Ericsson and Nokia Diverge on RAN Hardware Strategy
Telecom leaders are split between custom silicon and GPU architectures as they race to boost mobile network efficiency.
Updated on Oct. 21, 2026 in Semiconductors
Live Poll
Do you trust specialized, custom-built hardware over widely available, standardized technology products?
Ericsson and Nokia are diverging on the hardware foundation for Radio Access Network (RAN) infrastructure, with Ericsson favoring custom ASICs while Nokia shifts toward GPU-based architectures. Ericsson maintains its reliance on proprietary silicon for current network deployments, contrasting with Nokia's new $1 billion strategic agreement with Nvidia.
Why it matters
This split highlights a fundamental disagreement over whether future cellular networks are constrained by transmission timing or raw computational throughput. The outcome of this architectural divergence will define the performance and efficiency standards for 5G and future network generations.
Ericsson currently utilizes custom ASICs—application-specific integrated circuits designed for single-purpose efficiency—for its Layer 1 RAN software to maintain cost and functional competitiveness. In contrast, Nokia aims to leverage GPU acceleration to achieve a projected doubling of spectral efficiency by 2028.
The players
Ericsson
A global provider of mobile network equipment that relies on proprietary custom silicon to maintain current network margins.
Nokia
A telecommunications infrastructure provider that is transitioning to GPU-heavy architectures through a partnership with Nvidia.
Nvidia
A designer of graphics processing units that is partnering with Nokia to provide the computational foundation for next-generation RAN products.
Per Narvinger
The CEO of Ericsson who is managing the company's commitment to custom silicon despite pressure from GPU-centric industry trends.
The details
Ericsson utilizes custom silicon to manage Layer 1 RAN functions, which handle the physical transmission of data between basestations and devices. Their strategy centers on the belief that RAN algorithms are limited by transmission time intervals rather than processing power. Conversely, Nokia is pivoting toward GPU-based RAN, employing Nvidia chips to handle complex calculations, which the firm argues enables more efficient AI integration compared to traditional processor designs.
Timeline
Q2 2026: Nokia reported its mobile infrastructure operating margin.
October 2026: Ericsson CEO Per Narvinger defended the company's custom ASIC strategy following Q3 results.
2028: Nokia’s target year for doubling spectral efficiency through GPU-based products.
The Tech Race
The transition to GPU-based RAN represents a significant departure from the industry-standard reliance on custom ASICs for Layer 1 network processing. Nokia is attempting to leapfrog current designs by integrating Nvidia GPUs, while Ericsson doubles down on its existing silicon-based performance trajectory.
Network operators will soon choose between proprietary silicon-based ecosystems and more generalized GPU-accelerated infrastructure. These architectural choices will dictate how quickly providers can deploy new AI-native link adaptation features that are currently active on 60,000 Ericsson basestations.
The takeaway
The race to redefine spectral efficiency has split the industry into two camps: one betting on optimized, custom silicon and the other on high-performance general-purpose GPUs. Watch for the 2028 benchmark to see if GPU-based architectures successfully deliver the promised twofold increase in spectral efficiency.
Further reading
For more on the hardware driving connectivity, see our coverage in Semiconductors.
Live Poll
Do you trust specialized, custom-built hardware over widely available, standardized technology products?
