How 900 Billion Agentic AI Agents Will Transform Telecom Networks by 2035

The jump from traditional AI systems to agentic ones — those that operate autonomously and take initiative — brings new challenges. Instead of waiting for queries, AI agents constantly generate data, react to their environment, and communicate with each other. This upends what networks need to deliver.

For decades, mobile networks optimized for downlink traffic — delivering content to users. But agentic AI systems want something different: powerful uplink, minimal latency, and coverage everywhere. The shift goes beyond technology to architecture itself.

📡 From Downlink to Uplink: The Great Reversal

Huawei predicts agentic AI systems will need five times more uplink capacity than we have today. The reason? Multimodal interaction. Agents don't just process text — they combine video, images, audio, and sensor data simultaneously.

"For multimodal interactions, we need gigabit uplink capability," said Fang Xiang, Huawei VP. Smart glasses are a perfect example — they continuously record video and images, sending them upstream for real-time AI analysis.

The Upload Paradox

Traditional apps had sporadic upload spikes — livestreaming, video calls. But agentic systems will generate continuous upstream traffic. Picture billions of devices sending sensor data, voice commands, and contextual information simultaneously.

InterDigital already spotted early signs of this shift. Next-gen wearables collect voice, biometric, and contextual signals to support persistent AI interactions. IoT sensors continuously send operational and environmental data to AI models for analysis and automation.

⚡ Latency: The New Network Obsession

Latency takes on new meaning in the agentic AI era. For AI robots that want to seem human-like, end-to-end latency must stay under 400ms — the so-called "Doherty Threshold."

This 400ms limit isn't random. It reflects how long humans need to stay engaged in computer interactions. Below this threshold, the experience feels natural. Above it, it becomes annoying.

Agentic AI systems evolve from asynchronous cloud queries to real-time systems involving robotics, automation, and immersive experiences. This requires new network architecture from the ground up.

The Real-Time Interaction Challenge

Consider an AI assistant having a voice conversation — every delay becomes noticeable. Or an autonomous vehicle deciding to turn — latency could be a safety issue. Agentic systems create scenarios where timing is critical.

🌐 Coverage: Beyond Urban Centers

The third major change is coverage. Agent-enabled devices won't be limited to smartphones in urban centers. They'll expand to vehicles, robots, and a wide range of connected objects that need high-quality coverage in rural areas, highways, and uncovered zones.

"Agent-powered devices will expand beyond current limits," Fang said. "We must extend high-quality coverage to villages, roads, and uncovered areas."

This puts entirely new demands on network operators. It's no longer enough to cover densely populated spots — they must ensure consistent, low-latency connectivity everywhere.

🤖 RAN Agent: The Network's Brain

Huawei's answer to these challenges has a name: RAN Agent. It's an AI-driven system built on the company's telecom foundation model and designed for intent-driven network automation.

RAN Agent is supported by the RAN Digital Twin System (RDTS), which models physical network assets, devices, and environment. RDTS provides the real-time data foundation on which the Agent operates.

Combined, this means the RAN can adapt autonomously based on user requirements, optimizing the network for user experience, O&M efficiency, and energy usage.

How It Works in Practice

Picture this scenario: thousands of AI agents in an area start exchanging large volumes of multimodal data. Traditionally, this would create congestion and require manual intervention from network engineers.

With RAN Agent, the system detects the traffic pattern change, analyzes requirements, decides appropriate resource allocation, and executes changes automatically. All this happens in real-time, without human intervention.

🔧 The Technology Foundation of Change

Huawei unveiled a series of new products to support this transition. The GigaGreen Plus series incorporates new antenna architectures and materials that improve performance and energy efficiency.

"With the help of innovations in new materials, advanced antenna architectures, and engineering, it extends coverage by 15%, sets new benchmarks for energy efficiency, and reduces size and weight by 30%."

Fang Xiang, VP Huawei Wireless Solutions

The series includes the tri-band ultra-wideband MetaAAU and a 256-transmit antenna unit operating in the 6 GHz band. They're designed to support 5G-Advanced deployments capable of delivering up to 10 Gbps downlink and 1 Gbps uplink speeds.

The Chiplet Architecture of Tomorrow

Huawei also revealed its next-generation UBBPi baseband platform, using chiplet-based architecture and near-memory computing for enhanced performance. The system doubles both capacity and energy efficiency compared to the previous generation.

A key feature is "all-scenario cell-free" networking. It includes distributed access points to eliminate cell boundaries, promoting seamless performance. Advanced coordination algorithms allow the baseband to optimize resources across time, frequency, and spatial domains.

📊 The Numbers Speak for Themselves

Market statistics emerging are impressive. According to ABI Research, smart glasses shipments will reach 70 million by 2030, with cellular-enabled devices representing over 12% of shipments.

Uplink pressures are already visible in video-heavy applications like livestreaming and real-time video collaboration. But agentic AI systems will create continuous upstream data exchanges from connected devices, creating sustained pressure on uplink capacity.

Distributed Intelligence Architectures

The solution, according to InterDigital research, is moving toward distributed intelligence architectures. AI workloads must be orchestrated across on-device processors, edge, and cloud platforms based on their complexity.

This means we can't rely solely on cloud computing. Intelligence must be embedded deeper into network infrastructure to ensure AI-enabled applications operate efficiently without sacrificing performance.

🚀 Pioneer Experiences

Aircom already moved forward developing raNora, a standalone Agentic AI platform designed for day-to-day radio planning and engineering workflows. The system introduces a governed, telco-trained multi-agent architecture that brings structured reasoning and execution to operational environments.

raNora includes two high-impact agents: the Database Agent for network data queries and compliance audits, and the Coverage Agent for coverage analysis and automated site placement recommendations.

"The transition toward autonomous networks requires AI that can operate within governed workflows — not just generate insights," said Khurram Chaudhry, VP Products & Engineering at Aircom.

Reality in Operational Environments

Early results are encouraging. Instead of ticket-driven processes that limit efficiency, agentic systems can execute repetitive tasks while humans oversee policy and edge cases. This doesn't replace engineers — it frees them for more creative work.

NTT DOCOMO also started commercial deployment of an agentic AI system for network maintenance, using one of the world's largest datasets. Results show significant improvement in operational efficiency.

🔮 Coming Challenges

Despite progress, significant challenges remain. First is security — billions of AI agents communicating autonomously create new vulnerabilities. Traditional security models based on perimeter defense won't work.

Second challenge is interoperability. With so many different agent types and vendors, how do we ensure everything works together? We need new standards and protocols.

Third challenge is human skills. Agentic systems require new skills from network engineers — more AI/ML knowledge, less manual configuration. The industry must invest in training and education.

🎯 Frequently Asked Questions

How do agentic AI differ from traditional AI systems?

Traditional AI systems wait for queries and respond. Agentic AI takes initiative, plans actions, and executes tasks autonomously. They can communicate with each other and adapt to real-time conditions.

When will we see mass deployment of agentic AI in networks?

First commercial deployments already started in 2026. Huawei predicts by 2035 we'll have 900 billion agents. Scaling will be gradual, starting from specialized use cases.

How does energy consumption get affected?

Paradoxically, agentic systems can reduce energy consumption through better resource optimization. Huawei's GigaGreen Plus series promises 30% reduction in size and weight while doubling energy efficiency.

The change agentic AI brings isn't just incremental — it's transformational. We're talking about fundamental restructuring of network architectures, from radio layers to cloud backends. Operators who adapt early will gain competitive advantage. Those who delay risk falling behind in a rapidly changing market.

The question isn't whether this era will come — it's when and how fast we can adapt. Early signs show the future arrived sooner than we expected.