IDC：2024年工业大模型应用市场规模约12.1亿元人民币

Core Insights - The industrial large model application market in China is projected to reach approximately 1.21 billion RMB in 2024, with enterprises focusing their budgets on platform and foundational model systems driven by a mindset of not wanting to fall behind, shifting evaluation criteria from "is it available" to "is it valuable" [1][3] Group 1: Market Trends - 2025 is identified as a critical year for the implementation of industrial large models, with improvements in open-source and domestic foundational model performance, stable multimodal understanding, and tool invocation becoming commonplace in production environments [3] - The focus is shifting from pilot projects to replication, with companies demanding higher controllability, including data localization, auditability, versioning, and risk mitigation becoming essential requirements [3][4] Group 2: Technological Evolution - The evolution path for the next two to three years is clear, with large models transitioning from "answerable" to "executable," integrating long context and function calls with RAG and knowledge graphs to enable cross-system actions under controlled permissions [4] - The engineering focus is moving from "heavy platform, light scenario" to "light platform, heavy scenario," emphasizing the delivery efficiency through replicable scenario packages [4][5] Group 3: Strategic Focus for Technology Providers - Technology suppliers are advised to concentrate on five high-certainty intelligent agent categories: multimodal knowledge retrieval, document generation and compliance review, mass object communication, experience extraction and root cause/process optimization, and cross-system querying to automatic execution [5][6] - Each category should have clear KPIs to measure milestones, such as first-pass yield, OEE, and quality improvement, with business value tied to SLA rather than just interface availability [6] Group 4: Business Model Transformation - The business model is evolving from one-time project-based approaches to "platform subscription + scenario packages + savings sharing," ensuring recurring revenue to cover ongoing operations and version governance [6] - There is a need for early adaptation and mutual certification with ERP/MES/PLM/SCADA/DCS and key OT devices to ensure data sovereignty, model registration, behavior tracing, and compatibility with domestic software and hardware [6] Group 5: Competitive Landscape - The success of industrial large models will depend not on parameters but on replicable engineering capabilities, with the ability to transform high-value scenarios into auditable, maintainable, and measurable intelligent product packages being crucial for market leadership [6]