New FADU Drive Enables Direct File Access for GPUs
The file-semantic SSD avoids CPU overhead to deliver a 2.5 times performance increase over existing Gen5 storage.
Updated on Oct. 15, 2026 in Semiconductors
Live Poll
Do you believe new AI storage technologies will effectively reduce national energy consumption and costs?
FADU unveiled the first file-semantic solid-state drive at the 2026 Open Compute Project Global Summit, a technology currently at the demonstration stage. The drive enables GPUs and DPUs to process files directly by bypassing CPU routing.
Why it matters
This architecture aims to reduce compute overhead, latency, and power consumption within AI infrastructure. By shifting file management tasks to the storage controller, the design addresses a major bottleneck in large-scale data processing.
The 16-terabyte drive, powered by the sixth-generation FC6161 controller, consumes 28 watts during sequential writes and 5 watts in standby. It achieves a 2.5 times performance increase versus Gen5 storage technology.
The players
FADU
A developer of storage controllers and solid-state drive technology focused on power efficiency and low-latency data architecture.
The details
The controller uses custom firmware to manage file-to-address translation, file searches, and access permissions directly on the drive. By embedding file requests into standard Non-Volatile Memory Express (NVMe) commands—the industry-standard protocol for high-speed storage access—the system eliminates the need for the CPU to intermediate file requests between the storage and high-performance computing components like GPUs.
Timeline
October 2026: FADU unveiled the file-semantic SSD at the Open Compute Project Global Summit in San Jose.
The Tech Race
This development moves beyond current Gen5 storage standards by shifting file management logic closer to the physical data. It sets a new benchmark in the race to optimize AI infrastructure by reducing the traditional dependency on CPU-based file routing.
Enterprise data centers and AI infrastructure operators will be the first to integrate this hardware once it moves from demonstration to production. The technology effectively replaces standard block-level storage in workflows requiring high-speed direct file access by GPUs and DPUs.
The takeaway
FADU intends to scale this technology by providing reference designs to original equipment and design manufacturers. Observers should track the release of these reference designs, which will determine how quickly this file-semantic capability enters the commercial server market.
Further reading
For broader trends in high-performance storage hardware, see our latest coverage on Semiconductors.
Source note: This article includes information reported by The Korea Times.
Live Poll
Do you believe new AI storage technologies will effectively reduce national energy consumption and costs?
