If Nokia is to survive as a mobile kit vendor deep into the 6G era, the western world’s big telcos need to embrace its idea that Nvidia’s chips are a desirable and economical feature of the radio access network (RAN). Co-developed with Marvell Technology, Nokia’s latest custom silicon product is the last it plans to sell. In a few years, there will be no alternative to Nvidia.
What many RAN observers regard as a massive gamble on those Nvidia graphics processing units (GPUs) probably explains why Nokia has been so keen to advertise telco interest in its forthcoming AI-RAN offers. After its latest Radio World get-together, held in the Finnish city of Oulu on Lapland’s doorstep, it reeled off the names of various operators giving AI-RAN a test drive. Some 19 have now signed up for “initial deployments,” claimed CEO Justin Hotard in a LinkedIn post. But the three companies he identified in that post are less well known – and none is a telco.
Pulsone (a brand used by Cohere Technologies), DeepSig and ZaiNar are among a cohort of startups designing RAN software they hope can be deployed on Nokia-built networks. To the consternation of many telcos, kit vendors remain the network’s gatekeepers, and a small number account for almost all commercial business. But Nokia has looked the most willing to unlock its doors and allow other developers inside.
A similar degree of openness was always baked into the AI-RAN pitch that came out of Nvidia. Besides positioning GPUs as a general-purpose substitute for custom silicon in the RAN, the world’s largest chipmaker has assembled its own framework for a RAN software stack. Branded Aerial, it naturally works on Nvidia’s CUDA software platform. Provided they are CUDA-literate, other developers are free to take it and tinker, adding their own special ingredients to the recipe.
Using Aerial as a foundation is precisely what Nokia has done. “It provides a base for our teams and Nvidia to go out and test some of the models,” said Pallavi Mahajan, Nokia’s chief technology officer, when she spoke with Light Reading in the summer. “I think having that as a base is super useful, which is why we don’t look at Nvidia as just a silicon provider.”
App attack
Yet even before Nokia teamed up with Nvidia around this time last year, it had seemed more receptive than certain rivals to some of the concepts aired by the open RAN movement, led by the O-RAN Alliance. Most famously, that group has standardized the fronthaul interface that joins radios to baseband units, where Nvidia’s technology figures. The effort means telcos can, in theory, combine one company’s radios with another’s baseband technology, instead of buying all products from the same vendor system. Meanwhile, a separate O-RAN initiative aims to ensure RAN technology can host third-party network apps.
Those apps could perform jobs like boosting capacity, improving spectral efficiency or powering down basestations at quiet times of the day. They come in several flavors, but only one is being wholeheartedly consumed. So-called rApps would sit on a platform called a non-real-time RAN intelligent controller (RIC) and be good for tasks that do not have to be executed at low-millisecond speeds. Ericsson, Nokia and Samsung, the three biggest RAN vendors without Chinese origins, all offer rApp support. But Ericsson and Samsung have been far more resistant to the xApp, the second flavor.
The key difference, supposedly, is that xApps work much faster to carry out necessary tasks. For that reason, they require a near-real-time RIC, and neither Ericsson nor Samsung has sounded supportive. Judging by their earlier public statements, both vendors seem to believe xApp-type capabilities are best delivered by their own baseband technology.
That stance originally distinguished them from Nokia. In 2023, it flagged trials of an open RAN interface called E2 that would allow a near-real-time RIC to be linked to Nokia’s baseband units. Subsequently, however, there was no observable progress, as the company acknowledged in late 2025, during a press event in Oulu. “xApps has been of less interest lately,” said Tommi Uitto, the former president of Nokia’s mobile networks business group.
James Crawshaw, a principal analyst with Light Reading sister company Omdia, had a similarly gloomy take on xApps around the same time. “Ericsson doesn’t advocate xApps and is only pushing the rApp story,” he said. “Moving to xApps, where you’re changing things at very, very fast control loops – that’s a future project.” The x, whatever it was supposed to represent, now has the look of a cross that has toppled off a crypt.
Meet dApps and the real-time RIC
But a new strain of network app has recently emerged. The more dapper-looking dApp has been presented by parts of the industry as an option for tasks that demand even speedier execution and a real-time RIC. It is today seen as an optimal route into the network by Cohere, one of the startups that Hotard named in his LinkedIn post. Having previously designed an xApp for its capacity-boosting universal spectrum multiplier (USM), Cohere has now come up with a dApp for the same product, as described in a blog published in July.
Where xApps needed the E2 interface, dApps would use a newer one called E3 that Nokia has swung behind. “We’ve taken the whole concept of APIs and decoupling layers in a very serious fashion, and what it now enables us to do is bring in an O-RAN-compliant E3 interface,” said Mahajan. That means Nokia will be able to support apps for sensing, more “geo-precise” location monitoring and improved spectral efficiency, she explained.
Partnership with Cohere could make Nokia’s target of doubling spectral efficiency by 2028 on much of the existing infrastructure look more achievable. Telcos including Vodafone, Australia’s Telstra and Bell Canada have publicly endorsed its USM technology after field trials, and all three have now made investments in the startup. On a recent Telecoms.com podcast, Cohere CEO Ray Dolan also identified Nokia as the vendor partner during trials in the Bell Canada network.
Those appear to have used a common, off-the-shelf server that can be deployed next to a basestation at a mast site to host Cohere’s technology. But a dApp would be closer to a direct integration and probably come with service benefits. It would achieve what xApps did not, said Dolan on the Telecoms.com podcast.
As Cohere explained in its blog, E2 operates primarily within the control plane, the network’s traffic cop, as opposed to the user plane, the actual traffic. “The E3 interface extends this model into the user-plane domain, allowing dApps to influence scheduling, precoding, and channel estimation with sub-millisecond latency.”
As ‘complicated as brain surgery’
None of this, however, means dApps lie just around the corner. “rApps were easy, xApps were hard, dApps are really hard,” acknowledged Dolan very candidly. “It’s a great idea that is nothing but a great idea.” The engineering of dApps is as “complicated as brain surgery” and the industry has a huge amount of work left to do, he added.
What’s more, the other big kit vendors have been relatively mute about dApps. Given their stated qualms about xApps, both Ericsson and Samsung would have a hard time explaining why dApps are more acceptable. “We view dApps as an emerging concept, not a replacement for rApps,” said Ericsson by email. “rApps remain our established approach for automation use cases within our architecture.”
“For xApps and dApps, we continue to monitor and research, but there is no product commitment to support dApps today,” it continued. “Our position stays aligned with our target architecture and industry requirements and needs.”
Ignoring the Chinese, Ericsson remains the world’s biggest vendor of current 5G technology and is naturally more risk averse. That contrasts with the make-or-break strategy of Nokia under its current boss, following market share losses in the lucrative US before he joined the business.
Their radically different approaches might also raise questions about the portability of apps between vendors. Nokia is evidently trying to build some kind of ecosystem around Aerial. But the CUDA software platform is native to Nvidia’s GPUs, meaning apps based on it cannot be migrated to other types of hardware. Today, neither Ericsson nor Samsung is actively pursuing a GPU RAN strategy.
Were Nokia to thrive with a dApps-led approach – and even reverse the market share losses of recent years – its competitors might be forced to respond. But many telcos see little need to replace the hardware they currently use. That seems less of an issue for Ericsson and Samsung than it is for Nokia, whose software development track requires customers to invest in GPUs.
Unfortunately, no rApp, xApp or dApp will address the problem that open RAN was initially intended to fix – the shortage of economically viable suppliers building the main RAN parts. The opening of the fronthaul interface was supposed to foment competition. Yet operators are still heavily dependent on a handful of big vendors.
Spending in that market fell from $45 billion in 2022 to $35 billion last year, according to Omdia’s research. But it remains significantly higher than expenditure on apps, with Omdia predicting that rApps will generate annual revenues of only several hundred million dollars by the end of the decade. Apps could feasibly make the infrastructure options far more valuable. But they won’t broaden the choice.
