Ericsson warns AI shift will force radical redesign of telecom networks

Historically, mobile networks have been built for consumption. They are designed to push massive amounts of data down to users—think of streaming video or browsing web pages. But the rise of generative AI, particularly on wearable devices like AI-enabled smart glasses, is flipping this equation.

Speaking at the India Mobile Congress (IMC) 2026, Nitin Bansal, managing director of Ericsson India, said these new applications will force telecom operators to redesign their infrastructure around uploading data rather than downloading it.

"If you're using AI glasses, you need an enhanced uplink because you're sending a lot of data, photos and videos on the uplink," Bansal said. As users feed real-time visual and sensory data to cloud-based AI models, the upload capacity of mobile networks will become the primary bottleneck.

Designing networks for AI

Ericsson calls this shift "networks for AI"—engineering telecom architecture specifically to meet the high-capacity, low-latency upload demands of AI devices.

To cope with this, telecom operators must also rely on "AI for networks." This involves using machine learning inside the infrastructure itself to improve spectral efficiency, manage power consumption, and allow base stations to support more users without needing a proportional increase in physical hardware.

The pressure to modernise is already driving manufacturing. Ericsson recently announced that it has produced more than 15,000 antennas and 25 million network components in India since July 2025. The company's domestic manufacturing push is explicitly geared toward supporting the transition to 5G Standalone networks, which utilise a dedicated 5G core rather than relying on older 4G infrastructure.

Processing at the edge and agentic AI

While Ericsson is focused on carrying the data, semiconductor and consultancy leaders are looking at where the actual processing will happen.

Anku Jain, managing director of MediaTek India, pointed out that the next major wave is "agentic AI"—autonomous software agents running directly on consumer devices like smartphones and laptops to handle complex tasks. To prevent networks from being entirely overwhelmed, much of this processing must happen "at the edge" rather than in the cloud.

Jain highlighted MediaTek's recently launched Dimensity 9600 Pro chipset, which features a dual neural processing unit (NPU) designed to run large language models directly on-device. He added that the Indian government's semiconductor initiatives are helping to build a domestic ecosystem that will eventually catch up to manufacturing advanced silicon nodes.

What comes next for telecom operators

This shift opens up new business models for telecom companies. Berjesh Chawla, managing director and lead for communications, media and technology at Accenture India, said agentic AI will allow operators to make their own network and IT operations increasingly autonomous.

Beyond efficiency, Chawla argued that telecom companies have an opportunity to move past being simple "dumb pipes" of connectivity. By leveraging their existing enterprise relationships, billing systems, and growing data-centre footprints, they can offer integrated AI solutions, sovereign AI hosting to meet local data residency rules, and advanced cybersecurity.

As the industry looks toward the eventual rollout of 6G, these dynamics will only intensify. "Data centres need to be connected," Bansal said, noting that robust 5G networks are needed to link the massive server farms powering AI workloads.