After losing Nokia, crisis-hit Intel seeks network assets buyerAfter losing Nokia, crisis-hit Intel seeks network assets buyer

As Nokia's boss, Rajeev Suri presided over one of the most disastrous bets the telecom industry has ever seen, when the Finnish kit vendor staked all its chips on a losing hand. Under Suri, succeeded by Pekka Lundmark in 2020, Nokia had based its first wave of 5G radio access network (RAN) products on Intel's silicon. Development of the required 10-nanometer chips subsequently hit delays. In the telecom sector, Nokia was the standout victim.

Since then, it has hauled its way back to 5G competitiveness, phasing out the expensive field programmable gate arrays (FPGAs) that provided a stopgap after the Intel failure. But relations between the two companies were irreparably damaged. Nokia's immediate response was a pivot from Intel to Marvell Technology – by now well documented – for the baseband system-on-chip (SoC) technology dedicated to Layer 1, the most computationally demanding and hardware-dependent chunk of RAN software.

Suri alluded to that shakeup in February 2020, months before Nokia identified Broadcom, a developer of radiofrequency (RF) chips, as a third supplier of 5G components. "What has changed is that we no longer work with only one supplier but with two other SoC partners making custom silicon for us in mobile networks and they all have some unique assets, be it in RF or baseband," he said at the time.

While Intel looked to have permanently lost some business, the company retained its position as the supplier of chips hosting other RAN software. But in Nokia's latest range of 5G products, named after various birds of prey, Intel has been clawed out, according to multiple sources. Instead, Nokia is using chips based partly on the blueprints of Arm, a UK designer offering an architectural alternative to Intel's x86 platform.

Putting silicon up for sale

In a 5G market that has been shrinking, the loss of Nokia as a RAN customer represents a major setback for Intel, which had not responded to a request for comment at the time of publication. Its broader business is in crisis. Due to publish third-quarter results later this week, the chipmaker is expected to report its biggest quarterly revenue drop in five quarters, implying sales will fall by more than 8% compared with the year-earlier period.

Intel has lost market share to AMD, its x86-wielding competitor, and is threatened in various sectors by Arm-based rivals. So far, the company has also had no answer to the dramatic rise of Nvidia in the fast-growing market for artificial intelligence (AI) chips. Efforts to build a successful foundry business are meanwhile gobbling resources. Intel slumped to a $1.6 billion net loss for the most recent second quarter, after managing a $1.5 billion profit the year before.

In response, Intel is cutting about 15% of jobs, a figure that would translate to more than 18,000 employees, based on numbers for the most recent fiscal year. Almost every week brings fresh reports and speculation about a sale of assets to other silicon parties. Qualcomm, famous as a provider of modems and processors used in mobile phones, has been linked with an unlikely takeover of the whole business. Apple was this week floated as the latest prospective buyer.

Such a mammoth acquisition still seems improbable for all sorts of reasons, but Intel is known to have approached others about selling non-core parts of its business, including some of its network assets. Candidates have included Marvell, which is believed to have turned down an offer. Last month, the rival chipmaker was reported to be a "likely bidder" for Altera, the FPGA-making subsidiary of Intel.

But a sale of the more purpose-built chips used in 5G basestations is also possible, and Ericsson, Nokia's main European rival, is seen as a potential acquirer. Just as Nokia has retreated from Intel's RAN silicon, Ericsson looks increasingly entangled with it. That would give the Swedish vendor a strategic interest in buying assets to safeguard supplies. And Nokia's switch to other silicon suppliers inadvertently clears a major obstacle to an Ericsson takeover.

Ericsson entanglement

The structure of the RAN market makes Intel heavily reliant on Ericsson as a customer for its purpose-built silicon. Despite the efforts of some telcos to cultivate alternative kit makers, today's RAN market is dominated by a handful of big vendors. Data from Omdia, a Light Reading sister company, gave Huawei, Ericsson and Nokia, the three largest suppliers, a whopping 75.1% of the entire global market in 2023, an increase of 0.2 percentage points on the 2022 figure. ZTE and Samsung, the fourth- and fifth-biggest players, respectively, claimed another 20%.

US government sanctions would today prevent Intel from catering to Huawei's 5G needs. But even before those sanctions came into effect, Huawei was using basestation chips made by HiSilicon, its wholly owned subsidiary. Subtracting Huawei and Nokia from the equation would leave Intel with an addressable market that accounts for less than 50% of total RAN product sales. What's more, while Intel identified ZTE as a basestation chip customer back in 2020, any contracts with Chinese companies look susceptible to a further tightening of sanctions in the current geopolitical climate. And Samsung, the number-five player, relies heavily on Marvell, not Intel, for silicon in its purpose-built 5G.

Clearly, Intel's 5G business has massively undershot its original expectations. In October 2020, the chipmaker was eyeing a 5G silicon market it expected to be worth $25 billion by 2023. Yet the entire market for RAN products generated revenues of just $40 billion last year, when sales were down 11%, according to Omdia. This year, they are expected to fall another 7% to 9%. Meanwhile, total sales at Intel's network and edge group, which covers much more than just 5G, tumbled 31% last year, to $5.8 billion.

The division also registered a $500 million operating loss, compared with a $1 billion profit in 2022, and the telco part of it has been a drag on results so far this year. Excluding that business, network and edge sales for the first half would have risen a tenth year-over-year, said David Zinsner, Intel's chief financial officer. As it happened, revenues fell 7%, to about $2.7 billion. One source, who spoke on condition of anonymity, reckons the wireless division is bleeding about $300 million a quarter.

Ericsson, which said it does not comment on rumors and speculation, has always been relatively coy about the nature of its relationship with Intel. But it may be exposed in several areas. In Layer 1 (L1) – where Nokia relies on Marvell, which in turn uses Taiwan's TSMC as a foundry – Ericsson has long claimed that all design work happens in-house. For planned products, this leaves Intel with a manufacturing-only role. "Ericsson designs the Ericsson Silicon SoCs for its RAN compute products (including the L1 processing) and Ericsson uses Intel as the manufacturer of those SoCs," the Swedish company previously told Light Reading by email.

For other parts of the RAN software stack, however, Ericsson has acknowledged its heavier reliance on Intel products originally codenamed Snow Ridge and subsequently released under the Atom brand. Snow Ridge, emphasized Ericsson, was designed and built by Intel, albeit with the Swedish vendor's input. "In the Ericsson RAN compute product it provides the processing for the L3 control plane, user plane, the OAM [operation, administration and management] handling and the IP [Internet Protocol] interface," said Ericsson in August last year.

Virtually non-existent

Seemingly less problematic, although a major industry talking point, is Ericsson's dependency on Intel in the tiny market for cloud or virtual RAN. Here, Ericsson has developed a separate software stack it can run largely on Intel's general-purpose Xeon-branded central processing units (CPUs). For an especially demanding L1 task known as forward error correction (FEC), Ericsson also relies on an Intel "accelerator," a more customized chip. Intel provides the FEC code as well as the necessary hardware.

No doubt, Intel would have expected this virtual RAN market to account for a much bigger share of revenues than it actually does. And because it is the world's dominant provider of CPUs, virtual RAN lift-off might have helped Intel realize much loftier revenue targets. But outside a few greenfield projects, Verizon remains the only brownfield telco to have made a significant investment in virtual RAN. Samsung, its vendor, also has a two-track software strategy, using Intel's CPUs in this market alongside Marvell for purpose-built 5G.

Yet the virtual RAN benefits for Verizon are far from obvious. It trails both T-Mobile and AT&T on 5G availability metrics scored by Opensignal, an independent monitor, and it appears to have no meaningful cost advantage over its competitors. For the recent third quarter, it reported an operating margin of 21%. Ignoring a big impairment charge at the wireline business, AT&T managed roughly the same.

In August, Remy Pascal of Omdia gave virtual RAN a 3% share of the market, and 10% of the baseband subsector. He predicted this share would double by 2028. But that does not guarantee business for Intel. Arm and Nvidia, which has marketed its graphics processing units (GPUs) as general-purpose successors to Intel's CPUs, could figure prominently in future deployments. Telcos, after all, would presumably prefer to have a market not dominated by one struggling chipmaker.

While Ericsson might have little interest in those virtual RAN assets, its potential takeover of purpose-built silicon products would obviously face hurdles. With Ericsson unlikely to offer chips to third parties, one option for merchant silicon would disappear. A foundry move by Ericsson is even more improbable. If it is worried about Intel's future, as other companies are known to be, then a strategic alternative would be to find another manufacturer and a different supplier of chips for the less computationally demanding and more hardware-agnostic RAN software. Nokia has shown it's possible.