By Light Reading
Publication Date: 2026-03-10 08:30:00
A still unanswered and rarely asked question in telecom is why a chipmaking behemoth like Nvidia sees interest in the 5G and 6G radio access network (RAN) business. It has, since early 2024, encouraged the industry to view its graphics processing units (GPUs), the hardware trainers of large language models (LLMs), as a dual-purpose option for RAN workloads and AI inferencing in telco networks. Then, in October last year, it plowed $1 billion into Nokia, which simultaneously revealed a RAN roadmap based on Nvidia’s chips. Very little of this makes obvious sense.
The deal with Nokia is taken by at least one industry veteran and another respected analyst as an unalloyed positive, a sign “the telecom industry is worth investing in.” But is it, seriously? Nvidia reported sales of about $68.1 billion in its most recent quarter, almost twice as much as the whole market for RAN products makes in a year. Only a small fraction of this amount is captured by the designers of baseband silicon. For a company of Nvidia’s size, it hardly seems worth the pain of engagement with an industry that is routinely criticized for its conservatism and inertia.
Nvidia is perhaps right to argue that GPUs for AI inferencing, when trained models are deployed, will need to be sprinkled across a bigger number of facilities, and not limited to a few giant data centers. This would reduce latency, the time a signal takes to go from…
