The use of Nvidia GPUs in the radio access network would have massive implications for telcos and their suppliers.
After selling gazillions of its graphics processing units (GPUs) to the hyperscalers, Nvidia has been on the lookout for new types of customer, and telcos seem to be high on the list. It is hardly a surprise. The global market includes hundreds of operators with facilities that could house those GPUs, and the sector is desperate for growth. Could GPUs supporting AI applications at the network's "edge" provide the hoped-for boost?
Nvidia obviously says yes and has dangled some juicy numbers in front of the telcos like a zookeeper enticing the animals with some choice cuts of meat. A telco that bites will be able to generate $5 in inferencing revenues over a five-year period for every $1 it invests, the chipmaker claims. Besides using these GPUs for sales-generating AI, the telco will be able to use the spare capacity to operate its mobile network in what Nvidia calls AI-RAN (for radio access network).
This would be a seismic shift for the industry, with ramifications for many other suppliers. Today's RAN infrastructure is largely purpose-built for the mobile job by Huawei, Ericsson and Nokia. With virtual RAN, which still accounts for a relatively small share of the market, Ericsson and Nokia (although not Huawei) still contribute the radios while combining their RAN software with general-purpose chips and compute platforms. The Ericsson appliance gives way to a Dell server and Intel processor hosting the Swedish company's software.
An AI-RAN sponsored by Nvidia would shatter these arrangements. ARC-1, an appliance the company showed off earlier this year, comes with a Grace Blackwell "superchip" that would replace either a traditional vendor's application-specific integrated circuit (ASIC) or an Intel processor. Ericsson and Nokia are exploring the possibilities with Nvidia.
But they cannot simply take software written for purpose-built or virtual RAN deployments and pair it with an Nvidia GPU. Developing RAN software for use with Nvidia's chips means acquiring fluency in compute unified device architecture (CUDA), Nvidia's instruction set. "They do have to reprofile into CUDA," said Soma Velayutham, the general manager of Nvidia's AI and telecom business, during a recent interview with Light Reading. "That is an effort."
An AMD alternative?
If AI-RAN became the de facto RAN option, Ericsson and Nokia would probably end up relinquishing their roles in chip development for baseband units. They would still, however, be active in the development of radios. Intel and other chipmakers targeting the virtual RAN would similarly face a much smaller addressable market.
Of course, this remains highly unlikely. Many telcos are still unconvinced by the economic case for AI-RAN, and GPUs have a nasty reputation – however justifiable it is – as energy hogs. Any telco concerned about the lack of alternatives to Intel in virtual RAN is bound to be just as worried about overreliance on Nvidia in AI-RAN.
Could AMD be an option? It is certainly valued by NScale, a UK business with a GPU-as-a-service offer, as an AI alternative to Nvidia. "AMD's approach is quite interesting," said David Power, NScale's chief technology officer. "They have a very open software ecosystem. They integrate very well with common frameworks." So far, though, AMD has said nothing publicly about any AI-RAN strategy.
The other telco concern is about those promised revenues. Nvidia insists it was conservative when estimating that a telco could realize $5 in inferencing revenues for every $1 invested in AI-RAN. But the numbers met with a fair degree of skepticism in the wider market. Nvidia says the advantage of doing AI inferencing at the edge is that latency, the time a signal takes to travel around the network, would be much lower compared with inferencing in the cloud. But the same case was previously made for hosting other applications at the edge, and they have not taken off.
Even if AI changes that, it is unclear telcos would stand to benefit. Sales generated by the applications available on the mobile Internet have gone largely to hyperscalers and other software developers, leaving telcos with a dwindling stream of connectivity revenues. Expect AI-RAN to be a big topic for 2025 as operators carefully weigh their options.
Here is a selection of our AI-RAN stories from 2024:
12/3/2024 Nvidia bid to reshape 5G needs Ericsson and Nokia buy-in
11/15/2024 Softbank goes radio gaga about Nvidia in nervy days for Ericsson
11/4/2024 T-Mobile emerging as Nvidia's big AI cheerleader
10/24/2024 Verizon execs hint at making money from AI infrastructure
10/16/2024 Ericsson plots hardware-agnostic 6G but warns of revenue pain
10/7/2024 Vapor IO hopes for an Nvidia-powered AI push at the edge
9/25/2024 Dell goes big on Nvidia but lauds Intel in new telecom pitch
9/20/2024 Nvidia shows what open RAN could really look like with ARC-1
9/19/2024 Nvidia and T-Mobile just tore up Intel's virtual RAN rulebook
6/5/2024 Nvidia emerging as toughest Arm wrestle for Intel in 5G
4/10/2024 Telcos will applaud Intel's charge at Nvidia as AI hype goes imbecilic
3/18/2024 How Nvidia could threaten Ericsson and Nokia in 5G and 6G
You May Also Like
.jpg?width=700&auto=webp&quality=80&disable=upscale)
