- From edge inference to NVIDIA STX, purpose-built KV cache infrastructure for consistent performance at scale.
Graid Technology, the pioneer in GPU-accelerated NVMe storage, today announced its Agentic AI Storage Portfolio: a purpose-built family of KV cache solutions designed to eliminate the storage bottleneck that stalls “always-on” production AI. The portfolio spans three deployment tiers: KV Cache Server, KV Cache Rack, and KV Cache Platform, all built on SupremeRAID™ technology. KV Cache Platform, the portfolio’s highest tier, is purpose-aligned to NVIDIA’s STX reference architecture, with native BlueField-4 DPU execution on the roadmap for H2 2026.
As agentic AI moves from experimentation to production, the infrastructure assumptions that underpinned single-shot inference have broken down. Models running continuous multi-step tasks and maintaining context across hours of operation generate KV cache demands that overwhelm GPU HBM. The result: latency spikes up to 18x, GPU utilization as low as 50%, and model-level failures, including hallucinations and reasoning degradation, that are difficult to detect and costly to recover from.
SupremeRAID™ addresses this directly, aggregating up to 32 NVMe drives into a single 280 GB/s virtual pool, bypassing the CPU via GPU Direct Storage, and delivering KV cache reads at 1.3ms- 77x faster than standard NVMe. The three portfolio tiers bring this capability to every deployment scale:
KV Cache Server – single-node NVMe acceleration for individual inference servers and edge AI deployments. Available now.
KV Cache Rack – rack-scale, partner-validated solutions co-engineered with leading server OEM partners for enterprise multi-GPU clusters. Available now.
KV Cache Platform – Purpose-built for NVIDIA’s STX reference architecture, with native BlueField-4 DPU execution and rack-scale storage expansion on the roadmap.
“A year ago, at GTC 2025, Jensen Huang predicted that storage would become GPU-accelerated for the first time. This year, NVIDIA turned that concept into an architecture with STX and CMX,” said Leander Yu, CEO of Graid Technology. “Our KV Cache Portfolio is built for precisely this moment, delivering the storage performance that agentic AI demands, at storage-tier economics.”
For enterprises and infrastructure teams evaluating agentic AI deployments, the full deployment architecture, technical specifications, and NVIDIA STX compatibility details are available in the solution brief: Graid Technology Agentic AI Storage Portfolio: Purpose-built KV Cache Solutions for Inference at Scale.
To learn more about Graid Technology’s AI offerings, visit graidtech.com/ai.
Media Inquiries:
Andrea Eaken, Sr. Director of Marketing, Americas & EMEA
andrea.eaken@graidtech.com
About Graid Technology
Graid Technology is building the storage backbone for the future of AI, enterprise, and high-performance computing. As the creator of SupremeRAID™, the world’s first and only GPU-based RAID, and the global steward of Intel® Virtual RAID on CPU (Intel® VROC), Graid Technology delivers flexible RAID solutions that maximize NVMe performance while ensuring resilient, scalable data protection for modern data infrastructure. Headquartered in Silicon Valley with global operations and R&D in Taiwan, Graid Technology is advancing RAID innovation for the next generation of data-intensive workloads. To learn more, visit graidtech.com.
SOURCE: Graid Technology Inc.