Volume 343, Issue 3 · 8 articles from 2 issues
DDN expects bookings growth well beyond 3x and pitches storage that keeps every GPU busy as inference spending overtakes training.
VMblog reports from DDN's IT Press Tour briefing in Palo Alto, where president and CRO Kevin Delane said bookings are running at about 2x and should finish well beyond 3x year over year, with more than 5,000 customers and over 1 million GPUs served.
With demand outrunning GPUs, power and floor space, DDN's goal is to keep every installed GPU at 99% to 100% utilization, and the briefing cited Gartner's view that inference spending passes training in 2026.
CPO Omer Asad detailed the technology, including Infinia returning cached context over RDMA directly to GPU memory.
Executives also acknowledged a branding gap that new logo-publicity efforts aim to close.
DDN says Nebul uses DDN Infinia to store and reuse KV cache, cutting GPU work for sovereign AI inference.
DDN announced at NVIDIA GTC Berlin 2026 that European AI cloud provider Nebul is using DDN Infinia to accelerate inference for sovereign workloads on its NVIDIA-powered infrastructure.
Infinia stores and reuses key-value (KV) cache, reducing the GPU work needed to serve models.
DDN says testing showed Nebul's workloads found answers faster and at lower cost.
Nebul measured results on its own platform with the same eight GPUs, model and workloads in both configurations.
DDN pitches the economics of AI storage, listing eight ways data platforms waste GPU time in training and inference.
At the 70th IT Press Tour, DDN focused on AI economics rather than raw performance, under the tagline 'We don't make GPUs faster. We make them work, profitably.'
Citing Gartner, it noted AI-optimized IaaS spending of $42 billion in 2026, with inference overtaking training at 55%.
DDN lists eight ways storage erodes GPU economics: GPUs waiting on data, checkpoint stalls, noisy neighbors and costly expansion in training, and prefill delays, agentic loops recomputing context, scaling slowdowns and weak tenant isolation in inference.
Its answer is two data engines, EXAScaler and Infinia, under one namespace and closely tied to NVIDIA platforms.
WWT and DDN publish a storage reference design for 50MW+ AI factories that saves power for GPUs and speeds KV cache loading.
World Wide Technology and DDN announced a storage reference design for AI factory campuses of 50 megawatts and above, aimed at feeding GPUs fast enough while leaving more of the fixed power budget for compute.
Built on NVIDIA's Vera Rubin NVL72 reference architecture, it targets more than 2.5 GB/s of read throughput per GPU, with one 2U DDN AI400X3M appliance rated at 190 GB/s for a full rack.
For inference, DDN Infinia on NVIDIA BlueField-4 STX shares KV cache across nodes; DDN reports loading a 112K-token cache in 2.1 seconds versus 57 seconds to recompute.
WWT integrates the components as one system.
DDN describes how M2M Tech's smart-city system on DDN Infinia and NVIDIA turns traffic camera alerts into actionable context.
DDN uses a busy Seattle intersection to show the gap between detecting a problem and knowing what to do about it.
Traditional systems fire an alert and leave operators to check feeds, permits and crews by hand.
M2M Tech built a smart-city solution for municipalities on DDN Infinia and NVIDIA accelerated computing to close that gap.
DDN argues enterprise AI is shifting from training to inference, where value depends on keeping data moving at the pace of models and GPUs.
DDN explains how European NeoCloud Nebul stores KV cache on DDN Infinia to make inference faster, more resilient and more efficient.
DDN says Nebul, a European NeoCloud running AI workloads for organizations whose data must stay in Europe, uses DDN Infinia beneath its inference stack.
Infinia stores the model's key-value (KV) cache where the whole GPU fleet can reach it, so GPUs restore context instead of rebuilding it.
DDN says this makes inference faster, lets a replacement GPU pick up where the last left off, and wastes fewer GPU cycles.
The post gives the example of a software provider serving Dutch municipalities.
DDN marks the 10th anniversary of NVIDIA DGX by tracing how feeding data to GPUs shaped the DDN-NVIDIA relationship.
DDN marks the 10th anniversary of NVIDIA DGX-1, introduced in 2016, which it says began a new chapter for the company.
The problem linking the two firms is that GPUs deliver their speed only if data reaches them fast enough.
DDN says decades of solving data problems in demanding computing environments became more relevant as accelerated computing changed infrastructure needs.
The post traces how that history shaped DDN into the company it is today.
DDN and NVIDIA demonstrate Milvus with AiSAQ on DDN Infinia serving a billion-vector index at 98% recall.
DDN says it and NVIDIA demonstrated Milvus with AiSAQ on DDN Infinia at billion-vector scale while keeping production-grade recall and latency.
Infinia served a billion-vector index in place on shared storage at 98% recall, with consistent tail latency as concurrent requests increased.
Combined with NVIDIA cuVS and GPU-accelerated index builds, indexing time fell from days to hours.
DDN argues any growing RAG or similarity search system will eventually reach a billion vectors.