Nvidia and Softbank to integrate AI-servers into telecoms cell towers

A telecoms network with servers integrated into the cell-towers to enable them to run AI networks is the aim of a collaboration between Nvidia and Softbank.

In trials, SoftBank has demonstrated that its Nvidia-accelerated AI-RAN technology achieves carrier-grade 5G performance while running AI inference workloads.

“Our cell towers used to just carry bits for telecommunication and internet surfing,” says Masayoshi Son Chairman of Softbank,  “now, with this network, we’ll densely connect them with each other and make them into one big neural brain and the infrastructure of intelligence for Japan.”

The beauty of the idea is that it will reduce latency by putting AI processing nearer to where it is used. This would enable future applications like autonomous cars, robot control and multimodal retrieval-automated generation.

For instance, for an AI robot to function properly, it typically must receive an input every tenth of a second, says SoftBank and, with cell towers located a few hundred metres apart in cities, an AI- enabled cell tower could deliver that.

SoftBank plans to put the technology on the market in the fiscal year running from April 2026 to March 2027.