RAN Silicon Rethink- Part II; vRAN and General-Purpose Compute

Key Takeaways

• RAN market shrank $10B, now $35B
• Samsung leads vRAN with 53k sites
• Intel Granite Rapids enables denser server footprints
• Nokia pivots to Nvidia GPUs, reducing custom silicon
• Ericsson sticks to proprietary silicon, but supports CPUs

Summary

The global RAN market fell by roughly $10 billion, dropping from about $45 billion in 2022 to $35 billion in 2024. Vendors are diverging on hardware strategy: Samsung and Nokia are embracing virtualized RAN on general‑purpose CPUs, while Ericsson clings to proprietary silicon and Intel processors. Samsung now leads vRAN deployments, supporting 53,000 sites worldwide, especially with Verizon and Vodafone. Intel’s Granite Rapids Xeon and emerging GPU options are shaping the next wave of AI‑enhanced RAN.

Pulse Analysis

The RAN market’s $10 billion contraction reflects broader macro‑economic pressures and a maturing 5G rollout. As operators face tighter budgets, the economics of custom ASICs have eroded, prompting the big five vendors to reassess their roadmaps. Samsung’s aggressive vRAN push, backed by contracts with Verizon and Vodafone, illustrates how a software‑centric model can capture market share once dominated by purpose‑built baseband chips. Meanwhile, Nokia’s partnership with Nvidia signals a hybrid approach, leveraging GPU acceleration to offset the loss of dedicated silicon performance.

At the heart of the vRAN surge is the rise of high‑density, general‑purpose processors. Intel’s Granite Rapids Xeon platform, with its expanded core count, allows a single server to replace two legacy units, cutting both power consumption and rack space. Operators testing the platform report parity with traditional baseband performance, while AMD’s 84‑core offerings and Arm‑based Nvidia Grace CPUs present alternative pathways for future AI workloads. However, offloading compute‑intensive tasks like Forward Error Correction remains a bottleneck, giving Intel’s dedicated FEC accelerator a competitive edge.

For the telecom ecosystem, the move toward commoditized compute reshapes capex planning and opens doors for AI‑at‑the‑edge services. Vendors that can seamlessly integrate GPUs for beamforming or inference—without sacrificing the reliability of x86 servers—will likely dominate the next phase of 5G and emerging 6G deployments. Ericsson’s commitment to proprietary silicon may limit its flexibility, whereas Samsung’s COTS‑based strategy positions it to scale rapidly across North America and Europe. As operators prioritize cost‑effective, software‑defined networks, the balance will tip further toward general‑purpose CPUs, with GPUs playing a supplemental, use‑case‑specific role.