Ericsson: 6G Development Is Now Built Around AI From the Ground Up, Not as an Add-On

By Hector Herrera | May 10, 2026 | Telecom

6G is gaining commercial momentum — and the reason, according to Ericsson's May 2026 technical brief, is that AI is being designed into network architecture from the beginning rather than grafted onto existing radio systems the way it was with 5G. That structural difference, Ericsson argues, is what closes the gap between the AI-in-networks promises operators have heard for years and the operational cost savings they've been waiting to actually see.

Context

5G networks deployed AI as a layer on top of existing radio access network (RAN) infrastructure. The underlying architecture — fixed spectrum allocation, static radio configurations, rule-based traffic management — was built for human-managed networks. AI tools were then applied to optimize within those constraints. The result has been modest efficiency gains but nothing like the transformation vendors promised.

6G is being architected differently. Ericsson's technical brief, published ahead of MWC 2026, describes a network where AI inference is native to the radio layer — where spectrum decisions, congestion prediction, and energy management are made by AI models running at the edge in real time, not handed off to centralized compute after the fact.

What AI-Native 6G Actually Means

In a traditional RAN (Radio Access Network — the infrastructure connecting devices to the network backbone), spectrum is allocated based on pre-configured rules and manual optimization cycles. An AI-native RAN changes this in three concrete ways:

Dynamic spectrum allocation: AI models can predict which cells will face congestion 30–60 seconds ahead and pre-position spectrum resources accordingly, reducing dropped connections and improving throughput without adding hardware.

Real-time energy optimization: 5G base stations consume significant power even during low-traffic periods. AI-native architectures can power down individual radio elements during idle windows and spin them back up before demand spikes — a task that requires millisecond-level inference that rule-based systems can't achieve.

Predictive fault detection: AI-native networks can identify degrading hardware signatures before outages occur, enabling maintenance crews to replace components proactively rather than responding to failures.

These aren't theoretical — they're capabilities that Ericsson and other vendors have demonstrated in controlled environments. The AI-native 6G architecture is designed to make them operational at scale.

What Operators Are Driving

Ericsson's brief cites survey data showing that 85% of telecom operators name opex (operating expense) efficiency as their primary AI deployment objective. That's a meaningful data point: operators are not deploying AI in networks primarily for revenue growth or service differentiation — they're deploying it to cut costs on networks that are increasingly expensive to run.

AI data center power demand, which operators must provision for themselves and for hyperscaler tenants, is increasing those operational costs. An AI-native 6G architecture that dynamically manages energy at the radio level is directly responsive to that pressure.

Why This Matters

For the telecom industry: The 5G cycle disappointed many operators on ROI. 6G represents a second chance, but only if the efficiency gains are real and deployable at scale. Ericsson's argument — that AI-native architecture is the structural change that makes those gains real — is a significant strategic claim that will be tested as commercial 6G deployments begin in the 2028–2030 window.

For enterprise customers: AI-native networks will eventually enable network slicing (dedicated, guaranteed bandwidth for specific applications) and ultra-low latency services that current networks can only approximate. Industries like autonomous logistics, remote surgery, and real-time industrial control have use cases that require those guarantees.

For equipment vendors: Ericsson, Nokia, Huawei, and Samsung are competing to define what AI-native 6G looks like. The technical specifications getting locked in now — in standards bodies like 3GPP — will determine which vendors' architectures become the foundation of the next decade of global network infrastructure.

The China Dimension

Huawei's AgenticCore stack, announced in late April 2026, represents China's approach to the same problem: AI embedded at the telecom network layer for autonomous operations. With Chinese operators deploying Huawei equipment domestically and in emerging markets, the 6G standards competition is also a geopolitical contest over which architectural approach becomes the global baseline.

Ericsson's brief is partly a technical argument and partly a market positioning document — making the case that the Western-led standards track for AI-native 6G is further along and more operationally credible than alternatives.

What to Watch

The 3GPP Release 20 and Release 21 standards processes, which will formalize key AI-native 6G specifications, are the critical milestones. Watch for commercial 6G trials from South Korean and Japanese carriers in 2027, which will be the first real-world stress tests of these architectures. Closer term, operator capex guidance in Q2 and Q3 2026 earnings calls will show whether AI-native network investment is actually accelerating or still largely aspirational.

Source: Ericsson Blog — 6G Gains Momentum MWC 2026