- SoftBank and Nvidia are the two big players behind the AI-RAN Alliance
- Intel isn't a member of the alliance, which launched two years ago
- Intel says existing standards bodies and its Xeon CPUs already support AI in the RAN – and it’s not in a hurry to join yet another group
While much of the telecom industry is rallying around the AI-RAN Alliance, Intel is notably absent, preferring to wait this one out – for now.
In some ways, it’s not surprising. When the AI-RAN Alliance launched in 2024, a lot of people were taken off guard. SoftBank and Nvidia were driving the bus, and other founding members Amazon Web Services (AWS), Arm, DeepSig, Ericsson, Microsoft, Nokia, Northeastern University, Samsung Research and T-Mobile jumped on board.
Today, the alliance consists of 132 members – and Intel is still missing. But the chip maker says its decision isn’t in opposition to AI in the RAN so much as it is about practicalities.
“I am not against AI-RAN Alliance at all,” Cristina Rodriguez, VP and general manager of Intel’s Network & Edge Group, told Fierce at Mobile World Congress 2026 in Barcelona. “It hasn't been something that we have decided to join yet. We definitely see the importance of AI and the possibility of deploying AI today in the live network with the capability that we have today.”
In a formal statement, Intel said established standards bodies like 3GPP, the O-RAN Alliance, Telecom Infrastructure Project (TIP) and 6GIC already provide strong governance for network AI, and broad standards-focused groups help prevent fragmentation.
“Intel remains active across these forums and continually evaluates how to best support industry innovation and partner needs. Regarding the AI Alliance, we are assessing how its efforts align with existing work and where our involvement makes sense for the industry,” the company told Fierce.
Intel’s ‘practical’ AI strategy
Intel argues that AI can already deliver tangible benefits in the RAN using existing architecture, particularly server-based deployments powered by its latest processors.
The company’s Xeon 6 system-on-chip (SoC), which launched last year, includes built-in AI acceleration through advanced matrix extensions, Rodriguez said. The latest generation increases the core count to as many as 72 cores, up from 42 in an earlier release.
By having 72 cores, “we can reduce the number of servers so we go from more than one server per site to the possibility to have one server per site,” she said. “That's major from the point of view of total cost of energy, complexity and power consumption.”
The great GPU debate
Intel’s stance highlights an important debate within telecom: whether RAN infrastructure will require GPUs, a concept that Nvidia certainly is bullish on. (Remember, there’s a reason people referred to MWC 2026 as “Nvidia World Congress.” The GPU titan was everywhere.)
But Intel is careful not to frame this as a GPU vs. CPU debate. “GPUs obviously play a very important role in AI, no question about that,” Rodriguez said.
However, the use cases have yet to emerge that justify putting large language models in the RAN. “I don’t believe the industry has seen the use cases that justify having additional components, such as a GPU, on the RAN,” especially at the cell tower, she said.
“You don’t need to throw more resources than what you need,” she said. “If you can do your workload with the architecture you have in place, then you don’t need to have additional components.”
AI RAN: What’s left to learn
AI is being used today to optimize and improve the RAN, often leveraging traditional machine learning algorithms, and it’s yielded value for many years now, said Roy Chua, founder of research firm AvidThink.
“I think we all recognize that the use of AI in the RAN has been going on for some time,” he said. “There’s no doubt about that.”
What’s not so well known is running non-RAN workloads on the RAN infrastructure, whether through sharing GPU and other computing resources or having them co-located and running side-by-side.
“The need for the far edge versus the near edge for workload placement is unclear – we've grappled with this before many years ago during the first wave of edge computing for the telcos (MEC and others),” he said.
Likewise, using GPUs as a computing platform to replace existing systems-on-chips that combine very specialized ASICs with multi-core high-efficiency CPUs – that's been demonstrated in proof of concepts (POCs) by many vendors but it’s still a work in progress, he said.
GPU bulls and bears
Ericsson, Nokia, Huawei and Samsung all have varying opinions on this, as does Qualcomm, Marvell, Intel, Nvidia, Arm, he noted.
“Whether they are bullish or bearish on GPUs in the base stations will depend on finding monetizable non-RAN workloads for (B2B or B2C apps) and the sophistication of orchestrating and managing such workloads along with running a reliable RAN and the cost/energy efficiency of using large number of GPUs to replace today's optimized ASICs,” Chua said.
For its part, Intel is happy to participate in AI RAN via multiple industry groups.
“AI is going to be everywhere,” Rodriguez said. “This is an old industry and we have multiple forums. We have the Telecom Infrastructure Project. We have 3GPP and we’re part of all of those. I think there are so many places in the industry to talk about it and to figure out what is the best solution and how do we continue advancing AI.”
Read all of our coverage from Mobile World Congress 2026 in Barcelona here.