- Nvidia historically has not been a big player in wireless
- The company sees an opening in 5G and even 6G
- GPUs in wireless networks is its goal for now, but the company is not talking about chips in handsets
A few years ago Nvidia was better known for its graphical processing units (GPUs) targeting the gaming industry rather than its artificial intelligence (AI) chips. Of course, that’s all changed now, and Nvidia is seemingly everywhere all at once.
That includes wireless, although its entry is relatively new. Nvidia wasn’t a household name when 3G and 4G were deployed, but it’s jumping into the fray for the 5G and 6G eras.
The company already has one notable customer for its Aerial product. Japan’s NTT DoCoMo is the first-ever telco in the world to deploy a high-performance 5G virtual radio access network (vRAN) from Fujitsu built on the Nvidia Aerial vRAN stack and Nvidia Converged Accelerators.
DoCoMo expects the solution to reduce total costs by up to 30% and lower the power consumption at base stations by up to 50%.
Earlier this year, Nvidia released the second edition of its “State of AI in Telecommunications” survey, which queried more than 400 telecom industry professionals around the world. No surprise there, the survey showed enthusiasm for both generative AI and AI in general is booming, with telecom executives expecting AI to improve both revenues and cost savings.
“We’re obviously bringing generative AI to the telcos to help them transform their operations and drive out costs,” commented Chris Penrose, global head of business development for the telco space at Nvidia, on the sidelines of the Mobile Future Forward event last week.
Penrose spent about 30 years at AT&T, including as president of its IoT business, before joining edge AI startup FogHorn in 2020. He joined Nvidia in 2022 after Johnson Controls announced it was acquiring FogHorn.
During his time at AT&T, the operator was connecting a lot of “things” but also eyeing the next step: What to do with all those things once they’re connected. “I had the opportunity to work with some of these AI concepts early on,” he noted.
One of those was the use of sensors on video cameras so that they could count the number of people entering retail stores and make adjustments for security purposes and the employee-to-customer ratio.
No devices for Nvidia for now
Nvidia works with hundreds of telcos around the world in various capacities. For its AI solutions, it’s focused on improving the customer experience and driving operational efficiencies in the network. But it’s not in devices.
Rival Qualcomm often talks about how on-device AI enhances privacy and security by keeping it on the handset, reducing reliance on cloud-based solutions.
Nvidia isn’t doing that and that’s OK. “We’re good with AI everywhere,” Penrose quipped.
AvidThink founder Roy Chua said he doesn’t know if Nvidia will ever go down the handset chip path on its own again. Nvidia tried in the past with the Tegra platform, he said. That was around 2009/2010 and it didn’t quite take off. The platform ended up in automobile infotainment systems and gaming devices.
Now, Nvidia reportedly is collaborating with MediaTek, a strong supplier of silicon for handsets, on a joint PC solution to be available in late 2024/early 2025. “Could that relationship be expanded to handsets? Who knows?,” Chua said.
AI everywhere
Whether AI is in the handset or the network isn’t an either/or question, as they serve different purposes.
AI in the phone is about changing the background of a photo or using AI to map out a journey. AI in the network is more about how the network performs to make the handset work better or to save energy by turning down base stations in off-peak hours, said Recon Analytics analyst Daryl Schoolar.
“Because they do two different things, you’re going to want AI in both places,” he said. “Just because I put salt in my salad doesn’t mean I’m not going to put salt on my French fries.”
The sales pitch he’s heard for putting GPUs in the base station is it’s a more efficient way of doing the processing for 5G Massive MIMO in mid-band spectrum than Intel’s solution.
Prospects for the RAN
In terms of Nvidia’s broad designs on the RAN market, it’s competing with the likes of Intel, Qualcomm, Marvell and Huawei, as well as custom silicon manufacturers, like the startup EdgeQ that's currently focused on small cells, Chua said. However, Nvidia is touting its GPU as more general purpose and not only able to serve specialized RAN functions and any AI-driven optimizations, but also double as additional computing power for other AI workloads, he said.
“The reality is that AI acceleration will be needed on the handset, in the RAN, and upstream of that as well,” Chua said. “The AI accelerators will power predictive AI models to optimize resource utilization (spectrum, power), improve quality of service and overall reliability. And similarly, the AI accelerators can be used for generative AI functions like image enhancement, voice processing, other content generation and enhancement, on-device reasoning and planning.”
However, some analysts question whether operators are ready to embrace Nvidia’s GPUs.
Joe Madden, chief analyst at Mobile Experts, said the Nvidia solution looks great from a technical point of view, “but I don’t think it fits the business model that the operators and hyperscalers prefer. Either the solution or the business model will need to evolve before it can get traction,” he told Fierce.