Artificial intelligence is no longer just an application riding over the network; it is forcing network operators to rethink how their infrastructure is built, run and protected, according to a Cisco executive who testified before a U.S. Senate panel on July 30. Bob Everson, Cisco's chief architect of provider mobility, told the Subcommittee on Telecommunications and Media that AI is changing both the volume of network traffic and the way traffic behaves. The hearing, titled “Intelligent Networks: Powering Artificial Intelligence and Transforming Communications,” examined how the rapid adoption of AI has affected the networks that carry data, video, voice and machine-to-machine traffic.

Sen. Deb Fischer (R-Neb.), chairwoman of the subcommittee, opened the hearing by saying that widespread AI use has forced networks to evolve, requiring more capacity and more complex designs so AI can run efficiently. She also pointed to hundreds of billions of dollars in private network investment and various federal broadband programs aimed at deployment and maintenance. The panel also heard from representatives of U.S. Telecom, Vanderbilt University and the Nebraska Public Service Commission, illustrating the breadth of interest in AI-ready networks.

AI changes more than volume

Everson said Cisco has measured a fourfold increase in AI inference traffic over an eight-month period. Traditional networks were optimized for content flowing downstream, but AI is more two-way and uplink-intensive, he said. Prompts, context, sensor data and agent activity travel back toward AI models, and the connections stay active longer than normal web transactions. AI agents increase the demands further. In Cisco testing, an agent generated 450 percent more traffic than a person performing the same task, and roughly 70 percent of that additional traffic was inference. These changes require more than extra bandwidth; they change how network architects route data, place compute and manage latency.

On campus and branch networks, customers have reported a 34% increase in traffic tied to AI workloads over the last 12 months, and they expect a 96% increase in the coming year, Everson said. Half of enterprise customers tell Cisco that AI demand is concentrated on their Wi-Fi networks, and 73% of organizations either face or expect to face capacity limitations within the next 24 months. Everson linked those numbers to rising east-west traffic, latency-sensitive applications and continuous automated AI traffic. He said the FCC's decision in 2020 to authorize the full 6 GHz band for unlicensed Wi-Fi use has become more valuable as AI workloads spread.

From data centers to the edge

The surge in AI-related traffic is not confined to hyperscale data centers. Everson said enterprises are increasingly deploying small language models, open-source models and specialized models for vision, voice and other tasks, and some of these models will be distributed far from centralized clouds. That distribution changes performance expectations. A model used for factory-floor vision may require results in milliseconds; an AI assistant in a hospital may need to pull patient records securely without sending them to an external cloud. Network designers therefore have to balance capacity, latency and compliance at the same time.

In his prepared testimony, Everson identified several areas shaped by AI:

• Infrastructure investment: AI is driving a shift toward edge computing, and service providers need to plan for AI-native traffic profiles. That planning must account for technical requirements, cost, data sovereignty and security.
• Technical performance: Physical AI use cases such as robotics, autonomous vehicles and industrial automation may require sub-millisecond decision-making. If an autonomous robot has to send data to a central cloud and wait for a response, round-trip latency can be too high for safe real-time operation.
• Cost control: AI operations generate large volumes of data. High-definition video analytics for public safety, for example, can produce terabytes of data each day. Backhauling that data to a central cloud is expensive and can cause congestion.
• Data sovereignty and security: Enterprises and governments are increasingly cautious about moving sensitive data across the public internet to a third-party cloud. Edge processing allows them to keep data in place while still gaining AI benefits.

The promise of AI-native networks

While AI workloads create challenges for network operators, Everson said there is also a significant opportunity to use AI to boost performance and reliability, build security into the network and handle greater complexity. Agentic AI, he explained, will change the nature of traffic, but it also gives operators tools to run networks at machine speed and deliver better results.

One example is what Cisco calls AgenticOps, a set of AI-driven operations capabilities that can make networks self-healing. Everson said Cisco's AI-native tools allow the network to reroute traffic, adjust capacity or reconfigure nodes when the system detects performance degradation or an impending hardware failure. That can increase uptime for mission-critical services.

Everson also highlighted the role of AI in addressing the network industry's talent shortage. By automating repetitive tasks such as ticket resolution, configuration updates and routine maintenance, AI tools can lower the barrier to entry for junior analysts and allow experienced engineers to focus on higher-level architecture, security strategy and threat hunting.

Networks, he said, are moving toward AI-native platforms that become the fabric of intelligent connectivity rather than a simple pipe. As operators push compute toward the network edge, including at cell sites, applications can run directly on the network instead of in remote clouds.

One emerging application is Integrated Sensing and Communication (ISAC), which uses wireless communications and radio-frequency sensing to detect an object's position and path through radio waves. ISAC can work in low-light conditions, through smoke and around obstructions, making it useful for autonomous systems, robotics, smart facilities and public safety, according to Everson.

Recommendations for lawmakers

Everson closed his testimony with three policy recommendations for the subcommittee. First, he encouraged Congress to accelerate development of the U.S. AI-native stack, including compute, core networking and applications. He cited AI-WIN, a collaboration involving Cisco, NVIDIA, MITRE, Orion Development Company, Booz Allen and T-Mobile, as an example of work aimed at creating a secure American-led path from 5G-Advanced to AI-native 6G.

Second, he urged modernization of permitting and infrastructure processes. As computing becomes more distributed, allowing faster and more efficient deployment is essential. Everson also said that if Congress considers the future of the Universal Service Fund, it should account for the evolving costs of AI-ready networks so rural and urban communities can share in the benefits.

Third, he called for a balanced spectrum policy. The 800 MHz of licensed spectrum recently made available by Congress is essential for high-capacity, high-u

Source:Network World News