Nobody wins from open RAN except the public clouds

Descending the auditorium steps at the FYUZ event in Madrid this week, attendees were enthusiastically told that a surprise guest would appear for the inaugural session. "Tom Cruise?" one former film studies student hopefully enquired. What no one expected to see was the giant screened face of Fredrik Jejdling, the head of Ericsson's mobile networks unit, hovering above the massed ranks of telecom executives like an alien spaceship. Ericsson is not even a member of the Telecom Infra Project (TIP), the organizer of the show, and it has long been seen as an enemy of open RAN, the topic that consumes TIP's attention.

Jejdling seemed keener on talking about Vonage. Ericsson bought the software business for $6.2 billion last year and yesterday wrote down the value of its investment by $3 billion, blaming the dire performance of Vonage peers on stock markets, among other things. But he was able to muster some faint Swedish enthusiasm – the kind you saw when Bjorn Borg won a point at Wimbledon – for open RAN after his company recently promised to make radios compatible with a certain flavor of the technology.

About 1 million are already in the field, ready for a software upgrade, he told the audience. Operators, then, should expect to see that software by late 2024, posited Santiago Tenorio, the network architecture director of Vodafone and TIP chairman. "I don't want to commit to any particular time," replied Jejdling with a nervous chuckle.

There is predictable skepticism about Ericsson's intentions and commitment. Open RAN's new interfaces could hurt the Swedish vendor by allowing operators to buy only some of its radio access network parts and link those to other suppliers' components. When they shop with Ericsson today, it is all or nothing. But Jejdling's close encounter with open RAN-kind means both Ericsson and Nokia – the other big Nordic vendor – have beamed down and are voicing support for the concept. And each has probably realized it is under little direct threat in this strange new land.

Investment gaps and duplication

RAN challengers are almost nowhere to be seen. Parallel Wireless, a RAN software developer, is not even a reference anymore, Tenorio told Light Reading last week, after it reportedly cut most of its workforce last year. Chinese vendors have been evicted from the most important markets for Ericsson and Nokia, if they were ever there in the first place.

Others have weakened their research-and-development (R&D) focus by expanding into new areas. Ericsson's RAN market share outside China has risen by six percentage points since 2017, it boasts. Nokia's 5G share is up by the same amount in just 18 months, it claims, citing Dell'Oro research.

The scale of the challenge was neatly summed up by Robert Soni, the RAN vice president of AT&T, who estimates that Ericsson and Nokia spend about $5 billion on R&D in this market each year. "Those numbers that are spent by the incumbents are staggering," he said. And about one in three dollars goes just on integrating their own products, Soni estimates. Others will succeed only if they pool resources, in his opinion.

But the industry is awash with overlapping initiatives – government-backed test centers, TIP's facilities, operators' individual efforts. Meanwhile, vendors are foisting their own pre-integrated partner solutions on telcos. "The whole idea was to give the operator the ability to make the choice," said Soni.

The death of general purpose

The virtualization aspect of open RAN is not what it was cracked up to be, either. In recent days, Nokia has attacked the notion that many central processing units (CPUs), supplied largely by Intel, are "general purpose." Instead, it believes they feature a lot of customization to support 5G needs.

Nokia's own cloud RAN products include the same customized silicon from Marvell used in its purpose-built 5G. Hyperscalers now routinely use customized silicon to offload demanding workloads from CPUs, said Joel Brand, Marvell's senior director of product marketing. Why should something with all the specific needs of 5G buck the trend?

Equally suspect is the description of servers including these chips as "common, off-the-shelf" equipment. A server sold to Walmart is not the same as one designed for telcos and compliant with very specific network equipment-building system (NEBS) guidelines, according to Geetha Ram, the head of RAN compute for HPE.

Brand thinks the industry must go even further. "The RAN today is based on custom appliances," he said during a FYUZ panel session. "Those custom appliances are built to operate outdoors, without active cooling, with passive cooling. They are designed for maximum reliability and availability. If we want to succeed to move that to a server environment, we really need to think what a vRAN server looks like." Yet all this would rather undermine some of the original arguments about open RAN economies of scale.

Cloudy economics

None of this means absolutely nothing changes and nobody wins. But open RAN's likely beneficiaries are probably not what the telecom industry intended. Clues lie in the increasing reference to "cloud" rather than virtual RAN and in recent tie-ups between the Nordic vendors and the hyperscalers. Of particular interest is a new deal between Ericsson and Google Cloud.

Under that arrangement, Google would bring its own computing equipment, described as "black boxes" by Ericsson, into a telco's facilities. The public-cloud giant would essentially replace the telco as the investor in that equipment. Ignoring the radios and antennas, the role of Ericsson, previously involved in hardware development, would be limited to contributing the RAN software. Rather than buying appliances and servers, the operator would pay Google on a usage basis.

Dish, a greenfield open RAN operator in the US, already has this kind of relationship with AWS. Decisions about CPU offload to accelerators and other technical features could eventually be taken away from the telcos. AWS, for instance, is working to put RAN workloads on its own Graviton processors, built with the blueprints of Arm.

"We defined an open fronthaul interface so now the processing can be done by a hyperscaler," says Brand. Whether the hardware is Dell, HPE or Graviton, hyperscalers can put their cloud platforms at the site hosting the distributed unit, responsible for baseband processing. "This is not increasing the variety of vendors, because there are only a few of them. Whether it's good for the industry, bad for the industry, will affect Nokia, will affect Ericsson, will affect Verizon, will affect Vodafone, I don't know."

This is not a fait accompli, of course. Hardly any telco workloads run on hyperscaler platforms today. A crop of alternatives – Red Hat, VMware and Wind River – exists for building a single private cloud that can manage all workloads. But the paradigm has changed now the hyperscalers have entered the private cloud space.

No one seriously thinks Red Hat can move in the opposite direction and establish itself as a public cloud. Nor can it develop chips, build large language models from scratch and do all the other stuff that might attract a telco to a hyperscaler. Nokia insists Red Hat is now its preferred cloud partner, but its anyRAN strategy is also about ensuring its software can work "on any partner's cloud infrastructure." Earlier this year, it showed that its RAN software, split between Marvell and Graviton chips, could run on an AWS platform.

Cloud RAN of this nature would see the public clouds advance even further into the telco sector. It would be especially awkward for supporters of "fair share," the controversial argument that Big Tech should contribute to network costs because demand for their applications is forcing telcos to invest bigger sums. Adding capacity in mobile often means putting in extra compute resources, but these would be rented from Big Tech, not owned by the telcos. Pricing negotiations between landlords and tenants don't often work out well for the latter.