Traveling to this year's Mobile World Congress in Barcelona, Dennis Hoffman could not help but notice the rise of AI and decline of open RAN. "All the big signs and banners and adverts around open RAN had been replaced by AI," said the general manager of Dell's telecom systems business, on a call this week with Light Reading. "That seems like the pivot."

It seems like the pivot by Dell, too. The release about its latest telecom move is heavy on references to Nvidia, an AI chips powerhouse, and includes not a single mention of open RAN (for radio access network). Instead, the big maker of server equipment wants to talk about the multi-faceted "AI for telecom" program it has just launched. It's a response to the frenzy of interest in AI as telcos grapple with the concept and try to figure out where the technology should fit (if anywhere) in their networks. Dell thinks it can help.

"I think it's our position in the value chain," said Hoffman, when asked why. "We're above the silicon layer, so we're needed to package and deliver the silicon innovation to the end user. We're also the place where the virtualization software and bare metal application runs." Architecturally, of course, Dell is found just about everywhere, from the giant data center to the network "edge" and user device. It gives Dell a bird's eye view of the whole system.

But the industry's shifting priorities also reflect the contrasting fortunes of the two big chip companies that sponsor open RAN and AI. Intel, the dominant provider of central processing units (CPUs) for use in virtual network infrastructure, looks bloodied in the open RAN corner. Across the canvas is a grinning Nvidia, a silicon titan of AI.

Intel's share price has halved this year, and its last financial report was a troubling read for investors, showing a net loss of $1.6 billion for the second quarter, compared with a $1.5 billion net profit a year before, with sales relatively stable at about $12.8 billion. Nvidia's stock is up 144%. Its second-quarter profit soared 122% year-over-year, to $16.6 billion, and its revenues grew 122%, to $30 billion.

The relative health of each company, of course, has little to do with the telecom sector. Yet the telecom sector has reacted. Industry executives have recently talked less about open RAN than they have about AI-RAN. On stage at T-Mobile's capital markets day last week, CEO Mike Sievert wondered aloud why open RAN architecture still proved so elusive. Moore's Law is dead, responded Nvidia boss Jensen Huang, in a veiled dig at Intel. CPUs can't keep up with more customized chips, he said, promoting Nvidia's graphics processing units (GPUs) as a kind of general-purpose successor.

DisArming Dell

As one of Intel's very biggest partners, Dell obviously has a far more charitable assessment. Sapphire Rapids, an Intel platform intended partly for a virtual RAN, needs fewer servers and less "off-processor acceleration" than the older Ice Lake did to handle the same requirements, said Hoffman. He expects Granite Rapids-D, a forthcoming Intel platform, to be even better.

But he agrees that Nvidia would bring a "silicon diversity" that is currently missing from the sector. "It's all beneficial. The issue, however, is there has to be a software ecosystem. Everything is written to run on x86," he said, referring to the architecture at the heart of Intel's silicon products. AMD remains the only other big manufacturer of x86 chips.

Arm, the main alternative to x86, is used by Nvidia to make Grace-branded CPUs that could theoretically host the RAN software developed by Ericsson, Nokia, Samsung and others. Oracle-backed Ampere Computing and AWS, through its Graviton processors, are also Arm licensees. Yet Hoffman's remarks worryingly imply there is still an absence of viable Arm-based alternatives to Intel. Despite much industry talk about cultivating them, the big RAN kit vendors have not coded for Arm-based silicon, according to the Dell executive. "If they do, you can bet we'll be making subsystems to serve it," he said.

Not just for chatbots

In the meantime, and as part of its "AI for telecom" strategy, Dell is now promoting an Intel-based PowerEdge server that includes an Nvidia GPU. "This is not RAN-dependent," said Hoffman, shifting the conversation away from the topic of open virtual RAN. Instead, the idea is to introduce Nvidia's powerful chips into parts of the telecom network where they could have various uses, from improving network performance to supporting new applications for customers.

Dell, moreover, can share several examples of work it has already done with telcos in this space. In partnership with South Korea's SK Telecom, it has built a "chat agent" that combines AI with existing business support systems. Perhaps more interesting is the use of generative AI – the flavor of the technology most associated with Nvidia's GPUs – for the very telecom-specific purpose of root cause analysis.

"You have lots of message flows, log files, structured and unstructured data," said Manish Singh, the chief technology officer of Dell's telecom systems business. "One of the use cases there is to use generative AI, which is very good at doing anomaly detection. The end result of that is expediting network troubleshooting. Think of it as a network engineer copilot."

His other network-specific example is fault prediction, usually handled by older machine-learning tools. The "big unlock" with generative AI, he explains, is the technology's ability to look at not just structured but also unstructured data. "That's where generative AI really shines," said Singh. "We are already seeing 90% plus accuracy of fault prediction." In the RAN, specifically, generative AI can also help to analyze system data and derive insights.

This all chimes with the messaging recently put out by Nvidia. The challenge is making it pay. GPUs, everyone knows, are expensive and power hungry. To justify investment in them at the edge, an operator would probably have to realize multiple benefits, using GPUs to host network software and AI applications. And those would probably need to include revenue-generating and not just cost-saving applications. Skepticism is rife.

"A big driver of Dell's telecommunications business, and the decision we made to form it back in 2020, was that the world's network operators own an awful lot of the edge," said Hoffman when challenged on this point. "And so, if we're going to be successful in edge computing, we need to be successful in helping enable telecom to capitalize on this opportunity."

Nvidia, meanwhile, threatens to be a mightily disruptive force. Last week's update revealed ARC-1, an appliance or server – apparently deployable in data centers or at mobile sites – that includes all the hardware and software needed for the baseband part of the RAN. An Arm-based Grace CPU would handle some RAN functions, leaving more demanding software to a Blackwell-branded GPU. If the technology works, and the economics add up, tomorrow's RAN ecosystem could look very different.