NVIDIA Forms 6G Alliance with Global Telcos
NVIDIA and a coalition of global telecom giants—including T-Mobile, SoftBank, and Deutsche Telekom—have committed to building 6G networks on open, AI-native platforms. The alliance aims to standardize on GPU-accelerated, software-defined radio access networks (RANs). This move positions NVIDIA's hardware as foundational for the next generation of AI-powered edge intelligence and IoT.
This initiative is part of the broader AI-RAN Alliance, which launched in February 2024 and now includes over 130 members such as AWS, Arm, Microsoft, Nokia, and Samsung. The group's work is split into three key areas: using AI to improve the radio access network (AI for RAN), integrating AI and RAN processes for efficiency (AI and RAN), and deploying AI services at the network edge (AI on RAN). The fundamental driver for this shift is that current 5G networks were not architected to support the coming wave of "physical AI," which will require connecting billions of autonomous machines, vehicles, and sensors. This creates a new category of "AI traffic" that requires networks to function as a distributed AI compute fabric, not just a system for human-centric data and voice. This AI-native infrastructure is envisioned to enable agentic AI to manage network operations autonomously. Researchers are already designing architectures like an "Agentic-AI Core" where a large-scale model fine-tuned on telecom knowledge, dubbed NetGPT, can generate and execute network procedures from natural-language intents without human intervention. This would allow the network to self-optimize and self-heal. For enterprise governance, an AI-native 6G network allows for continuous, automated security and compliance. Agentic AI systems could constantly validate the live network against formal specifications from bodies like 3GPP and NIST, automating risk management and ensuring that deployed functions adhere to regulatory requirements. The move to a software-defined, GPU-based architecture unlocks new enterprise business models. Since RAN infrastructure is typically built for peak traffic, its compute capacity is often underutilized. This new model allows for co-hosting AI workloads, enabling telcos to offer "GPU-as-a-Service" or process AI tokens at the edge during off-peak hours. NVIDIA's infrastructure-focused alliance is not the only one shaping the 6G landscape. Semiconductor rival Qualcomm is simultaneously building its own coalition centered on the "demand side," bringing together over 30 device makers and IoT companies, including LG Electronics, to develop AI-based 6G technologies for a new generation of hardware. Live field trials are already moving AI-RAN from the lab toward commercial reality, with commercial 6G launches expected around 2030. In one benchmark, a software-defined AI-RAN running on a single NVIDIA GH200 server achieved a throughput of 36 Gbps with latency under 10 milliseconds.
