Over the past three decades, the information and communications industry has focused on connectivity — linking people to people, people to things, and things to things in the digital world. However, as the global mobile AI wave accelerates, the industry is poised for structural changes. With 3.7 billion daily minutes of AI human-machine calls and over 1.3 billion monthly active users of mobile AI applications, the information interaction model is evolving from human-to-human to human-machine and machine-machine. Agents are also growing at an unprecedented rate, deeply penetrating various industries and ushering in the Agentverse — a world where humans, machines, and intelligent agents seamlessly connect and collaborate. Just as PCs gave way to the mobile internet, transforming lives and leading us into a golden age of mobile telecommunications, the Agentverse will similarly redefine the future of mobile communications.
The rise of agents fuels the token economy and industry productivity
Currently, the rapid iteration of foundation models is driving the exponential growth of inference capabilities. AI is evolving from passive tools into agents capable of autonomous planning, long-term reasoning, and proactive task execution. As agents break through application silos and integrate into diverse physical devices like smart glasses, connected vehicles, and embodied robots, connectivity is transforming. It is no longer just about transferring information, but also about decision-making and collaboration. Each device becomes an interactive anchor point, bridging the digital and physical worlds like never before.
The popularization of agents is driving a massive surge in token usage. According to the National Data Administration, average daily token calls in China have skyrocketed from hundreds of billions to trillions in just two years. As agents like OpenClaw increasingly move to the device side, a single complex device-cloud interaction can involve up to millions of token calls. It is predicted that the number of active agents will surge from 28.6 million in 2025 to 900 billion in 2035. This huge number of agents will reshape production and daily life, enabling every individual, every family, and every enterprise to access intelligence on demand.
Personal agents reshape individual productivity: Agents will act as digital assistants for every user, providing long-term, stable companionship while improving work efficiency and quality of life. For example, in recruitment, the AI agent of Liepin (a leading human resource service company in China) has demonstrated remarkable capabilities: It can process 100 resumes in 40 seconds, accurately recommend potential candidates in 4 minutes, boost interview rates by 60%, and reduce recruitment costs by 90%. In children's education, agents power AI toys with deep emotional interaction and personalized learning capabilities, helping children expand their vocabularies by 30% through targeted teaching.
Home agents build premium smart spaces: Agents are gradually penetrating various home terminals, from smart speakers, central control screens, and fitness mirrors to traditional home appliances such as refrigerators and ovens, enabling intelligent control and personalized services. Global carriers are now actively deploying in this high-value space. At MWC 2026, Chinese carriers showcased smart screens and home robots. European carriers like DT, Orange, and VDF are focusing on AI TV boxes, driving the transition from single-point intelligence to system intelligence in homes and comprehensively improving the quality of home life.
Industry agents accelerate digital and intelligent upgrades: By integrating AI agents' autonomous decision-making with the physical capabilities of embodied robots, industry agents are unlocking immense productivity when applied to core production processes. For example, China Southern Power Grid worked with China Mobile and Leju Robot to deploy embodied robots for power grid inspection. Leveraging a low-latency, high-reliability 5G-A network, the robots can use high-definition devices to read meter data in real time and send the results back to the cloud for analysis by foundation models. This enables long-distance autonomous obstacle avoidance and efficient fault detection, boosting inspection efficiency by 84%.
Three characteristics of agents drive a full upgrade of network capabilities
As physical entities across industries merge into the Agentverse, long-context inference and multi-agent collaboration are becoming the new norm. The traditional mobile network model — best-effort and downlink-prioritized —falls short of agent requirements. Therefore, to support real-time connection and collaboration between numerous agents, the network needs a comprehensive structural upgrade.
First, long-context inference demands greater uplink capacity.
When generating research reports and simulating programming processes, agents require background residence and long-context inference. Their context data accumulates with each interaction, and inference accuracy depends heavily on understanding long contexts. Today, the maximum context size has reached over 10 million tokens, making continuous uplink transmission of ultra-large data blocks a key challenge. For example, a single interaction involving 400,000 tokens between a mobile phone and a foundation model generates 1.4 MB of uplink traffic. If the context reaches 10 million tokens, the uplink traffic of a single interaction can exceed 45 MB. This high-capacity demand strains the uplink capabilities of existing base stations.
Second, real-time concurrency of multi-modal interactions stimulates demand for higher uplink rates.
In interactive scenarios like real-time translation and exhibition tours, real-time multi-modal perception and continuous human-computer interaction are crucial to user experience. In a typical AI tour guide scenario, the uplink traffic ratio often exceeds 60% during multi-user concurrency, as high-frame-rate video streams need to be sent to the cloud for visual understanding. To ensure accurate environmental recognition, uploading a single frame of HD imagery requires over 400 KB. The network needs much higher uplink bandwidth. Without it, uplink transmission quality will be impacted and multi-modal token transmission will become asynchronous, leading to irrelevant responses or interrupted conversations.
Finally, efficient execution of embodied AI calls for end-to-end experience assurance.
In scenarios like industrial control, autonomous driving, and embodied AI, agents require high response speeds and service continuity. For example, humanoid robots at major events operate at very high frequency. Even brief network congestion or jitter can cause inference disorders or total system shutdowns. Latency-sensitive applications demand stringent time to first byte (TTFB), low end-to-end latency, and minimal jitter. The network must therefore be capable of maintaining very strict and fixed thresholds for bit error rates and transmission latency to ensure high-assurance, end-to-end deterministic performance.
Building agent-oriented mobile networks for diverse Agentverse requirements
As interaction objects expand from humans to agents, interaction forms are evolving to become multi-modal, high-frequency, and continuous concurrency. To tackle these new network challenges, Huawei has achieved breakthroughs in three key areas: Network capability upgrade, user experience assurance, and intent interface openness. These collaborative innovations comprehensively support diverse Agentverse requirements.
First, building agentic mobile broadband (MBB) networks to create a multi-dimensional capability foundation
Sufficient uplink bandwidth is the cornerstone of the intelligent world. Currently, most networks' uplink edge rate is only 2–3 Mbit/s. 5G-A demands 20 Mbit/s uplink, and 6G demands 50 Mbit/s. To meet these needs, Huawei has introduced M-MIMO devices to deliver higher bandwidth and more channels. By leveraging technologies like global 3D intelligent coordination, flexible beamforming, and intelligent pairing, the network can dynamically allocate spectrum resources and optimize performance across timeframes, frequency, and area. This will double the quality experience of edge users and lay a solid foundation for a smooth transition to 6G.
Second, developing the innovative token block pipe to ensure agent service experience
The traditional QoS mechanism, designed for bit streams, effectively ensures video services but falls short in terms of handling AI's burst low-latency data transmission and multi-modal concurrency. To address this, Huawei provides the token block pipe mechanism, which is capable of dynamically constructing hierarchical transmission paths to cope with bursts of blocks and 100 ms latency services. Specifically, the mechanism provides high-speed, low-latency, low-jitter channels for visual stream blocks, and reserves dedicated, low-latency and high-reliability lanes for control stream blocks. This ensures differentiated assurance at the modal level.
In addition, just as the mean opinion score (MOS) was used to measure voice communication quality, Huawei has been working with authoritative organizations such as the International Telecommunication Union (ITU) and China Communications Standards Association (CCSA) to promote the initiation of AI-MOS experience evaluation standards. These standards will map the service quality of agents to network KPIs, for quantifiable end-to-end experience and the closed-loop management of network transmission quality.
Third, enabling multi-agent integration and ecosystem openness through A2A-T intent interfaces to drive carrier business success
Traditional networks rely on API calls, limiting service scalability. In the Agentverse, the communication network itself is an agent. China Mobile and Huawei have jointly launched the industry's first intent-driven interface, the Agent-to-Agent Telecom (A2A-T) intent interface, to facilitate communication between network agents and other agents. This interface not only provides carrier-grade security and reliability, but deeply understands multi-modal service intents and automatically converts them into network scheduling capabilities. This means that the time required for interface interconnection and deployment, which previously took months, can now be reduced to just days. Whether it is autonomous driving or fully automated smart factories, their dedicated agents can now access powerful network resources on demand.
Building the network foundation to embrace the Agentverse
Agent capabilities are evolving at astonishing speed, doubling every three months. The operational logic of various industries is being thoroughly reshaped by the Agentverse. Against this backdrop, our current vision of the future may seem overly conservative.
In this intelligent revolution, intelligence without network connectivity is like an information silo. Computing power sets the upper limit of an agent's intelligence, while network connectivity defines its action boundaries. 5G-A is already widely commercialized globally and will continue to unlock the limitless potential of agents. Over the next five years, the world will witness the explosive growth of mobile AI. This major shift will bring a golden opportunity to rebuild industry infrastructure and tap into a trillion-dollar business space.
Huawei looks forward to collaborating with global carriers, device ecosystems, and partners across various industries to build agent-oriented networks. Let this be the start of a new chapter. Together, we will create and embrace an Agentverse where agents are everywhere and empower every aspect of our lives.
Pessoal 
Enterprise 
Carrier 
Huawei Cloud 
Digital Power 
