Losing AT&T is big blow to Nokia, but move makes joke of open RAN

AT&T's now-confirmed decision to phase Nokia out of its radio access network (RAN) and become even more heavily reliant on Ericsson is close to a disaster for the Finnish vendor. Coming just three years after it had similarly lost a major 5G contract at Verizon – to Samsung, in that case – it will leave Nokia RAN-less with two of North America's biggest operators. That first loss made sense. This one is a headscratcher, and likely to cost Nokia as much as $1 billion in 2025.

The financial estimate came in a research note issued by Simon Leopold, an equity analyst at Raymond James. AT&T uses Nokia across about a third of its existing RAN and has historically accounted for between 3% and 5% of Nokia's quarterly sales, he reckons. To be clear, $1 billion would equal precisely 4% of Nokia's entire revenues last year. Speculation about the AT&T move had already sent the company's share price down 6.5% in Helsinki on Monday.

Nokia's own estimate is that AT&T has accounted for between 5% and 8% of mobile net sales so far this year. That would mean between €367 million (US$397 million) and €588 million ($636 million). Previously announced cost cuts will partially mitigate the impact, but it will need another two years to achieve its financial target of a double-digit operating margin after AT&T's decision, it said in a detailed statement.

Its 2020 setback at Verizon was no big surprise to those who were tracking its performance. It had been let down by Intel on the delivery of 10-nanometer chips for inclusion in its first 5G products and then hastily switched to field programmable gate arrays (FPGAs), an expensive and power-hungry type of chip but one it could more speedily develop. Product competitiveness suffered, margins fell and Nokia lost RAN market share.

When the chips are down

But these FPGA-based products were phased out of the 5G portfolio in late 2022. Nokia has stood by Intel for some of its RAN needs but introduced Broadcom for radio-unit chips and Marvell Technology for the most computationally demanding stuff (Layer 1) on the baseband side. Since the first quarter of 2022, it has recorded bigger 5G market-share gains than any other vendor, according to Dell'Oro, capturing 29% of the market outside China by the third quarter, said Nokia in its statement. It has looked to be over its problems.

Nokia has also appeared far more proactive than Ericsson in open RAN, an apparent strategic priority for AT&T. The industry at large, including many of its open RAN new entrants, is certainly less skeptical about Nokia's commitment to this concept. In today's statement, Nokia drew attention to a recently secured open RAN deal with NTT Docomo in Japan. It also said that AT&T's decision "was driven by reasons specific to AT&T" and that the US operator "believes Nokia has highly competitive products and services" in RAN along with "an accomplished R&D capability."

What, then, is going on? Earl Lum, the EJL Wireless Research analyst who originally sniffed out AT&T's ditch-Nokia plan, believes AT&T may have taken issue with the fan units in Nokia's latest massive MIMO (advanced 5G) radios. But if these are prone to malfunction in AT&T's hotter markets, as Lum apparently thinks, then how has Nokia managed to land so much 5G business in scorching India, where its radios have been widely deployed this year?

Lum also reckons Nokia was slow to replace Intel-based products. Yet this would be an equally perplexing explanation for AT&T's move. If, by Intel-based products, Lum means the FPGAs, then at least some of these are thought to have come from Xilinx (now owned by AMD). More importantly, Ericsson today has a much closer relationship than Nokia does with Intel, having put the Santa Clara-based chipmaker at the very heart of its cloud RAN strategy. If AT&T's $14 billion, five-year plan is to build a virtual as well as an open RAN for 70% of its traffic, then Intel will be everywhere.

AT&T's press department declined to answer a question on this, and Igal Elbaz, its chief technology officer, seemed deliberately vague on his call with US reporters yesterday. The implication was there, however. AT&T not only said Ericsson will provide cloud software but threw Intel's name into the mix of other vendors announced. Dell, Intel's big server buddy, was also on the list. Meanwhile, Roger Entner, an analyst with Recon Analytics, seems to have made up his mind that AT&T has backed "lookaside," a technique that involves running most Layer 1 software on the general-purpose central processing units (CPUs) that Intel makes.

"Inline," the alternative Nokia prefers, means putting Layer 1 on the custom chips developed by Marvell and using these as so-called "accelerators." The Finnish vendor insists its approach is more power-efficient and that CPUs gobble energy. If AT&T disagrees, then it is moving in the opposite direction from many of the Internet players, which increasingly use custom silicon in place of CPUs for specific workloads. Ericsson itself has made seemingly contradictory statements, occasionally insisting that its custom silicon beats CPUs and elsewhere championing those CPUs.

Cloudy logic

Alternatively, AT&T might think cloud economics will offset the CPU costs and agree with Verizon that Nokia's solution is not really "cloud-native," as the expression goes. In a white paper co-authored last year, Verizon and Ericsson wrote that inline "runs counter to the cloud RAN principles of portability and flexibility, effectively tying the MNO to a particular solution."

Nevertheless, Intel's critics – Nokia among them – think it no different. Although widely used in computers and servers, and compatible with common programming languages and software tools, Intel's x86 architecture is not a standard and only one other big chipmaker (AMD) supports it. The CPUs Intel is developing for Ericsson also look increasingly "purpose-built," Nokia recently told reporters. Among other things, they use non-programmable accelerators for the most resource-hungry Layer 1 software.

Virtual and open RAN is at least partly supposed to be about fomenting competition in the supply chain. The irony is that AT&T will be even more dependent on a single vendor after the switch away from Nokia. It does identify Fujitsu alongside Dell and Intel in its list of suppliers, but the Japanese company's role is altogether unclear. Fujitsu would most likely contribute radios, and yet AT&T evidently plans to buy lots of these from Ericsson. Its dual-mode 5G radios will support C-band and 3.45GHz spectrum, said Elbaz.

There are plenty of other unanswered questions. Will AT&T put its RAN on the same Microsoft-provided cloud platform supporting its 5G core? Mention of Ericsson's cloud software suggests not, but that would potentially create silos and make cloud savings harder to realize. Ericsson's massive MIMO radios will all include equalizers, it has said, even though some would prefer to see this feature kept in the baseband. But all baseband units must include equalizers – and work with more basic radios – to be deemed open RAN-compatible. Won't this apparent duplication of features drive up costs for AT&T?

There is speculation that Ericsson turned Vito Corleone, making AT&T an offer it couldn't refuse – one Nokia would not match. As for open RAN, there has long been a view it would change little, that incumbent vendors would pledge openness and commercial realities would persuade big telcos to keep using them. It seems the case here. But that will be no consolation for Nokia. In Helsinki today, the mood is likely to be as dark as a Finnish sky in winter.