- Nokia CTO and AI Officer Pallavi Mahajan says AI-RAN is all about decoupling hardware from software
- Regarding the vendor's claims about spectral efficiency gains, Mahajan says one of Nokia’s secret weapons is an asset called Bell Labs
- Nokia’s collaboration with Nvidia is another way it’s achieving the kinds of innovation the industry needs, according to Mahajan
When CTO and AI Officer Pallavi Mahajan presented Nokia’s AI-RAN vision last week, she said: “We are providing innovation at software speeds.”
What exactly does that mean? Fierce has been hearing this mantra a lot since Mobile World Congress 2026 in Barcelona: Telecom needs to move as fast as software. Mind you, telecom historically is not known for moving fast. But Nokia is putting a lot of emphasis on software these days.
So we asked Mahajan: Can telecom/wireless move at the speed of software? Like, is that even possible?
“I think so and we’re starting to see that,” she said.
What about those GPUs
Typically, when people look at Nokia’s AI-RAN strategy, they think about putting GPUs in the RAN.
Despite Nokia’s tight relationship and $1 billion investment from Nvidia, “that isn’t our thinking,” she said. “Our fundamental thinking behind AI-RAN was let’s decouple hardware from software.”
Similar to other industries, like servers, the minute you decouple hardware from software, innovation goes up.
That’s not to say hardware is no longer important. It is, but “we believe that software is where most of the innovation is going to be and we want to make sure that we capture that market,” she said.
Mahajan joined Nokia last fall from Intel, where she was general manager of Data Center and Artificial Intelligence Software. Prior to that, she worked at HPE and Juniper Networks. At Nokia, she also oversees Bell Labs.
She spoke with Fierce this week alongside Udaya Mukherjee, who leads Nokia’s RAN and Core technologies. He also worked at Intel for 25 years prior to joining Nokia about three months ago; he gets credit for architecting the first fully virtualized 5G infrastructure.
The role of Nokia Bell Labs
One of Nokia’s secret weapons is Bell Labs, according to Mahajan. Nokia launched its AI-RAN concept in October 2025, but Bell Labs has been working on it for the last four or five years, she said. That’s how Nokia can make its claims about 20% to 25% improvements in spectral efficiency, with a path to 2x spectral efficiency by 2028.
Traditionally, engineers would “tweak it here, tweak it there, manually” to achieve an improvement of 0.5 dB here or 1 dB there, Mukherjee said. When you introduce AI, “that’s a quantum leap, especially on a multi-user MIMO,” he said, noting that Nokia’s spectral claims have been proven in its Dallas lab and via trials in the Chicago area.
“What we are seeing is not like one algorithm or one model that we have brought in that gives us the line of sight to 2x. It’s a combination of many of these things, which have been independently tested in Bell Labs,” Mahajan said.
Nvidia reigns supreme
A big “aha” moment for Nokia was Nvidia’s RTX 4500/Blackwell-derived solution, which fits within existing power and thermal limits, an important consideration for Nokia’s AirScale deployments. That means brownfield operators can upgrade capacity by adding a new card to existing racks without changing cooling systems or altering power requirements, she said.
Prior to this, “if we were to talk of spectral efficiency, we would have said, ‘I'm going to develop a custom silicon which will provide 20%, 30%, 40% spectral efficiency.' But we are not saying that we will do that. Now we are saying you know what? Nokia today provides you 20 to 25 percent spectral efficiency,” she said.
By end of 2027, “we'll provide you 50% spectral efficiency. By end of 2028, we'll provide you 100 percent spectral efficiency. And in none of these conversations are we asking our customers to go about and upgrade their hardware. It's the same hardware that will also take them to 6G. That is what I mean by innovating at the pace of software,” she said.
In the U.S., Nokia Cloud AI-RAN is already in 5G field trials with T-Mobile, its lead trial partner. Outside the U.S., the Finnish vendor cites collaborations in various stages with operators including SoftBank, Deutsche Telekom, Telia, Orange, BT and Indosat Ooredoo Hutchison.
The 5G-to-6G bridge is where AI-RAN gets especially interesting. Take Integrated Sensing and Communication (ISAC), for example. It’s associated with the 3GPP’s move to the 6G standard, but it can be achieved using 5G technology, according to Mukherjee. “We are actually showcasing sensing right now,” he said.
“We are seeing a lot of interest in sensing,” especially from a defense use case point of view, Mahajan said. Sensing can be used, for example, to guard an area so no unauthorized intruders are allowed in, whether it’s drones or something else.
Analyst: Back off the GPU angle
CCS Insight analyst Ian Fogg said too much attention has focused on the GPU part of the Nokia/Nvidia initiative.
Granted, Nvidia’s role in Nokia’s AI-RAN proposition and Nvidia’s investment in Nokia points people to GPUs and hardware.
However, “those people forget that one of Nvidia’s key differentiators over others is the wide industry support for Nvidia’s CUDA software architecture,” he said.
Operators are looking to deploy a platform that’s cost effective. “Nokia is aiming to improve the performance and efficiency of its AI-RAN platform in the future. If Nokia can deliver software upgrades that boost performance, then an operator may be able to delay hardware upgrades on existing sites or avoid adding new cell sites because its existing hardware is able to support more traffic. Avoiding the need to add new cell sites is especially attractive for wireless operators because running more cell sites increases operational costs,” he said.
Has Nokia actually figured out a way to “move as fast as software” and if so, how?
Fogg said parts of Nokia have long been software-focused – in core networks, for example. The fact that Nokia last year moved RAN and core networks into the same business unit should help Nokia with the shift to increased use of software in its portfolio, he noted.
Two major AI-RAN trends
Mobile Experts founder Joe Madden said he sees two major trends embedded in the term “AI-RAN.”
When it comes to AI-for-RAN, or optimization of the network using AI, “Ericsson is clearly in the lead with commercial offerings and proven performance in field trials. Nokia, on the other hand, is leading on AI Edge Computing using GPU horsepower in the network to run AI inferences and develop new revenue for customers. Comparing these two things is like comparing bicycles and cyclists,” Madden told Fierce.
All of this begs the question: Who are the most likely targets for Nokia’s AI-RAN?
T-Mobile, Softbank, Elisa, Indosat and a few others are good candidates, according to Madden. “I don’t expect Verizon or AT&T to adopt this approach; Verizon’s CTO Yago Tenorio has made public comments that are very skeptical of the AI-on-RAN architecture and AT&T has already decided to move away from Nokia. Personnel changes at these companies are unlikely to change their direction, as infrastructure development happens in a long, slow process,” he said.
Nokia’s claims of 2x improvement in spectral efficiency is “very exciting,” he said, but it’s “completely unproven … Nokia has a long way to go to prove that the performance is real. If it is, then the trajectory that I have forecasted for GPUs will change dramatically.”