Nokia has taken its biggest step yet to building an AI-RAN base station based on Nvidia compute, announcing that it will design a base station product based on a new Nvidia platform – the Aerial RAN Computer Pro (ARC-Pro). For the first time, Nokia said it is committing to developing a solution in which Layer 1 RAN processing is carried out in the Nvidia GPU, with no additional or separate accelerator. Nokia said the strategic partnership would add, “Nvidia-powered, commercial-grade AI-RAN products to Nokia’s industry-leading RAN portfolio.”
The two companies will build a design integrating Nokia’s anyRAN cloud RAN software and the Nvidia ARC-Pro into Dell PowerEdge servers, with T-Mobile U.S. confirming it will conduct trials of the design in 2026.
The news dropped just as Nokia and Nvidia announced the chip and AI company would be making a $1 billion investment in Nokia, giving it a 2.9% stake in the company. While much of the strategic part of that announcement focussed on the potential for Nokia to drive into data centre networking, AI and cloud market, the companies also said, “Nokia intends to accelerate development of Nokia’s 5G & 6G RAN software to run on NVIDIA’s architecture.”
The base station design announcement takes Nokia deeper into a direct integration with Nvidia, for the first time porting its RAN software, including the most processor-intensive part known as Layer 1, onto the GPU platform. Nokia previously confirmed in February that it was working with Nvidia and T-Mobile to explore the use of GPUs for RAN acceleration, but remained uncommitted to its roadmap at that point.
GPU is used for L1 including acceleration, for which in this case no SmartNIC is needed. Nokia anyRAN common software is running on ARC.
In January this year, Nokia and Nvidia were part of a SoftBank trial that integrated Nokia’s anyRAN cloud software with an Nvidia-based server. On that occasion, acceleration for L1 RAN software was carried out on Nokia’s SmartNIC plug-in, with L2/L3 on the Nvidia CPU. The GPU part of the Grace-Hopper platform was left free for other workloads.
“Unfortunately Nokia does not yet support GPU acceleration,” Softbank’s Ryuji Wakikawa, VP Head of Research Institute of Advanced Technology, said at the time. Aji Ed Head of Partner Cloud RAN Solutions, Nokia, said that Nokia was using the plug-in acceleration as a bridge to enable telcos to assess AI-RAN potential.
This new design now addresses that gap, and brings a new potential chip supplier into Nokia’s domain, as well as giving Nvidia one of the big three RAN vendors as a potential long term customer.
A spokesperson confirmed to TMN: “[The] GPU is used for L1 including acceleration, for which in this case no SmartNIC is needed. Nokia anyRAN common software is running on ARC.”
Reworking Nokia’s RAN software for operation in a GPU environment represents a n evolution from its current strategy of accelerating lower layer RAN with a Marvell ARM-based chipset. It will also commit Nokia more fully to Nvidia’s vision of using spare or shared GPU resources for non-RAN tasks and workloads, or to run workloads that can assist the RAN.
The business model for AI-RAN is still uncertain, with many operators yet to be convinced that the multi tenanted GPU model is one that will fly commercially, doubting it will give enough returns to pay back what are expected to be the increased cost of rolling out GPU-based base stations on a large scale basis. As for AI for the network, in which a GPU platform might be leverage to run AI workloads that manage and optimise the RAN, operators and other chip vendors point out that such workloads do not need the overhead of the GPU to function efficiently.
Clearly those involved in this announcement see the partnership as a big deal. The companies brought out CEO quotes for its press release, from Jensen Huang, Justin Hotard and Michael Dell, as well as John Saw, president of technology and chief technology officer at T-Mobile.
Nokia has said it will embed Nvidia’s ARC-Pro “at the heart of a new solution” that will provide the ability to support distributed edge inferencing, as well as RAN workloads. It also said it would accelerate the availability of its 5G and 6G RAN software on the NVIDIA CUDA platform.
Strategically, given the timelines and likely workloads, the companies positioned the development of AI-RAN products as a 6G enabler, describing the development as 6G-ready, meaning that it will have the processing power to support 6G use cases, as well as an AI-native RAN itself.
Nvidia describes ARC-Pro as a 6G-ready accelerated computing platform that combines connectivity, computing and sensing capabilities, available as a reference for manufacturers. At time of writing there were no public materials on the ARC-Pro platform, and Nvidia said it wasn’t providing any more details at this time.
Comments
0