Super Micro Computer (SMCI) has introduced a broad portfolio of AI infrastructure systems built with AMD processors, graphics accelerators and networking technology. The company announced new GPU-accelerated servers, enterprise server platforms and rack-scale computing systems aimed at enterprise, cloud and high-performance computing deployments. The announcement deepens Super Micro's long-standing partnership with AMD and positions the company to capture demand from customers seeking alternatives to NVIDIA-centric AI server architectures.

SMCI stock ended trading at $31.03, up 1.54%, after the announcement of the expanded AI infrastructure lineup.

H15 Server Series Targets Enterprise and Cloud Workloads

Super Micro Computer launched its H15 server family based on 6th Generation AMD EPYC 9006 Series processors. The systems are designed for enterprise workloads such as cloud infrastructure, data storage, virtualization, high-performance computing and other demanding software environments. The EPYC platform competes directly with Intel's Xeon processors in the data center CPU market, where AMD has steadily gained share over recent generations.

According to the company, the H15 portfolio is intended to provide higher compute density without requiring additional power infrastructure. The platform supports up to 256 processing cores and 512 simultaneous threads, while also offering expanded memory capacity and improved input/output throughput. Super Micro said those upgrades are intended to support more complex enterprise applications and AI workloads.

The H15 server collection includes Hyper, CloudDC, GrandTwin, FlexTwin, Petascale Storage and SuperBlade architectures. The configurations are aimed at a range of deployment environments, including cloud service providers, corporate data centers, academic institutions and large-scale operators. The platforms can be configured with traditional air cooling or advanced liquid cooling, depending on customer requirements.

AMD Instinct GPU Servers Expand AI Capabilities

Super Micro also expanded its AI infrastructure lineup with PCIe GPU servers using AMD Instinct MI350P accelerators. The AS-5126GS-TNRT and AS-5126GS-TNRT2 systems can house ten GPUs in a 5U air-cooled enclosure, allowing customers to add AI processing capacity while continuing to use air-cooled data center designs. The air-cooled approach matters because many existing data centers are not yet equipped for liquid cooling, lowering the barrier to adoption for organizations with legacy facilities.

Each of the two server models uses AMD Instinct MI350P GPUs with up to 144GB of HBM3e memory per accelerator. The systems support reduced-precision computing formats commonly used in AI inference and training. Super Micro said the design is intended to increase performance while reducing infrastructure costs and data center footprint requirements.

The company also integrated AMD Pensando Pollara 400 AI network adapters across its AI infrastructure offering. The networking technology is designed to handle front-end communications, storage transfers and scale-out connectivity through standard Ethernet architectures. Super Micro said this approach allows organizations to scale AI clusters while maintaining compatibility with existing network environments. AMD acquired Pensando in 2022, and the technology has since been integrated across its data center product stack as a differentiator against competitor networking solutions.

Helios Platform Offers Rack-Scale AI Architecture

Super Micro introduced the Supermicro AMD Helios Platform for large-scale AI deployments. The rack-level system includes 72 GPU accelerators using AMD Instinct MI455X processors, paired with 6th Generation AMD EPYC CPUs. The platform uses liquid cooling to support higher performance and operational efficiency for intensive computing workloads. Rack-scale liquid-cooled AI systems have become a focus area across the industry as power densities exceed what conventional air cooling can handle.

The Helios Platform also incorporates AMD Pensando networking components and the ROCm software framework. ROCm serves as AMD's open-source alternative to NVIDIA's CUDA platform, and its continued development is central to AMD's effort to win developers and AI researchers who have largely built workflows around CUDA-compatible tools. The integrated system is designed to support large AI model training workloads and high-volume inference deployments. Customers can deploy configurations ranging from individual servers to multi-rack installations.

The product launches are part of Super Micro Computer's Data Center Building Block Solutions strategy, which combines server hardware, storage systems, networking equipment, liquid cooling technology, manufacturing capabilities and deployment services into integrated infrastructure offerings.

Super Micro Computer said it plans to demonstrate the new platforms at AMD Advancing AI Day 2026 in San Francisco. The showcase is expected to include a FlexTwin rack featuring 96 AMD EPYC 9006 processors.