Nvidia has a radical new AI-RAN plan – a 6G radio unit chip - Light Reading

NVIDIA is advancing a radical AI-RAN strategy by developing a chip that will be directly integrated into future 6G radio units (RUs). While NVIDIA’s GPUs have previously been used in RAN servers and a... Read original →

NVIDIA embedding GPUs into 6G radio units fundamentally rearchitects RAN’s hardware-software boundary, shifting PHY-layer processing from fixed ASICs to CUDA-programmable logic. This directly undermines Marvell and Intel’s baseband dominance and forces a redesign of thermal/power systems due to heightened compute density at the RU edge. If reliant on Taiwan, China-based 3nm EUV foundries, geopolitical supply risks could inflate inventory costs by over 20%. Samsung and Nokia may counter by accelerating in-house AI accelerators to avoid CUDA lock-in, while MediaTek faces strategic tension over licensing its 5G IP stack. Within 18 months, if NVIDIA demonstrates superior energy efficiency versus ASICs in massive MIMO deployments, software-defined RAN will surge; otherwise, it risks repeating FPGA’s failed ASIC-replacement narrative.