Ericsson CEO Defends Custom Silicon Over Nvidia GPUs in RAN

Ericsson's new CEO Per Narvinger has dismissed the idea of replacing the Swedish vendor's in-house chips with Nvidia GPUs, telling equity analysts that its custom silicon is an "extraordinary asset" and revealing that AI-native link adaptation already runs commercially on more than 60,000 basestations with three operators.

The remarks, made on Ericsson's third-quarter results call last week, address a question that has shadowed the industry since Nokia struck its $1 billion pact with Nvidia about a year ago. Under that deal, Nokia plans to design RAN products based on Nvidia's data-center GPUs, eventually displacing Marvell Technology as its custom silicon partner.

Narvinger sees no reason to follow. "We have disaggregated our software and hardware, meaning we have a lot of flexibility on what path to embark on in the future," he told analysts. "Having said that, right now, we see that it's an extraordinary asset to have your own ASICs. That really makes it possible for us to get our functionality fast. It makes it possible for us to be very cost competitive."

Why the silicon choice matters

Ericsson has designed its own application-specific integrated circuits for Layer 1, the most demanding slice of RAN software, throughout the 5G era. Nokia took a different path after early 5G setbacks with Intel, first partnering with Marvell and now planning a move to Nvidia.

The technical debate turns on timing constraints. Algorithms in a RAN must work within the transmission time interval, meaning key calculations cannot exceed 500 microseconds. Ericsson's engineers argue software built for those conditions simply does not need a GPU's processing muscle.

Nokia claims its GPU strategy can double spectral efficiency by 2028. Ericsson's own published gains from custom silicon are much lower. But several experts, including former Nokia CTO Marcus Weldon, expect the biggest improvements to be limited to a small number of use cases.

"A site already achieving near-theoretical efficiency under classical processing has less headroom for AI-driven improvement than a site operating well below its theoretical ceiling due to interference, channel estimation errors, or suboptimal scheduling," said Chetan Sharma, founder of Chetan Sharma Consulting, in a recently published white paper.

What is already deployed?

The commercial contrast is central to Ericsson's case. Nokia's approach requires operators to install GPUs in their networks, while Ericsson points to silicon already in the field supporting AI features.

"It's been around in the industry for as long as I've been, so more than three decades, and now replacing that with an AI model we get 10% gain in spectral efficiency," said Narvinger of link adaptation, a RAN algorithm.

Ericsson has not shut the door entirely. Narvinger said the company is "not ruling out any other platform," and its disaggregated cloud RAN track runs most Layer 1 functions on Intel CPUs, with only forward error correction offloaded to an Intel hardware accelerator. Ericsson says it could port that software to other CPU platforms after small tweaks, and during trials it has already moved software to an Arm-based Nvidia CPU.

Can the margins hold?

The economics remain the sharpest question. Critics doubt Ericsson can sustain custom silicon investment in what Narvinger himself calls a "flat" RAN market, since general-purpose chips serve much larger markets. Narvinger counters that Ericsson is building on years of development rather than starting from scratch, and that CPUs and GPUs carry the cost of supporting features that may never be used. Ericsson's latest earnings report included another warning about rising component costs driven by the AI boom.

The numbers give him some room. At the RAN-making mobile networks unit, which accounts for almost two thirds of Ericsson's revenues, sales rose 9% year-over-year to SKr38.7 billion ($3.9 billion) for the quarter. Gross margin slipped 0.2 percentage points to 49.8%, but operating margin improved 1.8 points to 21.8%.

Nokia, yet to report third-quarter results, posted a second-quarter operating margin of 11.6% at its mobile infrastructure group, down from 12.2% a year earlier. Some analysts also suggest Nokia's Nvidia deal reflected economic necessity rather than architectural conviction, after Marvell reportedly balked at custom silicon returns following Nokia's loss of US market share. If ongoing GPU RAN trials with operators such as T-Mobile US deliver measurable gains, Ericsson's ASIC defense will face a harder test in the quarters ahead.