Fujitsu unit 1Finity launches AI-Native VRAN to run AI jobs on phone tower hardware

What was announced

Fujitsu's telecom equipment arm, 1Finity, says its AI-Native VRAN is now available. The full-stack platform is designed so a single compute footprint at a mobile tower or regional edge site can carry radio traffic and local AI inference at the same time, turning carrier infrastructure into something that can host paid workloads rather than just pass calls and data.

According to a Fathom News report republished on dev.to, the stack is built on NVIDIA Aerial, a software platform that uses GPUs and networking software to virtualize radio access workloads in place of some fixed-function hardware. 1Finity named no customers and published no throughput numbers, so the announcement currently rests on the vendor's own description.

Why carriers might care

Carriers have spent years testing open RAN and virtual RAN to reduce vendor lock-in and lower upgrade costs, with mixed results, because performance and power trade-offs limited broad rollouts. The new pitch changes the argument. Instead of selling vRAN as cheaper plumbing, vendors are offering it as a route to edge AI revenue, a story aimed at chief executives rather than network engineers.

The underlying logic is utilization. Tower and edge sites already have power, cooling, backhaul and real estate. If those racks can run AI jobs when radio demand is light, the operator earns more from assets it already owns. Edge scenarios with strict latency requirements, such as factories, campuses and smart-city deployments, are the natural fit, since enterprises increasingly want data processed nearby rather than in a distant cloud region.

The engineering catch

Radio networks depend on predictable timing, and AI workloads arrive in bursts and draw heavy power. According to the report, the central technical challenge is scheduling both together so that AI inference never takes away the timing guarantees the radio side needs to keep calls and data sessions stable. Mixing the workloads also raises harder questions about latency, heat, uptime and workload isolation in environments that were never designed for experimentation.

The most likely early deployments, the report suggests, are private wireless networks in factories and ports, where one operator controls the radio environment and can match AI demand to a specific site. Public mobile networks are a far larger opportunity but far less forgiving of anything that disrupts service.

Nvidia's angle

For NVIDIA, the report notes, telecom is a rare infrastructure segment large enough to take serious GPU volume beyond the hyperscale cloud providers. Aerial gives the company a path into carrier budgets without waiting for every operator to build a standalone AI cloud of its own.

Why it matters

The launch is a concrete signal of where edge AI infrastructure is heading: into the radio access network itself. If 1Finity's claims hold up, phone towers become a distributed compute layer that enterprises can rent for low-latency AI, and carriers gain a second revenue stream from hardware they already run. But the announcement stops short of the evidence buyers need, with no named operators and no benchmark figures, and the scheduling problem it must solve is real. Until deployments show up outside labs, AI-Native VRAN is best read as a credible direction for telecom infrastructure rather than a proven product.