Expensive AI hardware sits idle roughly 30-40% of the time waiting for data. That's the problem Nvidia and DDN are finally solving with a new partnership that lets graphics processors grab information directly from storage instead of waiting for CPUs to do it.
DDN's Infinia platform now integrates with Nvidia's Storage-Next initiative to enable GPU-initiated data operations. The practical effect: reduced latency, less CPU overhead, better utilization. For data centers running thousands of GPUs in parallel, even single-digit percentage gains in efficiency translate to millions in savings annually.
The two companies have been aligned since 2016, when DDN first certified with Nvidia's GPUDirect Storage technology. That early work allowed data to bypass system memory entirely, moving straight from storage to GPU. This new collaboration takes it further. GPUs now actively pull what they need rather than passively waiting to receive it.
Sven Oehme, DDN's CTO, framed the update around cost. Jason Hardy at Nvidia's storage division emphasized smooth integration between compute and data infrastructure. The tech fits into Nvidia's 2026 platform roadmap alongside the Rubin processor and BlueField-4 DPUs.
Performance benchmarks will surface at the Flash Memory Summit. For infrastructure teams running large-scale AI operations, the timing matters. Model sizes keep expanding while training costs remain brutal. Any efficiency gain compounds quickly when you're burning through GPU hours at scale.
This article covers technology partnerships and infrastructure developments. It's not financial advice or a recommendation to buy or sell any security.
