DDN partners with NVIDIA to push GPU-initiated data access for next-generation AI

DDN, a provider of data intelligence and AI storage infrastructure, said it is collaborating with NVIDIA to advance “GPU-initiated data access” for next-generation AI workloads. The announcement, published August 4, frames the work around NVIDIA’s Storage-Next initiative, which targets next-generation approaches to storing, retrieving, and managing data for AI systems.

Under the collaboration, DDN is positioning its platform and technology portfolio to help organizations pursue a model where accelerators such as GPUs can initiate data access in a more direct way. The company’s release does not lay out specific engineering details in the announcement, but it ties the effort to performance and efficiency needs that arise when AI training and inference increasingly depend on large, fast-moving datasets.

The announcement is also notable because it targets the “data access” layer rather than only compute. In modern AI systems, throughput and latency are often bottlenecks not just for GPUs but for the end-to-end pipeline that feeds them, including where data resides, how it is requested, and how quickly it can be delivered to the compute that needs it.

DDN did not provide deal terms, implementation timelines, or named customer deployments. It also did not specify whether the collaboration is focused on a particular product line, such as software-defined storage, data management, or performance-optimized data services, beyond referencing its broader “AI and data intelligence solutions” positioning.

The NVIDIA involvement, according to the release, is anchored to Storage-Next, an initiative NVIDIA uses to describe work aimed at next-generation storage and data infrastructure for AI. NVIDIA is the company behind the GPUs and networking technologies that many AI platforms build around, so initiatives like Storage-Next typically matter to system architects who are trying to reduce friction between storage and GPU compute.

For the sector, GPU-initiated data access reflects a broader industry push toward more “data-aware” architectures. As AI models grow and training runs become more distributed, organizations have sought ways to streamline data movement, improve scheduling, and reduce the overhead that can occur when data access is coordinated separately from compute.

At this stage, the announcement leaves several questions open. DDN did not disclose which specific software interfaces or reference architectures would be used to enable GPU-initiated access, nor did it provide benchmark results, partner integration guidance, or public availability dates for any related components.

Investors and customers may want to watch for follow-on updates that specify technical milestones, integration pathways with NVIDIA platforms, and any performance targets. NVIDIA and DDN may also provide additional clarity on how the approach would be deployed in production environments, particularly for training clusters and inference deployments that have different latency and throughput requirements.