Supermicro Expands Solutions for AI-RAN and Sovereign AI Infrastructure

Supermicro announced expanded support for infrastructure solutions designed for sovereign AI platforms and AI-RAN (Artificial Intelligence – Radio Access Network). The new systems are developed to deliver high performance, energy efficiency, and scalability for telecommunications networks and AI infrastructures. Some of these technologies were presented during Mobile World Congress Barcelona 2026 – one of the largest events for the telecom industry.

The company’s solutions use the latest CPU and GPU technologies that support the development of AI-oriented telecommunications networks and infrastructures for so-called AI factories – environments for training and running artificial intelligence models.

AI-RAN infrastructure for next-generation networks

The use of artificial intelligence in telecommunications networks is increasing rapidly as operators seek greater efficiency and automation. AI-RAN integrates intelligent capabilities directly into the radio access network, enabling optimization of spectrum, energy consumption, and network performance. In addition, this model creates a distributed infrastructure capable of supporting AI applications both for operators and end users.

Such an architecture requires infrastructure solutions tailored to the specific requirements of telecom networks and capable of scaling in a cost-efficient way. Supermicro is expanding its portfolio of AI-RAN systems developed in accordance with the NVIDIA Aerial RAN Computer architecture, supporting new technologies such as NVIDIA Blackwell and the upcoming NVIDIA ARC-Pro platform.

New server systems for AI-RAN

Among the solutions presented by the company are several server platforms designed for AI and telecom infrastructures.

• ARS-111L-FR is a compact 1U short-depth server developed for distributed RAN workloads. The system uses the NVIDIA Grace CPU C1 and supports high-speed connectivity through NVIDIA ConnectX Ethernet adapters. In addition, the server provides space for up to two low-profile GPU cards, including the NVIDIA L4 GPU, designed for AI workloads.
• ARS-221GL-NR is a high-performance 2U system based on the Grace Superchip that supports up to two double-width GPU cards. The system has been updated to support the NVIDIA RTX PRO 6000 Blackwell Server Edition, enabling the execution of more complex AI tasks in telecommunications networks.
• ARS-111GL-NHR is a 1U system with the NVIDIA GH200 Grace Hopper Superchip, which includes an integrated GPU and a 900 GB/s NVLink connection between the CPU and GPU.

The combination of high-speed chip-to-chip connectivity, along with support for NVIDIA BlueField DPUs and ConnectX Ethernet adapters, makes the system suitable for both research workloads and AI inference at the edge.

Sovereign AI and new opportunities for telecom operators

The concept of sovereign AI refers to the ability of organizations to develop and manage AI solutions within their own infrastructure, controlled data environments, and regulatory compliance frameworks. The growing demand for such platforms creates new opportunities for telecommunications operators, who can offer AI infrastructure as a service located within national borders.

Deploying sovereign AI at large scale requires infrastructure that is powerful, scalable, and energy efficient. Supermicro’s Data Center Building Block Solutions (DCBBS) architecture enables rapid deployment and expansion of AI data centers. The modular design simplifies scaling, while optimized cooling and power solutions improve infrastructure efficiency.

Real deployments with global partners

During MWC, the company demonstrated real deployments together with several leading telecom operators and technology partners.

• Telenor presented Telenor AI Factory – the first sovereign AI cloud platform in Norway. The platform uses infrastructure with NVIDIA GPU acceleration, deployed entirely within the country, ensuring that data does not leave national borders. It is designed for companies, startups, and public organizations and supports the deployment of AI solutions while meeting requirements for handling sensitive data.
• SK Telecom is developing the Haein Cluster – a large-scale sovereign AI infrastructure for heavy AI workloads. The cluster includes more than 1,000 Supermicro AI servers equipped with NVIDIA Blackwell GPUs and is deployed in the Gasan AI Data Center. The platform provides GPU-as-a-Service for AI model training, inference, and development.
• Nokia demonstrated solutions based on Supermicro hardware. Among them is AnyRAN, a software package that enables operators to build RAN networks using cloud-native applications across different infrastructures. The ARS-111GL-NHR system has been validated to run AnyRAN. The company also presented an infrastructure design for combined AI/ML workloads based on Supermicro AS-8125GS-TNMR2 servers.
• Viettel High Tech uses Supermicro infrastructure in the deployment of a 5G network based on the Open RAN standard. The use of an open and modular architecture enables faster deployment of new technologies and reduced costs. Part of the infrastructure includes Supermicro 1U short-depth edge servers that provide performance and flexibility for modern wireless networks.