Nvidia targets 6G towers with new GPU-powered radio unit chip
Nvidia is developing a GPU for 6G radio units (RUs), aiming to replace custom ASICs in massive MIMO systems, which would expand its AI-RAN strategy. This move addresses power-intensive Layer 1 functions directly within RUs, a critical component of 6G infrastructure. Nvidia asserts that its GPUs are becoming essential as RU sophistication and processing demands increase.
Key Takeaways
- The new chip targets 6G ultra-MIMO configurations with up to 1,024 transmitters and receivers, which require 32x more processing power than current systems.
- Nvidia aims to replace traditional ASICs in the radio unit's physical layer (Low PHY) with programmable, software-defined GPUs.
- The hardware is designed for constrained environments, targeting power consumption below 100W and operating temperatures up to 100 degrees Celsius.
- The initiative leverages the Aerial architecture and CUDA platform, providing a unified software environment for approximately 6 million developers.
- Marvell confirmed its Octeon baseband processors will be enhanced to work directly with Nvidia GPUs for concurrent 5G/6G and AI workloads.
Why It Matters
The shift from custom silicon to programmable GPUs in the radio unit represents a fundamental change in wireless infrastructure, allowing operators to repurpose tower-top hardware for edge AI tasks beyond simple connectivity. By unifying the compute stack from the data center to the antenna mast, Nvidia is positioning itself to sideline traditional telco specialized hardware vendors like Ericsson and Nokia. This transition is essential for 6G's high-spectrum bands (7GHz), where fixed-function ASICs may lack the flexibility to manage complex beamforming and real-time AI modulation. Watch for the first field trials of integrated Nvidia-Marvell radio units in early 2027 to validate thermal performance in high-density urban environments.
Additional Context
The announcement follows the official launch of the AI-RAN Alliance at Mobile World Congress 2024, where Nvidia joined foundational members including Amazon Web Services, Arm, Ericsson, Microsoft, and T-Mobile. Per Nvidia filings in March 2026, the company deepened this commitment with a $2 billion strategic investment in Marvell Technology. This partnership focuses on integrating Marvell’s custom XPUs and optical interconnects with Nvidia’s NVLink Fusion platform to build 'AI factories' capable of handling the explosive growth in machine-to-machine traffic.
Relatedly, Nokia secured a $1 billion investment from Nvidia in late 2025 to accelerate its transition toward AI-native infrastructure. Per EE Times (March 2026), Nokia demonstrated a distributed GPU system at MWC 2026 that processed commercial 5G traffic alongside generative AI queries and live video captioning. These collaborative efforts suggest a broader industry pivot toward heterogeneous computing, where general-purpose GPU architectures manage the unpredictable requirements of 6G sensing and communications that traditional legacy systems were not designed to support.
Competitive pressure is also mounting from Intel, which launched its Granite Rapids-D processors in early 2026 to capture the virtual RAN (vRAN) market. However, per Light Reading reporting, Intel’s current roadmap lacks a dedicated radio unit component, potentially leaving a strategic gap that Nvidia’s RU-specific silicon is designed to exploit. With global RAN spending falling to approximately $35 billion in 2025 according to Omdia, the economic shift toward programmable, multi-purpose hardware may become the only viable path for sustaining future network densification.
Read full article at lightreading.com
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source