
• Gross margin trend for iQIYI segment
• Share of viewership from originals vs licensed

Counterarguments

• Efficiency gains can be offset by bidding wars for top creators and IP
• Competitors can also build studios and replicate processes

Erosion risks

• Users multi-home across platforms, diluting data advantage
• Regulatory limits on personalization/data use
• Content slate, not recommendations, dominates retention

Leading indicators

• Personalization engagement metrics (CTR, watch time per user)
• Churn rates / retention of paid members
• Share of traffic from recommendations vs direct

Counterarguments

• Recommendation systems are widely available and not a unique capability
• Data advantage is limited if viewers follow hit content across platforms