AIC’s F2026 DPU-Ready AI Storage Platform Powers World-Leading MLPerf Use Case

AIC, Inc. announced the latest success of its high-density F2026 2U JBOF, a PCIe Gen 5 architecture with DPU-enabled platform designed for next-gen AI and HPC workloads.

With capacity for up to 26×2.5” NVMe drives in just 2U of rack space, advanced baseboard management (BMC) for proactive monitoring, and flexible DPU integration supporting NVIDIA BlueField and other vendors, the F2026 is built to deliver performance, density, scalability, and energy efficiency.

One of the latest real-world validations of the F2026’s capabilities comes from ExponTech’ AI-native enterprise storage platforms. In the MLPerf Storage v2.0 benchmark, ExponTech deployed a single AIC F2026 configured with 24 ScaleFlux CSD 5000 SSDs and 4 NVIDIA BlueField-3 DPUs.

This compact configuration achieved:

• #1 global ranking in ResNet-50 training bandwidth, reaching 23.6GB/s per client.
• Support for 700+ A100 GPU cards and around 400 H100 GPU cards in training scenarios.
• Sustained 96%+ GPU utilization during AI workloads.

Thanks to its flexible architecture and proven real-world performance, the F2026 can scale to 1,024 nodes, delivering EB-scale capacity and 100s of TB/s throughput for large-scale AI training, LLM checkpointing, and enterprise HPC.

“The F2026 exemplifies how smart hardware design can redefine performance per rack unit,” said Michael Liang, chairman and CEO, AIC. “ExponTech’s MLPerf-certified results show how our platform enables partners to deliver world-class AI storage solutions — efficiently and at scale.”

The official MLPerf Storage v2.0 results are available at this webpage.