Nvidia is pushing AI-RAN — an architecture that runs radio access network workloads alongside AI jobs on the same GPU servers — in an attempt to place its hardware at the centre of mobile network infrastructure, Golem reports.
The pitch to carriers is efficiency. A cell site's baseband processing and a data centre's inference jobs are both massively parallel workloads; putting them on one pool of accelerators lets an operator buy hardware that is busy day and night rather than only when people are making calls. Nvidia has promoted that argument alongside carrier trials and partnerships, replacing purpose-built telecom boxes with racks of its servers.
Golem's framing is sceptical, and the title is pointed: this is an attempt to co-opt the mobile network business. Radio access networks are the most capital-intensive and most proprietary part of a mobile operator's estate, and the vendors who supply them — Ericsson, Nokia, Huawei — have held that position for decades. A GPU vendor that becomes the common substrate beneath both the network and the operator's AI ambitions would sit upstream of every layer above it.
Operators have heard versions of this promise before. Each generation of network technology arrived with a case for commoditisation and left with a new lock-in. The question AI-RAN raises is whether the compute layer Nvidia wants to own dissolves the old vendor boundaries or merely moves them up a level.
For Nvidia, the timing is convenient: telecoms is one of the few large infrastructure markets where its data-centre dominance has not already translated into a position.




