Nokia’s AI focus accelerates shift away from bespoke baseband silicon in RAN

Nokia’s strategic partnership with NVIDIA and a renewed push towards GPU-accelerated networks signals a disruptive industry move, challenging traditional telecom hardware models and redefining the RAN landscape amid rising virtualisation efforts by competitors like Samsung.

Fuse Wire ServicesNokia’s recent technology pivot away from bespoke baseband silicon toward GPU-accelerated, AI-native radio access networks is reshaping supplier dynamics across advanced mobile networks. According to the announcement of a strategic alliance with NVIDIA, the two companies formalised a $1 billion investment deal in October 2025 that introduces the NVIDIA Arc Aerial RAN Computer and a roadmap for AI-RAN and 6G-ready platforms, signalling Nokia’s intent to move Layer 1 workloads onto general-purpose and GPU-accelerated infrastructures. Industry commentary and Nokia’s corporate materials describe this as part of a broader realignment toward cloud-native, AI-enabled network architectures.

That shift contrasts with Samsung’s recent posture, which has placed virtualised RAN at the centre of its product portfolio. Samsung has converted vRAN from an experimental option into its principal commercial offering in many markets, advancing deployments with Tier One operators and scaling virtualised solutions where proprietary silicon would once have been assumed. The vendor’s public collaborations with European operators reinforce its status as the main alternative to the Nordic suppliers in many Western networks.

Samsung reports substantial field traction for its vRAN software, saying it supports tens of thousands of vRAN sites worldwide and naming major contracts within the United States and across Europe. Company briefings indicate a sizeable footprint with one large U.S. operator and ongoing rollouts in Europe that will rely principally on virtualised stacks rather than custom baseband boxes. These deployments, Samsung argues, demonstrate carrier-grade performance and operational maturity that underpins wider commercial confidence in software-defined RAN.

The economics driving this transition are straightforward: data-centre silicon benefits from broader scale and faster product cycles than bespoke telecom chipsets. Intel’s Granite Rapids Xeon family is highlighted by vendors as enabling higher core densities that shrink server footprints for vRAN workloads, while AMD’s higher-core x86 parts and Arm-based designs including NVIDIA’s Grace CPU are being evaluated as alternatives. At the same time, specialised functions such as Forward Error Correction remain computationally demanding and expose gaps in some general-purpose platforms’ acceleration features, prompting a mix of software optimisation and selective hardware offloads.

Nokia and Samsung are approaching GPU integration differently. Nokia’s roadmap, driven in part by its NVIDIA partnership, focuses on creating GPU-compatible Layer 1 software to accelerate intensive baseband tasks and to prepare for an AI-native 6G era. Samsung, by contrast, has emphasised targeted GPU use, chiefly for AI-driven radio optimisation such as beamforming, while continuing to rely on high-density CPUs for mainstream vRAN functions. Both vendors note that commercial off-the-shelf servers from major OEMs can host GPU cards, but they diverge over the extent to which GPUs will displace CPU execution for primary baseband processing.

Industry voices from Samsung have been explicit about the trajectory. “We’re still selling our purpose-built BBUs to a number of customers, but I do believe that it’s a matter of time,” Alok Shah, Vice President of Network Strategy at Samsung Networks, told Light Reading during MWC Barcelona. “Virtual RAN performance has reached parity,” he added, and later noted that “Granite Rapids has plenty of capacity to support AI algorithms on-platform.” Those remarks reflect Samsung’s conviction that general-purpose compute now delivers competitive performance while lowering the cost and complexity of building modem silicon.

Ericsson, meanwhile, continues to defend investment in its own silicon designs while keeping options open for customers preferring virtualised deployments on standard CPUs. Its commercial vRAN offerings today run on Intel processors and the vendor emphasises portability in its software engineering so operators can choose hardware over time. At the market level, the combined pressures of rising development costs for proprietary chipsets, a contracting device market and operator moves to virtualisation are eroding the case for new bespoke RAN silicon, even as some vendors maintain proprietary routes as a strategic hedge. Observers note that corporate restructurings tied to AI-targeted investments add further momentum to this industry-wide realignment.

Source Reference Map

Inspired by headline at: ^[1]

Sources by paragraph:

Source: Fuse Wire Services