T-Mobile and Ericsson Log 10% Efficiency Gains in AI-RAN Trial on Live 5G Advanced Network

T-Mobile and Ericsson reported that AI-native radio access network software — known as AI-RAN — delivered nearly 10% efficiency gains and up to 15% throughput improvements in large-scale commercial trials on T-Mobile's live 5G Advanced network. The trial moves AI-RAN from controlled lab environments into real customer traffic for the first time at this scale, and the results provide the commercial validation that has been missing from carrier conversations about AI network investment.

What AI-RAN Actually Does

Radio access network management — the software that controls how wireless towers allocate spectrum, manage handoffs, and serve thousands of simultaneous connections — has historically relied on static rules and parameter tables. A tower "knows" what to do in most conditions because engineers programmed those conditions in advance.

AI-RAN replaces static rules with machine learning models that adjust network parameters in real time. Instead of following pre-set rules for spectrum allocation, the system observes actual traffic patterns, predicts what's about to happen, and optimizes allocation dynamically. Instead of fixed handoff thresholds, the system learns when users are actually about to move between towers and prepares the handoff before latency degrades.

The theory has been sound for years. The gap has been live network validation — real traffic, not lab simulations — at scale.

The Trial Results

According to Telecoms Tech News coverage of the trial, T-Mobile and Ericsson's commercial-scale trial on T-Mobile's live 5G Advanced network produced:

• ~10% efficiency gains — the network does the same work with roughly 10% less spectrum and energy overhead
• Up to 15% throughput improvements — individual users get faster speeds, particularly in dense environments where the network previously struggled to allocate spectrum efficiently
• No new hardware required — the gains came entirely from AI-RAN software running on existing infrastructure

Ericsson has previously stated that AI can improve existing 5G network performance by 25–30% at full deployment scale, without hardware upgrades. The T-Mobile trial results are consistent with early-stage deployment numbers that would reach that range at full network integration.

The First AI-RAN Call

Separately, T-Mobile completed what it described as the world's first AI-RAN call — a live voice and data call running entirely on an AI-native radio access network — in Bellevue, Washington, in collaboration with NVIDIA and Nokia. The milestone is technically distinct from the Ericsson efficiency trial: it demonstrates that an AI-native RAN architecture can handle the full range of wireless communication functions, not just optimize parameters on an existing RAN.

The NVIDIA involvement is significant. NVIDIA's AI computing chips are increasingly being considered as the processing substrate for AI-RAN, running inference workloads directly in the network rather than sending data to centralized AI systems. If AI-RAN scales to full deployment, it creates a new class of NVIDIA customer — wireless carriers — buying GPU compute for network management rather than for AI model training.

What This Means for Carriers

The commercial case for AI-RAN comes down to a simple question: can you defer expensive hardware upgrades by improving performance through software? If 10% efficiency gains hold across the full T-Mobile network, the answer is yes.

Capital expenditure deferral is a powerful financial argument. A major carrier spending $5–8 billion annually on network infrastructure can defer significant capital by improving existing 5G performance through AI-RAN rather than building out new capacity. At scale, the software investment pays for itself by extending the useful life of existing hardware.

Competitive differentiation is the second argument. If AI-RAN delivers 15% throughput improvements in dense environments — stadiums, city centers, airports — the carrier deploying it can offer measurably better service in exactly the locations where premium customers are most likely to notice and switch.

The Broader AI-RAN Race

T-Mobile and Ericsson aren't alone. Nokia has been running AI-RAN trials with multiple carriers. Samsung has published AI-RAN results with US carriers. Qualcomm is developing AI-RAN processing specifically for its baseband chips. The space is moving from a theoretical conversation to active commercial deployment within a 12-month window.

The geopolitical dimension matters too. As Western carriers seek to reduce dependence on Chinese equipment vendors, AI-RAN software running on Western-designed hardware becomes part of the network security argument, not just the performance argument.

What to Watch

The key question is whether T-Mobile moves from trial to full network deployment of AI-RAN software on its Ericsson infrastructure — and on what timeline. A full-network deployment decision from T-Mobile would trigger similar decisions at AT&T and Verizon, which are running their own trials. Watch T-Mobile's Q4 2026 earnings call and any capital expenditure guidance that references AI network software, which would confirm whether the trial results are translating into deployment commitments.