Supermicro Unveils AI-RAN Servers with NVIDIA Grace CPUs and Two Blackwell GPUs
Supermicro showcased three new AI-RAN systems at MWC Barcelona—ARS-111L-FR, ARS-221GL-NR and ARS-111GL-NHR—featuring NVIDIA Grace CPUs, up to two Blackwell GPUs and 900GB/s NVLink for telecom deployments. Collaborations with SK Telecom’s 1,000-node Haein GPU cluster and Telenor’s sovereign AI cloud platform underline demand for secure, scalable AI infrastructure.
1. Product Launch at MWC Barcelona
Supermicro expanded its AI infrastructure portfolio at MWC Barcelona, unveiling three AI-RAN systems tailored for telecom networks: ARS-111L-FR (1U, short-depth, NVIDIA Grace CPU C1, two low-profile GPU slots), ARS-221GL-NR (2U, Grace Superchip, dual Blackwell GPUs) and ARS-111GL-NHR (1U GH200, 900GB/s NVLink interconnect).
2. Technical Specifications
The ARS-111L-FR supports NVIDIA ConnectX Ethernet adapters for high-speed networking, while the 2U ARS-221GL-NR accommodates NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs; the ARS-111GL-NHR integrates a Grace Hopper Superchip with BlueField DPUs and delivers 900GB/s NVLink bandwidth for edge AI inferencing.
3. Partner Deployments
SK Telecom’s Haein Cluster uses over 1,000 Supermicro servers equipped with NVIDIA Blackwell GPUs to provide GPU-as-a-Service for training and inference, and Telenor has launched Norway’s first sovereign AI Factory with in-country NVIDIA GPU-accelerated infrastructure to guarantee data localization and compliance.
4. Sovereign AI Strategy
As operators pursue data sovereignty, Supermicro’s modular Data Center Building Block Solutions enable rapid, energy-efficient scaling of in-country AI data centers while addressing telecom network requirements, positioning the company to capture new revenue streams in secure AI services.