Nutanix Unified Storage is now NVIDIA-Certified at the enterprise level, confirming its ability to feed data to AI factories at speeds of 160 GB/s across 1,024 GPUs.
Back
Nutanix storage earns NVIDIA certification as AI factories scale to 1,024 GPUs
Nutanix Unified Storage achieved enterprise-level NVIDIA certification, confirming its ability to feed data to AI factories at speeds of 160 GB/s across 1,024 GPUs without I/O bottlenecks.
"This NVIDIA certification confirms that Nutanix Unified Storage delivers the full-stack interoperability, linear scalability, and reliable data velocity that modern AI workloads demand," Thomas Cornely, executive vice president of Product Management at Nutanix, said.
The certified configuration runs on a 10-node all-NVMe cluster using parallel NFS and GPUDirect Storage over NFS with RDMA, creating a direct data path between GPUs and storage. Performance scales linearly from 10 GB/s read and 5 GB/s write for 32 GPUs to 160 GB/s read and 80 GB/s write for 1,024 GPUs, using NVIDIA Spectrum-X Ethernet with Spectrum-4 switches and BlueField-3 DPUs.
The certification addresses a key bottleneck in AI factory deployment: GPU utilization. Fragmented infrastructure and data silos can leave expensive GPUs idle while waiting for data. Nutanix also plans to support NVIDIA Vera BlueField-4 STX in the second half of 2026, adding silicon-level security for agentic AI workloads.
Why Storage Became the AI Factory Bottleneck
As enterprises scale from pilot GPU deployments to production AI factories with thousands of accelerators, the data pipeline has emerged as the primary constraint. GPUs can process data faster than traditional storage can deliver it, creating I/O wait states that reduce utilization rates. The Nutanix-NVIDIA certification aims to eliminate that gap by confirming a full-stack configuration where every component — from the NVMe flash to the network fabric to the GPU — is tested for interoperability. The solution supports a range of compute platforms, including x86 systems with NVIDIA RTX 6000 PRO Blackwell and H200 NVL GPUs, NVIDIA HGX servers with B200, H200, or H100 GPUs, and NVIDIA GH200 Grace Hopper Superchip configurations.
Nutanix joins a growing roster of storage vendors racing to certify their platforms for NVIDIA-powered AI. MinIO separately announced support for NVIDIA Vera BlueField-4 STX at GTC Taipei, positioning its AIStor platform as a secure data foundation for agentic AI factories. The competition centers on which platform can deliver the lowest latency, highest throughput, and strongest security guarantees while keeping GPU utilization above 90 percent in production environments.
For investors, the certification opens a new revenue channel for Nutanix in enterprise AI infrastructure, a market where IDC projects storage spending to exceed $50 billion annually by 2028. The planned BlueField-4 STX support prepares Nutanix for the next wave of agentic AI deployments, where storage security becomes as critical as storage speed. Nutanix, with more than 30,000 customers, now has a confirmed reference architecture to compete against NetApp, Dell, and Pure Storage for AI storage contracts.
This article is for informational purposes only and does not constitute investment advice.
[1] Nutanix Unified Storage Achieves NVIDIA Certification as Enterprises Race to Build AI Factories[2] Nvidia’s RTX Spark Silicon Brings Supercomputer Ambitions to Consumer Laptops[3] NVIDIA Vera BlueField-4 STX Brings Agentic AI Storage Processing With In-Silicon Security[4] MinIO AIStor and MemKV Deliver the Secure AI Data Path with NVIDIA Vera BlueField-4 STX
