The wave of intelligent agents is approaching! Agent connections reach hundreds of billions, and inference computing power has increased by more than 10 times

As the era of intelligent internet agents officially begins, human society is standing at a new growth milestone. Industry experts predict that, based on the current population of approximately 8.3 billion, the number of intelligent agents will surge to over 900 billion in the future, becoming the new main actors on the internet. This disruptive transformation not only demands unprecedented capabilities from communication networks but also will give rise to a new industry space valued at trillions of dollars.

At the recently concluded MWC 2026 in Barcelona, several industry experts shared their thoughts on the accelerating arrival of the intelligent agent world.

▍ Intelligent Internet with Hundreds of Billions of Connections and Over 10x Growth in Reasoning Power

China Telecom General Manager Liu Guiqing pointed out that artificial intelligence is entering a phase of autonomous execution and intelligent collaboration—Agentic AI—and has begun the era of intelligent internet agents. How to accurately grasp the transformative opportunities led by Agentic AI has become a common challenge for global telecom operators.

He shared three insights: First, Agentic AI will drive a dual leap—both quantitative and qualitative—in new digital information infrastructure centered on cloud-network integration. Quantitatively, the number of connected intelligent agents (Agents) will reach hundreds of billions, with reasoning computing power rapidly surpassing training power, achieving over 10x growth; qualitatively, the intelligent internet will evolve into a “distributed autonomous” architecture, with computing power shifting from centralized cloud to ubiquitous collaboration across cloud, network, and edge.

Second, the application of Agentic AI will shift from “+AI” to “AI+”. Large-scale deployment of Agentic AI will help enterprises move from single-point AI tools to deep integration across the entire industry chain and core processes, significantly improving operational efficiency and reducing carbon emissions.

Third, Agentic AI will push security boundaries from “cybersecurity” to “trustworthy intelligent agents.” As intelligent agents gain autonomous decision-making and tool invocation capabilities, becoming active entities in networks, this will enhance security defenses through increased automation and intelligence. At the same time, mechanisms for agent identity verification, behavior auditing, and behavior interruption need to be established to ensure actions are identifiable, constrained, and traceable.

▍ Uplink Traffic Share Will Significantly Increase

Huawei Wireless Product Line Vice President and Chief Marketing Officer Zhao Dong stated that in the future, there will be hundreds of billions of connected intelligent agents, and wireless networks will face unprecedented demands and challenges.

He believes that the rise of intelligent agents will reshape network connectivity on three levels:

First, interaction methods will evolve from single-modal to multi-modal fusion. Traditional communication mainly relies on text or voice, but intelligent agents need to process video, audio, images, and text simultaneously to enable more natural and richer human-computer interactions.

Second, different intelligent agents will achieve extensive collaboration. For example, intelligent agents on smartphones can invoke other apps, interact, and coordinate with other agents, forming a decentralized yet interconnected intelligent network.

Third, networks will upgrade from “data connection” to “intelligent connection.” Traditional network capabilities focus more on downstream experience, but in the world of intelligent agents, uplink transmission capacity becomes critical. When users interact with agents, they often need to upload images or videos in real-time and expect immediate responses. This requires networks not only to have sufficient uplink bandwidth but also to ensure low latency and deterministic experience.

Zhao Dong cited a practice at an art museum in Shanghai. The museum introduced AI agents as guides; visitors simply open an app on their phones, take photos or videos of exhibits, and receive real-time descriptions. Data shows that over 3,000 visitors use this service daily, leading to significant changes in network traffic patterns.

“Under traditional scenarios, uplink traffic usually accounts for about 10%, but in this intelligent agent scenario, uplink traffic has broken through 60%, reaching 63%,” Zhao Dong said. This indicates a shift in network demand. The real-time requirements of intelligent agents place higher standards on uplink speeds. For example, a high-definition image used for recognition can exceed 400KB, and with multiple users, the demand for uplink bandwidth multiplies.

Zhao Dong further revealed test results: in this scenario, if 10 users simultaneously use intelligent agents for exhibit explanations and expect responses within 2 minutes, each user needs at least 20 Mbps uplink speed. Even in China’s excellent network coverage, this requirement far exceeds current typical uplink capabilities. In many cases, actual speeds only maintain around 2-3 Mbps, a significant gap.

To address this challenge, operators are actively adjusting network resource allocation. Zhao Dong cited that China Mobile is exploring a 3:2 uplink-to-downlink ratio on the 4.9 GHz band to better meet the urgent uplink bandwidth needs of intelligent applications.

▍ Applications Still in Early Stages; Communication Standards for Agents Need Maturation

Omdia Senior Principal Analyst Yang Guang believes that the rise of intelligent agents will have profound impacts on telecom networks. He sees these impacts mainly on two levels: first, how intelligent agents empower the network itself; second, how networks will support the future of intelligent agents.

On one hand, telecom operators can incorporate intelligent agents into network maintenance and management to improve efficiency and reduce labor costs. However, practical implementation faces challenges. Currently, different equipment vendors provide network systems that operate independently; integrating intelligent agents requires effective communication between them, which is a key issue. Although the industry has begun to promote relevant standards, it will take time to reach maturity. Additionally, since intelligent agents rely on large models, their inherent “hallucination” issues could affect network stability and reliability. Operators and equipment vendors are actively exploring solutions, and progress has been made.

On the other hand, telecom networks will become the infrastructure supporting intelligent agent applications. Looking ahead, the number of intelligent agents may far surpass human users, becoming a new “user” category in communication networks. These agents will generate continuous traffic and may also drive new communication modes and service demands, creating substantial revenue opportunities for operators. From this perspective, the market potential for intelligent agents extends well beyond current human user scales.

However, Yang Guang also cautions that widespread adoption of intelligent agents will require a long development process. Currently, related applications are still in early exploratory stages. More importantly, communication between intelligent agents needs standardized protocols—just as humans need a common language to communicate. Issues such as agent identification, security, and regulatory mechanisms remain unresolved.

Of course, these uncertainties also present opportunities for the telecom industry. Yang Guang pointed out that telecom networks have long prioritized security, reliability, and traceability—core attributes essential for intelligent agent communication. Operators have extensive experience in ensuring communication quality, identity verification, and data privacy, and are capable of gradually transforming existing network architectures to support the complex communication needs of future intelligent agents.

“Telecom networks will not only carry the traffic generated by intelligent agents but also contribute technological strength to the overall development of societal intelligent agents,” Yang Guang concluded.

(Source: Cailian Press)