Micron has confirmed that its 9650 NVMe SSD, built on the PCIe 6.0 interface, has entered mass production, making it the first enterprise drive using the new standard to ship at scale.
Micron’s goal here was to create an ultrafast drive for the AI era. Traditionally, storage was engineered simply to accommodate ever faster processors and networking gear. In today’s AI environments, as models and applications devour ever more data, the ability for storage to handle increasing data transfer rates has become critical for performance.
To support this demanding environment, the 9650 leverages a PCIe 6.0 x4 connection and the NVMe 2.0 protocol to reach sequential read speeds of up to an impressive 28 GB per second and writes of 14 GB per second. In practical terms, that effectively doubles the bandwidth ceiling of PCIe 5.0 drives on the read side and meaningfully lifts write performance as well.
Random throughput also climbs, with read operations reaching roughly 5.5 million IOPS and random writes approaching 900,000 IOPS in higher-end configurations.
Micron touts these gains as offering advanced support for AI training and inference workloads. In other words, Micron, once a vendor of commodity memory chips, is now aggressively positioning itself as a leading supplier for the enterprise AI sector.
Liquid Cooling
The company is offering two primary versions: a Pro model and a Max model. Capacities range from 6.4 terabytes to more than 30 terabytes, depending on configuration. While peak sequential speeds are similar across the product lineup, endurance and mixed workload performance offer different performance levels. The Max models are tuned for heavier write cycles, delivering substantially higher total bytes written over their operational lifespan.
These performance improvements arrive within the same 25-watt power envelope common to the most demanding PCIe 5.0 enterprise drives. In an industry increasingly limited by energy availability, performance per watt matters enormously. But by moving more data without increasing power draw, the 9650 aims to improve overall system efficiency. Faster task completion at steady wattage can translate into lower aggregate energy use for certain workloads.
To deal with heat management at these fast data transfer rates, the 9650 supports both air-cooled and liquid-cooled deployments, with smaller E1.S form factors optimized for integration into liquid environments. As another sign of how AI is changing tech, storage is beginning to follow GPUs into liquid-assisted cooling designs as performance density climbs.
Geared for AI Data Centers
Micron says it has spent the past year and a half validating the drive across the developing PCIe 6.0 ecosystem, working with OEM partners to ensure interoperability. Broader deployment will depend in part on platform support. Enterprise server vendors and accelerator manufacturers must adopt PCIe 6.0 at scale before its bandwidth advantages are fully realized.
For now, the primary audience for the 9650 is AI-focused data centers rather than general enterprise IT. Consumer systems have yet to adopt PCIe 6.0, and mainstream platforms are not expected to support the interface for several years.
Meanwhile, demand from AI infrastructure is absorbing vast quantities of NAND flash and high-speed components, further concentrating next-generation storage in hyperscale environments. Micron is now at the forefront of this trend, working to keep pace as AI pushes storage to meet its voracious demands for data.
