Samsung and Verizon complete first AI-powered ISAC trial on vRAN

Verizon’s ISAC crowd-sensing test ran on ordinary commercial 5G hardware

In sum – what we know:

• Virtualized 5G trial – The Dallas test ran entirely on commercial 5G infrastructure and Samsung’s AI-powered vRAN software, with no dedicated sensing hardware added.
• Signals as sensors – ISAC reads changes in ordinary radio signals to generate near-real-time crowd-density heatmaps, avoiding cameras and lidar entirely.
• Scaling unanswered – One base station and six localized phones at a single staged event, with no accuracy, cost, or independent validation disclosed.

Samsung and Verizon are working hard on the next generation of cellular connectivity, and AI is seemingly at the core. The two companies announced that they’ve completed an Integrated Sensing and Communication (ISAC) field trial at a major international soccer fan event in Dallas, Texas, and they’re calling it one of the industry’s first practical implementations of a key 6G technology. More specifically, it’s apparently the first AI-powered ISAC crowd-detection trial to run on a virtualized radio access network, or vRAN, which replaces dedicated networking hardware with software running on general-purpose servers.

To be clear, this was not an actual 6G deployment. There is no 6G network to deploy on, after all. The trial ran entirely over commercial 5G infrastructure using 5G Advanced-related capabilities. Samsung and Verizon are making a case that software-led 5G networks can serve as a foundation for capabilities the industry expects from 6G, years before any full 6G radio-interface standard from 3GPP or the ITU is finalized.

The capability itself leans heavily on AI. The trial was less a radio demonstration than a test of AI inference at the wireless edge, using telecom signals as sensor inputs.

ISAC demonstrations do predate this trial — NTT and Sophia University showed pedestrian-flow sensing from commercial base-station signals in 2025, Ericsson demonstrated drone detection at its Plano, Texas headquarters in February 2026, and Huawei has been prototyping ISAC in the terahertz band since 2023. What’s new here is the combination — AI-powered crowd detection running on a virtualized RAN.

How ISAC turns cellular signals into a sensor

ISAC combines two jobs in a single network function. The network carries communications traffic, as usual, and simultaneously senses the physical environment around it. There’s no separate camera, lidar unit, or purpose-built radar involved. Instead, the system analyzes changes in radio signals caused by people and objects moving through an area. Verizon describes the concept as turning the cellular network into a massive, distributed radar-like sensor.

In Dallas, the output was a near-real-time heatmap showing crowd density and movement changes across the venue. Both companies say the system measured aggregate crowd counts rather than identifying individuals.

Dallas configuration relied on existing 5G hardware

The setup was fairly modest. A Verizon Cell-On-Wheels carried a Citizens Broadband Radio Service (CBRS) radio unit using 40 MHz of spectrum in the 3.5 GHz range. The radio unit acted as a sensing receiver, collecting signal variations from commercial 5G devices, and six localized Samsung Galaxy smartphones supplied the raw radio-signal data.

On the software side, Samsung combined its commercial AI-powered vRAN software, its virtualized core (vCore) software, and an ISAC application. All of it ran on Samsung’s “Network in a Server,” an edge-AI platform that packs network functions and accelerated compute into a single box. The platform processed the signal data locally and generated heatmaps viewable on commercial phones and a tablet.

Samsung’s key commercial claim is that the whole thing ran on a commercial 5G connection without operators replacing network equipment or bolting on dedicated sensing hardware. That’s a claim worth taking seriously, though it’s also worth remembering that one base station and six localized phones, staged live at a single event, is a long way from a stadium-scale deployment.

AI inference at the wireless edge

Raw radio-channel data doesn’t tell an event operator anything useful on its own. AI is what converts those signal changes into operational conclusions — crowd distribution, movement patterns, bottlenecks, and emerging congestion. That’s where the trial actually earns its “first” framing, and it reflects a broader industry shift toward so-called AI-native networks, meaning cellular infrastructure that runs AI workloads and uses AI for operation and sensing rather than merely shuttling data to distant clouds.

Placing accelerated compute at the network edge, right next to the radio infrastructure, is what makes this workable for time-sensitive tasks. Crowd management is the demo case, but Samsung points to robotics and autonomous-vehicle sensing as the sorts of applications where latency actually matters. Verizon, for its part, credits its earlier vRAN investment with making the trial possible at all.

“Our early investments in flexible, software-driven vRAN architecture are paying off today, giving us the agility to engineer and trial groundbreaking 6G concepts like ISAC over our existing infrastructure,” said Verizon CTO Yago Tenorio. The pitch, in other words, is that virtualized networks add capability through software rather than hardware replacement.

Verizon’s 6G strategy is explicitly tied to AI workloads more broadly, particularly AI-enabled wearables like untethered smart glasses, which Verizon says will demand a “more balanced uplink-downlink model,” generating as much uplink data as they download, while also requiring low-latency inference. The company separately demonstrated an AI Sports Companion prototype built on Meta AI glasses, its edge network, an NVIDIA DGX Spark system, a locally optimized small language model, and SportsDataIO statistics, delivering spoken live stats and win probabilities. Essentially, the idea is that future networks will need to run inference close to the user, and the radio network itself becomes one of the sensor inputs feeding it.

Privacy and deployment questions remain open

ISAC avoids cameras, but it introduces a different form of environmental monitoring by inferring activity from radio signals. Samsung and Verizon emphasize that the Dallas deployment measured aggregate density. Fair enough for this trial — but the privacy outcome of future systems depends on granularity, data retention, notice and consent, access controls, and whether sensing results can be linked to other data. ETSI’s ISAC work has already flagged a long list of privacy and security issues, which suggests the safeguards will need to be built alongside the technical standards, not after them.

The disclosed test was also narrow. One base station, 40 MHz of CBRS spectrum, six localized smartphones. Neither company disclosed accuracy rates, false-positive or false-negative rates, range, performance in a dense uncontrolled crowd, energy consumption, cost, or any independent validation. Running the prototype on existing 5G infrastructure genuinely could lower adoption barriers, and that’s the strongest card Samsung and Verizon hold. Whether it scales is the open question — one that hinges on standards, spectrum, network synchronization, edge-compute capacity, and whether the public accepts networks that watch as well as connect.