T-Mobile and Ericsson Claim World First With AI-Native Scheduler on Live 5G Advanced Network

By Hector Herrera | May 15, 2026 | Telecom

T-Mobile and Ericsson have moved an AI-native radio access network (RAN) scheduler into large-scale commercial trials on live 5G Advanced traffic — a claimed world first — delivering up to 15% downlink throughput gains and close to 10% spectral efficiency improvement over the rule-based systems that have governed cellular networks for decades. If the performance holds at commercial scale, it would mark the first time a deployed neural network replaces the scheduler at the core of how a major U.S. carrier allocates wireless capacity.

The significance goes beyond a benchmark number. Rule-based RAN schedulers are the invisible engine of mobile networks — they decide in microseconds which users get bandwidth, in what order, and at what transmission parameters. Replacing that logic with a neural network that predicts rapidly changing radio conditions in real time is a fundamental architectural shift, not an incremental upgrade.

What the Technology Actually Does

The AI-native scheduler Ericsson is trialing sits inside the radio base station hardware itself. Rather than following deterministic rules written by engineers — "give this user bandwidth if signal quality exceeds this threshold" — the neural network continuously learns from live radio conditions and predicts the optimal scheduling decision for the next transmission window.

According to Ericsson, the system incorporates AI-native link adaptation alongside the scheduler. Link adaptation controls how aggressively a base station encodes data — a setting that traditional systems update slowly based on feedback reports. The AI version predicts the right encoding parameters before that feedback arrives, shaving latency out of the control loop.

The combination of smarter scheduling and faster link adaptation is where the performance gains compound.

Scale of the Trial

This is not a lab demonstration or a controlled pilot in a single stadium. The trial runs on live network traffic across a substantial number of T-Mobile base stations, putting real user data — not synthetic loads — through the AI scheduler. That distinction matters because network conditions in the wild are far messier than anything a controlled test generates: interference from adjacent cells, users moving between coverage zones, sudden traffic spikes.

Running on real traffic with real users while claiming measurable throughput improvements shifts this from a research result to an operational claim that will be scrutinized by competitors and regulators alike.

Why the Numbers Matter to Competitors

Spectral efficiency — how much data a network can move per unit of radio spectrum — is the foundational constraint on carrier performance. Spectrum is finite and expensive. A 10% spectral efficiency gain is equivalent, economically, to unlocking 10% more of a carrier's licensed spectrum without purchasing any.

AT&T and Verizon are both investing heavily in AI RAN research, but neither has announced a live commercial trial at this scale. If T-Mobile achieves commercial deployment in Q3 2026 as targeted, it would hold a measurable performance advantage during a period when all three carriers are competing aggressively on 5G mid-band coverage quality.

The competitive window may be short — Ericsson will eventually offer similar capabilities to other customers — but being first into large-scale deployment typically means first to work through operational issues, and first to build the training data advantages that make AI systems improve over time.

The Broader AI RAN Trend

Ericsson is not the only vendor moving in this direction. Nokia has disclosed AI RAN research, and Samsung — a major supplier to T-Mobile — has an active AI RAN roadmap. The O-RAN Alliance, the industry body standardizing open radio interfaces, has been working on AI/ML integration specifications for the past two years.

What Ericsson's trial demonstrates is that the technology is ready to leave the specification document and run on production hardware. The question for the industry is whether proprietary AI schedulers running on vendor-specific silicon will coexist with or eventually dominate open, standardized approaches.

What to Watch

The commercial deployment target is Q3 2026. Watch for T-Mobile's network performance metrics in quarterly reports — throughput per MHz is the number that will either validate or complicate Ericsson's claims. Also watch for AT&T or Verizon announcements of competing AI RAN trials, which would likely come within months if T-Mobile confirms commercial rollout.

Source: Ericsson press release