Huawei proposes a new method to evaluate the quality of service of AI applications called AI MOS, modeled on the average opinion score used to measure voice service quality.
In summary, here’s what you need to know:
Future traffic changes – Huawei expects that AI agent applications will generate much more uplink traffic than traditional mobile services, forcing the evolution of networks designed around heavy use of downlinks.
connection surge – Huawei estimates that the number of AI agent connections could reach 100 billion, creating significant new infrastructure demands across mobile networks, cloud platforms, and edge computing systems.
New AI metrics – The company proposes AI MOS as a way to evaluate the quality of AI services and proposes an A2A-T protocol to support communication between agents across communication networks.
Huwei told MWC last week that the rapid emergence of AI agents will require significant changes in the way mobile networks are designed, especially in areas such as uplink capacity, service evaluation, and communication between agents.
Zhao Dong, the company’s vice president and chief marketing officer of wireless network products, speaking at a media and analyst roundtable, said the telecommunications industry is entering the “agent era” where AI agents interact with users, devices, and other agents in a variety of environments.
“We are already at the dawn of the agent era,” he said. “Agents will go to your phone, your car, your robot.”
He suggested that AI agents will significantly increase the number of connections on communication networks. Huawei estimates that the number of intelligent agent connections could reach 100 billion in the future, creating new infrastructure requirements for carriers.
One of the most important changes Zhao highlighted concerns traffic patterns. He said traditional mobile networks were primarily designed to handle downlink traffic, as most digital services involve downloading content. “The traditional functionality of the network is focused on the downlink,” Zhao says. “This time the agent needs more uplinks.”
He explained the changes using the example of an AI-powered museum guide experiment in Shanghai. In the trial, visitors used a mobile AI agent to send photos and videos to the system for analysis of their work. In this scenario, uplink traffic accounts for 63% of the total traffic, compared to traditional mobile networks where downlink traffic typically accounts for about 90%, Zhao said.
“This huge change means networks need to evolve,” Zhao said.
Huawei has previously discussed the use of AI to improve network operations. Zhao said the focus is now shifting to designing networks that directly support AI applications. “Simply put, last year we talked about AI for wireless networks, and this year we’re talking about networks for AI,” Zhao said.
Huawei describes the concept as an “agent-oriented network” and aims to support large-scale AI agent interactions and services. Huawei proposes a new method to evaluate the service quality of AI applications called AI MOS, which is modeled on the average opinion score used to measure voice service quality.
“AI MOS is similar to MOS for voice services,” Zhao says. “It’s a measure of service experience.” Huawei is discussing the concept with industry partners and standards bodies, including the International Telecommunications Union.
The company is also proposing communication-focused enhancements to its agent-to-agent communication protocol called A2A-T. “A2A is a communication protocol between agents,” Zhao said. “The telecom industry has defined A2A-T.”
Zhao also said that the development of AI agent services will involve collaboration across the broader technology ecosystem. “Terminal providers, chip manufacturers, vendors, and service providers are all working hard to promote the prosperity of the ecosystem,” Zhao said.
