From 2 million to 5 million: Intelligent agents move from concept to large-scale deployment, and the "six-layer full-stack architecture" of enterprise-level AI platforms – IDC – August 28, 2026.

In 2026, China's AI Agent market is undergoing a crucial leap from proof-of-concept to large-scale deployment. IDC data shows that the number of active enterprise AI agents will jump from nearly 2 million in 2025 to an estimated 5 million in 2026 , demonstrating strong growth momentum. At the same time, IDC research indicates that enterprises are no longer satisfied with building isolated AI applications, but are instead shifting towards building unified enterprise-grade AI platforms.

Against this backdrop, International Data Corporation (IDC) released its latest research report , "New Enterprise AI Platforms Are Emerging" , which provides an in-depth analysis of the technological evolution, construction paths, and best practices of enterprise AI platforms. The report points out that the value proposition of AI has shifted from "technical toys" to "operational assets," and its application depth has moved from "single-point question answering" to "embedded in core business flows." The role of employees is evolving from "system operators" to "intelligent commanders," and the organizational structure of enterprises is shifting from "a collection of people" to "human-intelligence collaboration." This trend has spurred the rise of "new enterprise AI platforms"—not simply a combination of models and applications, but a six-layer full-stack architecture encompassing the AI ​​foundation layer, data layer, intelligent agent development layer, process and business layer, application entry layer, and governance and security layer .

An IDC report points out that the construction of enterprise-level AI platforms currently presents two clear technological paths:

Path 1: Intelligent Process Automation. This approach focuses on the process itself, embedding AI capabilities into the entire BPM (Business Process Management) lifecycle. AI acts as a "digital employee node" within the process, performing tasks such as initial review, form filling, and risk identification. The advantages of this path are its rapid value realization and high integration with existing business systems.

Path Two: Data- AI Integration. Centered on data infrastructure, this path builds an end-to-end foundation for data and AI fusion, encompassing lakehouse integration, data governance, model training and inference, and agent development. This path is suitable for industry scenarios with large data volumes and complex business rules, aiming for long-term data assetization and the development of native AI capabilities.

Manufacturer Practices: Building Enterprise AI Foundations Through Multiple Pathways

The report provides an in-depth analysis of the enterprise-level AI platform construction practices of multiple vendors, presenting two main construction paths: "intelligent process automation" and "data-AI integration".

Intelligent process automation paths are centered on process and business orchestration. Aozhe supports enterprise AI nativeization with "AI + Data + Low Code," covering the entire chain from application generation to full-domain AI capability reuse through three core modules: AI Designer, AI Agent, and AI Discover. Its Cloud Hub platform serves large and medium-sized enterprises, while its Tritium Cloud platform is ready to use for SMEs. Yanhuang Yingdong proposes a full-stack capability of "AI + Data + Process + Application + Governance," using AI Agent and AI Workflow as the "engine," and has built four core products: the AI ​​Agent intelligent body platform, the aPaaS low-code platform, the bpmPaaS intelligent process platform, and the iPaaS intelligent integration platform. It also provides differentiated technical solutions according to AI risk levels . Inspur Tongsoft , with "process skill-based and data semantic" as its core concept, encapsulates complex processes into callable business skill packages, constructing knowledge graphs and business ontology to enable AI to understand enterprise data, achieving significant results in scenarios such as contract review at Sichuan Jiuzhou, safety operation and maintenance in metal mines, and process optimization in oil and gas fields.

The integrated data- AI path is based on data infrastructure. Primeton positions itself as an "enterprise-level digital intelligence foundation," with a system covering the entire chain of applications, data, and integration. Its MCP service release capability can seamlessly integrate legacy systems into AI, and its AI programming engineering solves the problem of code out of control. The "Prime Number Questioning" open computing process enables transparent data analysis. DeepTech positions itself as the "infrastructure for enterprise AI employees," pioneering the Ontology methodology and Token Factory pattern, increasing the number of effective tokens in general models from 5 to 9, achieving high-precision, zero-illusion implementation in scenarios such as manufacturing fault diagnosis, retail intelligent replenishment, and medical AI employees. Kejie Technology builds a full-stack AI infrastructure based on a lakeware architecture, proposing the concept that "computing power is the basic engine, algorithms are the soul, and data is the core fuel." Its KDP intelligent infrastructure platform and Keen AI artificial intelligence platform form a layered architecture, simultaneously compatible with traditional machine learning and new scenarios with large models. With CyberData, CyberAI, CyberEngine, and CyberGPT as its core product matrix, CyberAI builds an integrated Data+AI architecture that supports mixed scheduling of CPU, GPU, and NPU computing power, serving large central enterprises such as China National Petroleum Corporation and China Power Construction Corporation.

To learn more, visit: www.idc.com