Nokia spruces up its AI-RAN pitch with spectral efficiency claims

• Nokia and its partner/investor Nvidia are eager to get GPU-based AI-RAN technology into mobile networks
• Nokia has unveiled what it claims is the ‘industry’s first AI-native RAN platform’
• The pitch has some attractive deployment options for operators and some bold claims related to significantly improved spectral efficiency gains
• But questions remain about the economics of such deployments, including the impact on power consumption

As planned, Nokia has integrated its anyRAN software with Nvidia’s Aerial AI-RAN platform to create what the Finnish vendor is calling the “industry’s first AI-native RAN platform” and, in doing so, it is offering what appear to be mobile operator-friendly deployment options as well as making some very bold claims about projected spectral efficiency gains. 

There’s no doubt that Nokia and its partner (and investor) Nvidia have put a lot of thought into the ways in which mobile operators might want to deploy radio access network (RAN) equipment that is built around a GPU. They have been working hard on the AI-RAN pitch since Nvidia pumped $1bn into Nokia in October last year and at MWC26 in early March, Nokia announced that it had completed “successful functional tests of GPU-accelerated AI-RAN” with friendly network operators, such as Indosat Ooredoo Hutchison (IOH), T-Mobile US and SoftBank Corp, all of which are AI-RAN advocates and Nvidia partners.  

Now Nokia says its AI-RAN platform will enter pilot deployments at the end of this year and will be commercially available in 2027, “with a roadmap that leverages Nvidia’s programmable merchant silicon platforms”. 

And in pitching the new RAN deployment option to mobile operators, they have highlighted a very attractive feature that will certainly draw attention – the claimed ability to carry significantly more data traffic across existing licensed spectrum over the next few years. 

According to Nokia, the AI-RAN platform enables mobile operators to “unlock significantly more uplink and downlink capacity from the spectrum and radio infrastructure they already own”. Specifically, Nokia claims the AI-RAN platform “has already shown more than 20% spectral efficiency gains through AI-driven radio innovations,” though these innovations are not specified. “The company is on track to deliver 50% spectral gains by 2027 and more than 100% by 2028” – effectively doubling the capacity of existing spectrum assets – “helping telecommunication providers carry significantly more traffic in dense cells while reducing cost per bit and improving customer experience,” the vendor added. 

TelecomTV understands that those spectral efficiency figures are based on the vendor’s internal testing, validation and roadmap projections, with the results based on controlled conditions: Ultimately, real-world performance will have to be validated with operators during trials.

Still, that kind of proposition will undoubtedly attract the attention of all mobile operators. 

But it’s not a pitch they haven’t heard before: Specialist vendor Cohere Technologies has for years been touting its universal spectrum multiplier (USM), which can boost spectral efficiency on traditional and Open RAN deployments with a software upgrade. Cohere Technologies has been involved in multiple network operator trials, including with Vodafone and Bell Canada (which is deploying Nokia’s cloud RAN platform) – see this article from February 2025.  

But although Cohere Technologies and Nokia have engaged at various times in recent years, the spectral efficiency gains being claimed by Nokia are all its own work: The vendor told TelecomTV that the performance gains come from Nokia’s AI/ML-based RAN software and algorithms running on Nvidia’s accelerated compute hardware, and that the spectral efficiency gains are primarily driven by Nokia software combined with Nvidia GPUs.

More than one way to deploy AI-RAN

What will also attract mobile operators is that Nokia is offering three different ways to deploy its AI-RAN hardware, none of which involve rip-and-replace procedures. You can read more about the options in Nokia’s press release, but essentially they are:

For existing Nokia AirScale customers, a new GPU-powered AirScale capacity plug-in unit that integrates Nvidia’s Aerial AI-RAN platform into existing network infrastructure, “enabling a significant capacity step-change through a simple upgrade path while preserving existing network investments”. Nokia added that mobile operators can “introduce advanced AI capabilities, continuously improve performance through software and extend the value of deployed infrastructure,” with this option. 

A GPU-powered standalone AI-RAN node that “brings AI-accelerated RAN performance to any network environment and supports 4G, 5G and future 6G workloads on a common platform.” This node can be deployed as a standalone option, in a cluster of nodes or integrated with existing AirScale deployments to act as “a single logical base station”.  

GPU-powered AI-RAN COTS (commercial off-the-shelf ) server solutions delivered through ecosystem partners. “The platforms enable an open and secure supply chain while supporting deployment on industry-standard accelerated computing infrastructure, combining cloud-native flexibility with the performance requirements of AI-native radio networks.” 

According to Nokia, “these hardware platform options allow telecommunication providers to modernise at their own pace while preserving existing infrastructure investments, benefitting from a common software roadmap and accelerating innovation at software speed… [operators] can adopt AI-RAN in stages using the approach that best matches their deployment strategy, capacity requirements and installed base.”

Nokia noted that it has ongoing AI-RAN engagements with Deutsche Telekom, Telia, Orange and BT, as well as T-Mobile US, SoftBank Corp and IOH, among others, but that “not all deployment options have been publicly confirmed as having been tested by all operators.

Sticking points

On the face of it, these are all positive advances for Nokia’s AI-RAN pitch to mobile operators, but over the past couple of years – since the AI-RAN Alliance founders, including Nokia and Nvidia, first put forward the concept of GPU-based RAN systems – many mobile operators have been sceptical about the costs and energy consumption implications of such deployments: GPUs are expensive and are power-hungy. Mobile operators, meanwhile, are cutting their network investments and looking for any way possible to improve power efficiency. 

Nvidia has always countered such claims – see this article from March 2025 – and Nokia’s CTO Pallavi Mahajan told TelecomTV in March that the power and cooling issues were being addressed by Nvidia, which was working on optimising its Blackwell GPUs for telco workloads so that it “runs within the same guardrails of power” and  “within the same card rails of thermal”.

Now, Nokia is saying that the AI-RAN systems have been designed to “operate within broadly similar power envelopes to existing systems while delivering higher capacity,” offering “an improved energy efficiency per bit due to significantly higher throughput.”

As for the cost of such deployments, Mahajan talked about total cost of ownership (TCO) rather than pricing, saying that the AI-RAN platform being developed would provide operators with a performance and spectral efficiency “upshot” that couldn’t be matched by custom silicon and that offered an improved total cost of ownership (TCO). 

Today, Nokia told TelecomTV that the AI-RAN platforms are being offered as software subscriptions “that scale with operator needs” and that any associated hardware costs are set to be “broadly in line with existing AirScale and COTS solutions, with greater capacity and performance,” though “final product pricing is not yet disclosed.” Nokia also noted that for the cloud RAN deployments, “operators can use GPUs, such as NVIDIA RTX 4500 or RTX 6000, with flexibility depending on requirements”.

The vendor added: “The business case is driven by higher capacity from existing spectrum, lower cost per bit, and the ability to defer investment in spectrum and site upgrades,” which sticks with the TCO line. 

- Ray Le Maistre, Editorial Director, TelecomTV