Xsight Labs has announced a partnership with Hammerspace to develop an open architecture aimed at simplifying and improving data storage infrastructure for artificial intelligence (AI) workloads in data centers. The initiative is centered around the Open Flash Platform (OFP), which leverages Xsight Labs’ E1 800G Data Processing Unit (DPU) and Hammerspace’s orchestration software to create a new approach to AI warm storage.

The OFP design eliminates traditional storage servers and relies on direct connections between flash storage and the network. Storage devices communicate directly using open standards such as Network File System and Linux, which Xsight Labs claims unlocks higher performance, increases efficiency, and reduces operational expenditure (OPEX) by removing costly, power-intensive middle-layer storage servers. According to the companies, the OFP blueprint is intended to make flash storage with the NFP file system the backbone for AI infrastructure by reducing data path complexity.

For data center environments, the E1 DPU integrates with Hammerspace’s orchestration stack to enable Graphics Processing Units (GPUs) to access data directly from flash devices over the network, bypassing x86 server bottlenecks. This direct path is intended to accelerate AI training and inference workloads and allows for linear scalability with increasing flash capacity. Xsight Labs notes that its DPU was chosen for its Arm core density, high memory bandwidth, and 800 Gbps Ethernet connectivity, which meet high-performance requirements for flattened AI storage architectures.

The companies report that early access deployments of the solution are underway with strategic partners, and demonstration is planned for the 2025 OCP Global Summit taking place October 13–16 in San Jose, California. Limited volume shipments are expected in the fourth quarter of 2025, with full production systems available in early 2026.

Source: Xsight Labs