Nokia expanded its AI-RAN work with NVIDIA on October 21, 2026, describing plans for AI in radio processing, computing closer to data sources, and AI agents for network operations. Separately, Nokia reported a call on T-Mobile’s live network in Bellevue, Washington, and Supermicro announced its own infrastructure plans for AI-RAN deployments with Nokia and NVIDIA.
What Nokia expanded in its AI-RAN work
AI-RAN brings artificial intelligence into radio access network (RAN) functions and the computing that supports them. Nokia’s October 21 announcement described work with NVIDIA on AI for radio processing, computing nearer to data sources, and AI agents that could support operations across multiple network domains. Nokia’s AI-native anyRAN software is integrated with NVIDIA computing systems.
Nokia also reported that Elisa, BT, Deutsche Telekom, and Orange were preparing real deployments and commercial-service evaluations from 2027. That is a planned evaluation horizon, not a report of a broad commercial rollout.
How AI and radio workloads share computing
Nokia describes using accelerated computing for both radio functions and AI workloads. Running those workloads on shared computing infrastructure is intended to make use of that computing across different tasks. Nokia says latency needs, limits on devices, and data sovereignty can influence whether AI inference runs at the network edge, closer to where data is generated.
How AI and radio workloads share computing
At Mobile World Congress 2026, Nokia demonstrated a phone sending video over 5G to a GPU-equipped virtualized distributed unit (VDU), where a model analyzed camera frames. The demonstration illustrated one way to connect a mobile device, a radio network, and AI computing.
What Nokia reported about T-Mobile’s live network
Nokia said it, T-Mobile, and NVIDIA completed an AI-RAN call at a commercial cell site in Bellevue using T-Mobile’s live production network. Nokia described the setup as using a Massive MIMO radio, 5G anyRAN software, NVIDIA Aerial RAN Computer-based server infrastructure, a commercial 5G core, and commercial user equipment.
Supermicro’s separate AI-RAN systems
Supermicro announced infrastructure configurations combining its servers, NVIDIA Aerial RAN Computer, and Nokia radio-network technology and software. Its product plans cover several network locations and workloads:
- ARC-Compact: a short-depth 1U system intended for edge-network functions such as Service Management and Orchestration.
- Dual-node ARC-Pro: each node has an NVIDIA Grace C1 CPU and an NVIDIA RTX PRO 4500 Blackwell Server Edition GPU. Supermicro lists up to 1.6 PFLOPS of FP4 AI performance.
- ARC MGX 2U: a choice of dual NVIDIA Vera CPUs or dual x86 processors, alongside two NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs.
- ARC MGX 4U: up to eight NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. Supermicro separately describes 4U AI servers with dual x86 CPUs and up to 6 TB of DDR5 memory.
Supermicro said its NVIDIA ARC servers were generally available on October 21, 2026. It expected the ARC MGX 2U with dual Vera processors later in 2026, planned customer testing of 2U ARC-Pro later that year, and scheduled general availability of 2U ARC-Pro for the first half of 2027.
What the 10%–15% result measures
Nokia reported that initial laboratory measurements showed a 10%–15% improvement in uplink performance. Uplink is the direction data travels from a device toward the network. This is a laboratory result, separate from the Bellevue call on T-Mobile’s production network.
