Nutanix unified storage achieves NVIDIA certification as enterprises race to build AI factories
Nutanix has announced the Nutanix Unified Storage (NUS) solution is NVIDIA-Certified at the enterprise level. NVIDIA-Certified Storage is designed to enable enterprises and cloud providers to confidently deploy storage solutions that support the performance, security, and scale required for large-scale production AI workloads. Nutanix is also advancing AI-native storage with planned support for NVIDIA Vera BlueField-4 STX, reinforcing its focus on faster data access, greater storage efficiency, and simpler AI operations at scale.
As enterprises and cloud providers race to build AI factories to support production AI workloads, they require infrastructure that can keep data moving, maximise GPU utilisation, and reduce deployment risk. Success depends not only on access to powerful GPUs but on the ability to feed those systems with data efficiently and reliably. Fragmented infrastructure, siloed data, and inconsistent performance can slow deployments, limit GPU efficiency, and make AI harder to scale reliably.
With this certification, Nutanix is providing enterprises and cloud providers with a validated configuration to support enterprise deployment of AI infrastructure. The certification helps ensure NUS is validated for full-stack interoperability with NVIDIA AI infrastructure, helping to reduce I/O bottlenecks and integration risk. By enabling linear scalability for the data-hungry demands of AI workloads, it helps ensure an organisation’s most valuable assets, its GPUs and data, are working at maximum efficiency in production environments.
Built on a 10-node, all‑NVMe cluster, NUS leverages enhanced parallel NFS (pNFS) and GPUDirect Storage over NFS with RDMA to establish a low-latency, high-throughput, and resilient data path directly between GPUs and storage—maximising utilisation while minimising downtime.
The result is a scalable foundation for enterprise AI that helps customers move from targeted GPU deployments to larger production environments while keeping storage performance predictable as AI workloads expand. To support large-scale AI performance, the solution uses NVIDIA Spectrum‑X Ethernet, including NVIDIA Spectrum‑4 switches and BlueField‑3 DPUs, and delivers linear scalability 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.
This resilient, zero-downtime architecture provides a flexible foundation for AI workloads, supporting training, fine-tuning, inference, and RAG pipelines across a wide range of compute platforms including x86-based systems (NVIDIA RTX 6000 PRO Blackwell, NVIDIA H200 NVL), NVIDIA HGX servers with B200, H200, or H100 GPUs, and NVIDIA GH200 Grace Hopper Superchip configurations.
