From breakthrough demo to deployment path: SCADA on production-grade PCIe Gen6 hardware at NVIDIA GTC 2026

Breaking performance records is one thing; turning them into deployable infrastructure is another. Nowhere is that gap more visible than in AI infrastructure, where GPUs increasingly outpace storage systems' ability to deliver data at scale. At SC’25, Micron unveiled a major milestone in AI infrastructure performance: 230 million IOPS achieved from a single server, using the NVIDIA SCADA (Scaled Accelerated Data Access) programming model, Micron 9650 PCIe® Gen6 SSDs, Broadcom PEX90000 PCIe Gen6 switches, and the H3 Platform Falcon 6048 PCIe Gen6 server. SCADA is a secure programming model and technology stack that enables GPU-initiated, high-throughput storage access for small-block workloads. It serves as the software cornerstone of the Storage-Next initiative, led by NVIDIA, which brings together Micron and industry partners to define a new class of infrastructure for accessing massive datasets beyond local memory limits.

At GTC 2026, Micron is showcasing SCADA at a clear inflection point, moving beyond a breakthrough demo toward a deployable architecture. 

From innovation to deployment

While the exceptional performance results demonstrated at SC’25 remain unchanged, the underlying ecosystem has matured significantly. All PCIe Gen6 hardware used in the demo is now shipping in production and is commercially available, providing customers with a practical and repeatable path to implementation. 

This transition enables joint Micron and NVIDIA customers to confidently leverage the architecture for developing their own SCADA proofs-of-concept (POCs) using commercially available components.

Micron 9650: The first PCIe Gen6 SSD shipping in production

At the core of the demo system is the Micron® 9650 SSD, the world’s first PCIe Gen6 SSD, now shipping in production (Micron 9650 SSD, the world's first PCIe® Gen6 SSD, reaches a new shipping milestone!). Designed to deliver record-breaking throughput and IOPS, the Micron 9650 is purpose-built for GPU-driven environments such as NVIDIA SCADA, with an architecture optimized for small-block operations critical to scaling next-generation AI workloads.  

Through extensive interoperability testing across the PCIe Gen6 ecosystem, Micron has worked closely across the ecosystem to ensure the Micron 9650 can be reliably deployed as a foundational element of next-generation AI and data-centric infrastructures. 

A practical reference design for SCADA POCs

With PCIe Gen6 production components now available across the ecosystem, Micron’s SCADA demo presented at GTC 2026 provides a clear path for customers to evaluate, prototype, and deploy SCADA-based architectures. The demo illustrates how GPU-initiated storage operations, combined with next-generation interconnects and high-performance SSDs, can unlock unprecedented I/O scalability for workloads such as vector databases, graph neural networks, and large-scale inference pipelines.  

“Today’s AI workloads require accelerated, intelligent storage infrastructure built to rapidly source and process data in seconds to fuel AI agent reasoning,” said Jason Hardy, Vice President, Storage Technologies, NVIDIA. “NVIDIA is working with technology leaders like Micron to engineer the next-generation components necessary to build full-stack AI data platforms designed to meet the industry’s most demanding AI requirements.”

Collaboration that enables the ecosystem

The SCADA demo is powered by production-ready server and PCIe switching platforms from H3 Platform and Broadcom, enabling scalable GPU-driven storage orchestration. NVIDIA GPUs orchestrate storage access using the H3 Platform Falcon 6048 PCIe Gen6 server, which integrates compute, storage, and PCIe Gen6 connectivity into a single, optimized architecture.  

System-wide scalability is enabled by Broadcom® PEX90000 PCIe Gen6 switches, which deliver ultra-low latency, high bandwidth, and the port density required to seamlessly connect GPUs and large numbers of NVMe SSDs. Integrated within the Falcon 6048 platform, these switches allow accelerators and storage devices to operate as a unified PCIe Gen6 fabric, supporting linear performance scaling.

Importantly, the H3 Falcon 6048 server platform and Broadcom PEX90000 PCIe Gen6 switches—alongside the Micron 9650 SSD—are now commercially available, reinforcing the SCADA demo as a deployable reference architecture rather than a forward-looking concept. Together, these production-ready components mark a meaningful step toward closing the gap between AI storage performance breakthroughs and real-world deployment.

Visit the Micron booth at GTC 2026 (booth #1407), March 16-19 in San Jose, CA, to see the SCADA demo in action and learn how the Micron 9650, together with production-ready PCIe Gen6 servers and PCIe switching, is helping shape the future of AI infrastructure.