VAST Data announced a new inference architecture that enables the NVIDIA Inference Context Memory Storage Platform – deployments for the era of long-lived, agentic AI. The platform is a new class of AI-native storage infrastructure for gigascale inference. Built on NVIDIA BlueField-4 DPUs and Spectrum-X Ethernet networking, it accelerates AI-native key-value (KV) cache access, enables high-speed inference context sharing across nodes, and delivers a major leap in power efficiency.
As inference evolves from single prompts into persistent, multi-turn reasoning across agents, the notion that context stays local breaks down. Performance is increasingly governed by how efficiently inference history (KV cache) can be stored, restored, reused, extended, and shared under sustained load – not simply by how fast GPUs can compute.
VAST is rebuilding the inference data path by running VAST AI Operating System (AI OS) software natively on NVIDIA BlueField-4 DPUs, embedding critical data services directly into the GPU server where inference executes, as well as in a dedicated data node architecture. This design removes classic client-server contention and eliminates unnecessary copies and hops that inflate time-to-first-token (TTFT) as concurrency rises. Combined with VAST’s parallel Disaggregated Shared-Everything (DASE) architecture, each host can access a shared, globally coherent context namespace without the coordination tax that causes bottlenecks at scale, enabling a streamlined path from GPU memory to persistent NVMe storage over RDMA fabrics.
“Inference is becoming a memory system, not a compute job. The winners won’t be the clusters with the rawest compute – they’ll be the ones that can move, share, and govern context at line rate,” said John Mao, Vice President, Global Technology Alliances at VAST Data “Continuity is the new performance frontier. If context isn’t available on demand, GPUs idle and economics collapse. With the VAST AI Operating System on NVIDIA BlueField-4, we’re turning context into shared infrastructure – fast by default, policy-driven when needed, and built to stay predictable as agentic AI scales.”
Beyond raw performance, VAST gives AI-native organisations and enterprises deploying NVIDIA AI factories a path to production-grade inference coordination with high levels of efficiency and security. As inference moves from experimentation into regulated and revenue-driving services, teams need the ability to manage context with policy, isolation, auditability, lifecycle controls, and optional protection – all while keeping KV cache fast and usable as a shared system resource. VAST delivers those AI-native data services as part of the AI OS, helping customers avoid rebuild storms, reduce idle-GPU resource waste, and improve infrastructure efficiency as context sizes and session concurrency explode.
“Context is the fuel of thinking. Just like humans that write things down to remember them, AI agents need to save their work so they can reuse what they’ve learned,” said Kevin Deierling, Senior Vice President of Networking, NVIDIA. “Multi-turn and multi-user inferencing fundamentally transform how context memory is managed at scale. VAST Data AI OS with NVIDIA BlueField-4 enables the NVIDIA Inference Context Memory Storage Platform and a coherent data plane designed for sustained throughput and predictable performance as agentic workloads scale.”
Experience VAST’s industry-leading approach to AI and data infrastructure at VAST Forward, our inaugural user conference, February 24–26, 2026 in Salt Lake City, Utah. Engage with VAST leadership, customers, and partners through deep technical sessions, hands-on labs, and certification programmes. Register here to join.
Discussion about this post