Ericsson AI RAN Delivers 10% Spectrum Efficiency Boost

Twenty percent faster downlink speeds. Ten percent better spectrum efficiency. No new hardware required — just smart software that learns from network data. Ericsson made it reality in 2025, and now in 2026, the numbers look even more impressive.

⚡ AI-Native Link Adaptation: The breakthrough

Bell Canada became the first carrier worldwide to test Ericsson's AI-native link adaptation in real-world conditions. Not lab tests or simulations — they deployed it on live RAN infrastructure with actual antennas and real traffic. The numbers tell the story. Spectrum efficiency improved by 10%. Downlink throughput jumped up to 20%. The kicker? Zero hardware changes.

What is AI-native Link Adaptation: An algorithm that dynamically adjusts network parameters — modulation, coding, power — in real-time based on signal quality and interference conditions.

Ng Thiaw Seng from Ericsson's Strategic Network Evolution team explained it clearly at the ET5G Congress 2026: "Using AI algorithms for link adaptation at the radio interface, we immediately improved cell edge bitrate and spectrum efficiency by 20%." This marks the first time AI runs natively in the baseband unit, not as an add-on.

📊 Why Spectrum Efficiency Matters

Spectrum is limited and expensive. Every MHz costs millions at auctions. If you can squeeze more bits into the same space, it's like having more spectrum without buying it.

10% Spectrum Efficiency Improvement

20% Throughput Increase

12M 5G FWA Subscribers Globally

In practice, this means less network "overbuilding" to support uplink-heavy AI applications. The more efficient your RAN, the fewer cell sites you need for the same coverage and capacity.

The Technology Behind the Numbers

The key lies in real-time adaptation. Traditional systems adjust parameters based on static rules or slow adaptation cycles. AI-native link adaptation continuously monitors: - Signal-to-Interference-plus-Noise Ratio (SINR) - Modulation and Coding Scheme (MCS) selection - Channel quality variations - Interference patterns It adjusts all these parameters in milliseconds. Instead of waiting 100ms to detect changes, it recognizes and responds in just a few ms.

🧠 Neural Accelerators in RAN: The Next Step

Embedding AI in RAN isn't just a software upgrade. It demands processing power — enter Graphics Processing Units (GPUs) or specialized AI chips in radio units.

"If it's related to the air interface and needs microsecond or millisecond response, then processing must happen at the radio. Otherwise, the entire investment won't pay off."
Ng Thiaw Seng, Ericsson

This means we'll see neural accelerators in every baseband unit. General-purpose tasks can be processed in cloud facilities with more powerful hardware. But for real-time RAN optimization, AI must live at the antenna.

The Math of Efficiency

Spectrum efficiency is measured in bits per second per Hz (bps/Hz). With 100 MHz bandwidth delivering 1 Gbps throughput, your spectrum efficiency is 10 bps/Hz. With AI-native link adaptation, that same 100 MHz can deliver 1.1-1.2 Gbps. Doesn't sound dramatic, but multiply that across thousands of cell sites and millions of users.

🚀 From 5G Advanced to 6G

Research happens at Ericsson's R&D center in Ottawa — the same facility working on 6G. This connection is deliberate. AI-native link adaptation serves as proof of concept for how future networks will operate.

Open RAN Compatibility

Works with standard Open RAN interfaces, enabling deployment in multi-vendor environments.

Real-time Optimization

Millisecond-level adaptation to radio environment changes — much faster than existing systems.

Ericsson is a founding member of the AI-RAN Alliance, which includes NVIDIA, Samsung, and other major players. Goal: make AI a standard part of RAN, not an optional feature.

What This Means for Carriers

For carriers like Verizon or T-Mobile, 10% spectrum efficiency improvement means: - Fewer new cell sites for the same capacity - Better performance in poor conditions (cell edge) - Lower energy footprint per MB of data But there's a cost. Neural accelerators, software licensing, integration complexity. The question is whether benefits justify the investment.

🎯 The Challenges That Remain

Despite impressive test performance, questions remain. First, results come from controlled environment testing with Bell. How will it perform in dense urban areas with heavy interference? Second, energy consumption. AI requires processing power — meaning energy. If neural accelerators consume an extra 50W per baseband unit, multiply that across thousands of sites. Third, complexity. Every new AI component adds a layer that can fail. Network troubleshooting becomes more complex when you need to understand what the AI did.

Real-World Data from the Field

Live trials with Bell showed benefits are greatest in challenging scenarios: - Poor signal quality areas (cell edge) - High interference environments - Rapidly changing channel conditions (moving vehicles) In ideal conditions, the difference is smaller. This makes sense — AI shines when traditional algorithms struggle.

📈 The Economic Bet

Spectrum efficiency isn't just technical — it's economic. Every MHz carriers buy at spectrum auctions costs millions. In recent US auctions, 5G spectrum reached $372 million total. If you can extract 10% more capacity from the same spectrum, it's like earning back 10% of your investment. For major carriers, this could mean millions in annual benefits. But AI neural accelerators cost money. Ericsson hasn't announced pricing, but estimates suggest they'll add $2,000-5,000 per baseband unit. With thousands of sites per carrier, that's serious investment.

ROI Calculator

Return on investment depends on: - Spectrum cost per MHz - Number of baseband units - AI accelerator costs - Operational savings from better efficiency According to analysts, break-even point for most carriers is 18-24 months. If AI equipment lasts 5+ years, benefits will be substantial.

🔮 The Future: AI Everywhere

AI-native link adaptation is just the beginning. The next generation will bring AI to every RAN component: **Predictive Maintenance:** AI algorithms that predict hardware failures before they occur **Dynamic Beam Management:** Automatic beamforming pattern optimization based on user density and movement patterns **Energy Optimization:** Smart shutdown/wakeup of radio units based on traffic patterns **Interference Mitigation:** Real-time interference detection and neutralization In 6G, specifications will include AI from the design phase. It won't be an add-on — it'll be a core requirement. The question isn't "if" this will happen, but "when" it becomes mainstream. Early adopters will gain competitive advantage — laggards will struggle to catch up.

AI RAN Ericsson spectrum efficiency 5G networks neural accelerators RAN optimization telecom AI network efficiency

Sources:

How Ericsson's AI RAN Achieves 10% Spectrum Efficiency Gains in Real Network Deployments