MIIT Policy: AI Embedding into Communication Networks for Intelligent Infrastructure

Over the past two years, AI discussions have centered on large models, agents, and application-layer innovation. However, a deeper shift is occurring: AI is moving into communication networks, compute nodes, and urban connectivity infrastructure itself.

On June 12, the Digital China Summit website disclosed MIIT's "Implementation Opinions on Innovative Development of 'Artificial Intelligence + Information Communication' (2026–2028)." This document goes beyond encouraging AI applications; it explicitly calls for embedding AI capabilities into the technical, product, and service systems of the information communication industry. AI is becoming part of the network, not just a terminal assistant.

Core Significance: Not Just "+AI" but "Into the Network"

Unlike typical "AI+industry" policies, this opinion positions the information communication industry as the foundational layer for AI deployment. The document targets a 2028 development pattern with significantly enhanced system capabilities, breakthrough technical products, and prominent empowerment effects, with a 2030 maturity goal. Coverage spans networks, terminals, compute, intelligent agents, 6G, integrated space-air-ground, 5G-A, edge intelligence, and industry empowerment. This represents an upgrade from "transmission pipes" to "intelligent infrastructure."

Why Industry Readers Should Care

Many assume network upgrades concern only operators and vendors, but the opposite is true. Today, many AI projects fail to deploy not because models are insufficient, but because three elements remain unconnected:

Whether data can enter systems in real time

Whether compute can be invoked at the right location

Whether business processes can close the loop under low-latency, high-reliability conditions

All three are fundamentally tied to information communication infrastructure. When AI enters the network layer, the change is not merely "smarter answers" but "faster system response, smoother process collaboration, and earlier warning and handling."

Three Concrete Changes Ahead

1. Network Operations Shift from Manual Inspection to Network Autonomy

The opinion proposes deploying network intelligent agents, advancing edge intelligence, and improving network O&M intelligence. Historically, network optimization, fault diagnosis, and traffic scheduling relied on experienced engineers. With AI embedded in the network, operations shift from "people finding problems" to "systems detecting anomalies first, proposing solutions, then humans confirming critical decisions." This technical change has practical impact: more stable, business-aware networks improve AI application availability.

2. 5G-A and Edge Intelligence Push AI to the Field

The document mentions enhancing 5G-A capabilities, promoting edge-intelligence collaboration, and building a "1-millisecond urban compute circle." This means AI capabilities will move from centralized clouds to locations closer to business scenes. This is critical for low-latency scenarios:

Video recognition and real-time alerting

Industrial inspection and equipment diagnosis

Vehicle-road coordination and low-altitude sensing

Large-scale terminal collaborative control

When networks, edge nodes, and model inference collaborate, AI becomes a "real-time system" rather than a "post-analysis tool."

3. Anti-Fraud, Harassment Governance, and Anomaly Communication Identification Emphasize "Forward Warning"

The opinion highlights using AI to govern spam messages, harassing calls, and telecom fraud. This signals a shift from post-incident investigation to collaborative early warning across communication, platform, and model sides. A scenario illustration: when abnormal call patterns, bulk device behavior, anomalous traffic features, or fraud script variants emerge, the system can rapidly identify them on the network side, combine complaint data, black/gray industry clues, business rules, and model judgments to advance risk alerts, number disposal, and behavior blocking to earlier stages. This does not mean AI "automatically solves all fraud," but it moves anti-fraud from "passive response" toward "active discovery, rapid linkage, layered handling."

Real Value Lies in Connecting Four Capability Links

Current projects and policy direction show a pragmatic turn: success requires advancing four links together:

1. Data

Stable, governable, updatable, and labelable data sources determine whether systems can run long-term.

2. Network

The ability to connect data, models, commands, and feedback on appropriate links and nodes determines whether the system "runs."

3. Compute

Compute is not just scale but location: what stays central, what goes to the edge, what needs real-time inference — directly affecting cost and effectiveness.

4. Scenario Loop

Whether the system produces results depends not on PPT capabilities but on actual integration into business, warning, and handling processes.

Three Practical Takeaways for Organizations

Viewing this document as an observation window yields three pragmatic judgments:

AI competition is shifting from "model capability competition" to "model + network + data + scenario" system competition.

Previously low-level concepts like 5G-A, edge compute, and network intelligent agents will increasingly affect upper-layer application delivery.

Intelligent upgrades in anti-fraud, digital governance, and industry regulation will prioritize front-end perception, real-time linkage, and cross-system collaboration over single-platform feature stacking.

Conclusion

AI is entering a new phase. The previous phase focused on model strength; the next phase focuses on whether systems can truly integrate into networks, reach the field, and form closed loops. MIIT's opinion signals clearly: the next round of AI competition is sinking from the application layer to the infrastructure layer. The solutions that succeed will not be "chatty AI" but AI systems that "can connect, perceive, judge, collaborate, and dispose." This may be the most noteworthy change for 2026 and beyond.