Kuaikeji, June 18 – Based on NVIDIA's official announcement and the public roster of the AI-RAN Alliance unveiled at MWC 2026, NVIDIA is forging ahead with a groundbreaking technical approach: employing general-purpose GPUs to supplant traditional application-specific integrated circuits in 5G base stations.

The AI-RAN Alliance (Artificial Intelligence Radio Access Network Alliance), under NVIDIA's helm, has expanded from its initial 11 founding members to a robust group of 132. Established during MWC 2024 with NVIDIA at the forefront, the alliance boasts industry titans such as Ericsson, Nokia, Samsung, and Qualcomm among its ranks.

NVIDIA posits that the fundamental operations of the communication physical layer align closely with AI deep learning computations. Leveraging the CUDA cores and Tensor Cores within GPUs, NVIDIA can not only execute parallel computations with high efficiency but also extend the technology from the cloud to low-power, distributed locations beneath cell towers.

In the face of NVIDIA's aggressive push, the two traditional industry behemoths have adopted divergent strategies. Nokia has opted for a full embrace of NVIDIA, transitioning its previously Marvell-customized RAN chips to NVIDIA's GPU platform and cultivating its Layer 1 RAN business on the CUDA software framework.

Conversely, Ericsson remains steadfast in its commitment to self-developed ASICs and has formed an alliance with Intel to bolster CPU solutions. With RAN R&D investment reaching $5.2 billion last year, Ericsson contends that ASICs possess irreplaceable cost and temperature control advantages, especially amidst declining operator RAN expenditures.

Industry analysts observe that this software-defined network offensive, spearheaded by the AI chip giant, is compelling traditional telecom equipment vendors to reassess their survival strategies. Should GPUs emerge as the default computing power foundation for wireless networks, traditional equipment vendors risk being relegated from system integrators to mere hardware assemblers.