Vodafone's Málaga Lab and the Silicon Fork Behind AI-RAN
- November 21, 2026
- 9 mins
- Technology
- ai ran ericsson network architecture nokia nvidia open ran telecom vodafone
Vodafone’s Málaga Lab and the Silicon Fork Behind AI-RAN
Something quietly important happened in telecom this week, and it wasn’t a flashy product launch or a record-breaking speed test. It was the opening of a lab in Málaga, Spain, and a set of trial announcements that, taken together, tell us where the next phase of mobile network building is headed — and where the industry still can’t agree.
Let’s start with the simple version. For years, the big promise of Open RAN was that you could mix and match equipment from different vendors instead of buying everything from one company. Now the industry is trying to bolt artificial intelligence onto that same idea. The question everyone is wrestling with is deceptively basic: where does the AI actually live, and what kind of computer runs it?
This week gave us three real data points on that question. Vodafone opened a dedicated AI-RAN test hub in Málaga. Its UK joint venture, VodafoneThree, announced trials with Nokia and NVIDIA. And Orange’s network chief gave a candid interview about how his company is choosing between two very different visions of how this all works. Let’s walk through each, because the contrast between them is the real story.
What Vodafone Built in Málaga
Vodafone’s new Málaga Lab is being pitched as a “multi-vendor AI-RAN innovation and validation hub.” That’s a mouthful, so let’s translate. It’s a place where Vodafone can bring together radio equipment, computers, and AI software from different companies and test them side by side, under controlled conditions, to figure out what actually works.
Kyriakos Exadaktylos, who heads RAN evolution in Vodafone’s group technology team, framed it this way: the lab uses Vodafone’s “strong foundation in Open RAN to create a multi-vendor AI-RAN innovation and validation hub.” The goal, he said, is to “establish independent, repeatable benchmarks that show where AI delivers real value in mobile networks.” That word — repeatable — matters. Right now, every vendor claims its AI does wonderful things. Independent, repeatable testing is how you find out who’s telling the truth.
The lab is organized around three distinct approaches, and understanding them is the key to understanding this whole story:
- AI-for-RAN: using AI to make the radio network itself work better — things like smarter beamforming (aiming signal energy at users rather than spraying it everywhere) and predictive resource allocation (guessing where demand will spike before it does).
- AI-and-RAN: testing whether radio network functions and AI applications can share the same computing hardware securely and efficiently. This is the ambitious one, because it asks whether a single box can do two very different jobs at once.
- AI-on-RAN: hosting AI applications — computer vision, robotics, and so on — close to end users at the edge of the network, where low latency matters.
On the hardware side, the lab is equipped with QCT servers running NVIDIA’s Aerial RAN Computer MGX 2U Short platform, alongside Intel Xeon 6 Granite Rapids servers. In plain terms: Vodafone is testing GPU-accelerated systems next to more conventional x86 processors, and it’s also comparing against purpose-built telecoms chips known as ASICs. Vodafone says it plans to assess silicon performance using Standard Performance Evaluation Corporation benchmarks, AI performance through MLPerf, and RAN performance through the AI-RAN Alliance.
There’s also a forward-looking item on the agenda: integrated sensing and communications, or ISAC. That’s the idea that the same radio waves carrying your phone call could also “see” the environment — for example, detecting intruders at a sensitive site. It’s an intriguing capability, though clearly still in the experimental stage.
VodafoneThree Joins the Trial Parade
Meanwhile, in the UK, VodafoneThree — the merged entity combining Vodafone’s and Three’s UK operations — announced it will work with Nokia and NVIDIA on AI-RAN lab testing and field trials starting in the first half of 2027. Trial sites will be chosen once lab testing wraps up.
The initial use cases are refreshingly concrete: machine-learning-based channel estimation and link adaptation. These are techniques that help the network judge changing radio conditions and adjust how data is sent in real time. If that sounds unglamorous, that’s the point — this is the unsexy plumbing that determines whether your video call stutters.
The framing here is important. The programme centers on embedding AI directly into how the radio network operates, rather than treating AI as a separate layer sitting on top. And notably, VodafoneThree is running this on a parallel track to its work with Ericsson, testing whether GPU-based RAN can deliver. That’s the crux: Nokia and NVIDIA on one side, Ericsson on the other, and a real-world experiment to see which approach wins.
The Fork in the Road: GPUs Versus Custom Silicon
This is where Orange’s perspective becomes genuinely valuable, because Orange’s network executive is willing to say out loud what a lot of operators are thinking privately.
Orange started its evaluation process at the beginning of this year and is now, in its own words, developing “a good understanding of the different Open RAN configurations we can consider.” The next step is understanding how AI-RAN implementations can increase capacity and performance. As the Orange executive put it, “The offerings from some of the vendors are becoming much clearer now.”
Here’s the fork. Ericsson believes in tightly integrating RAN software with its own multi-core custom silicon — ASICs. Its argument is that custom chips deliver highly optimized performance and, because of their multi-core architecture, have plenty of compute headroom to run AI workloads. Nokia, by contrast, is moving away from custom silicon toward general-purpose and merchant solutions. Nokia’s bet is that it can ride the ongoing cost and performance improvements in general-purpose computing, and in doing so create an open platform capable of running the most advanced AI algorithms — many of which, as Nokia points out, haven’t been invented yet.
That last point is worth sitting with. If you build custom silicon for today’s AI algorithms, you’re locked into today’s understanding of what AI in a network should do. If you build on general-purpose hardware, you’re betting that flexibility is worth more than peak efficiency. Neither bet is obviously wrong. They’re just different bets about how fast things will change.
Orange is hedging. “One thing I’m quite sure about is that, in all cases, we need to work with both vendors,” its network chief said. “There are pros and cons to both. What I really like about both approaches is that the companies are opinionated. They can explain very clearly why they believe their way of thinking is the best.” That’s a diplomatic way of saying Orange isn’t ready to pick a winner — and doesn’t think it has to yet.
Nokia’s ambition extends beyond just connectivity. Harri Holma of Nokia confirmed the company wants to create a more open radio platform, allowing third parties to develop applications using information generated within the RAN — interference data, positioning data, sensing data. In other words, Nokia wants the radio network to become a platform that others build on, the way smartphones became platforms for app developers. It’s a big bet, as the Orange executive acknowledged.
The Memory Crisis Nobody’s Talking About
One detail from the Orange discussion deserves its own section, because it’s the kind of thing that quietly derails technology plans: the memory crisis.
Orange noted that all vendors are affected by it, and that this is a concern because the operator wants to make sure it doesn’t have too much of an impact on future costs. This is a reminder that AI-RAN isn’t happening in a vacuum. The same global shortages and price pressures affecting the broader computing industry affect telecom too. You can have the most elegant architecture in the world, but if you can’t source the components at a sensible price, the business case falls apart.
For an industry that loves to talk about algorithms and spectral efficiency, it’s worth remembering that supply chains and component costs often decide what actually gets deployed.
Why This Week Matters
Step back and look at the pattern. Vodafone is building a neutral testing ground in Spain. Its UK joint venture is running trials with Nokia and NVIDIA while keeping Ericsson in the mix. Orange is deliberately keeping both vendor camps close. Three different operators, three different approaches, but all pointing at the same underlying reality: nobody has settled the question of what AI-RAN should be built on.
That’s not a failure. It’s actually a sign of a maturing market. A few years ago, AI-RAN was mostly slideware — impressive diagrams with arrows pointing at a cloud labeled “AI.” Now operators are running benchmarks, planning field trials for 2027, and arguing about silicon. Those are the arguments you have when you’re actually trying to build something.
The Málaga Lab’s emphasis on “independent, repeatable benchmarks” is the tell. Vodafone isn’t just testing whether AI works; it’s testing whether the claimed benefits survive contact with a neutral measuring stick. And by explicitly working with Nokia, NVIDIA, Intel, and others under one roof, it’s trying to avoid the trap of letting any single vendor define what “good” looks like.
For anyone following Open RAN, the message is that the multi-vendor principle is now being applied to AI itself. The disaggregation that once meant separating radio units from baseband units is now being extended to the question of where intelligence lives — in custom silicon at the cell site, in GPUs at the edge, or in some combination. The operators who get this right will be the ones who can add AI capability without locking themselves into a single company’s roadmap for a decade.
The trials starting in 2027 will be the first real test. Until then, watch the benchmarks, watch the silicon choices, and watch whether the memory crisis turns out to be a footnote or a headline.
Sources
- Vodafone builds AI-RAN test hub to improve mobile connectivity — Vodafone
- VodafoneThree Taps Nokia, NVIDIA for UK AI-RAN Trials — Telco Magazine
- Orange picks its way through AI-RAN battle — The Mobile Network
- VodafoneThree doubles up on AI-RAN with Nokia–NVIDIA trial
- VodafoneThree to test Nokia’s AI-RAN with Nvidia GPUs