Nutanix and Supermicro to Offer Joint Private AI Inference System in Early 2027

The rack-scale system pairs Supermicro servers carrying eight Nvidia Blackwell Ultra GPUs per node with Nutanix's agentic AI software, and Nutanix is pitching it to Middle East and African firms that want to control their inference costs.

By Nakoda News Desk

1 day ago·2 min read

Prefer Nakoda News on Google

Nutanix and Supermicro are building a jointly engineered AI infrastructure system for companies that want to run AI inference in their own data centres rather than through public cloud services, according to an announcement reported by Intelligent CIO Middle East. The companies said it should be available in early 2027 through Supermicro and its channel partners.

The system combines the Nutanix Cloud Platform with Supermicro servers, Nvidia GPUs and Nvidia Spectrum-X networking. It runs on Nutanix Agentic AI, the company's software stack for provisioning infrastructure, running models and governing AI agents. The stack also includes the Nutanix Kubernetes Platform and Nutanix Enterprise AI, which delivers models as a service through a secure Agent Gateway, plus an integrated catalogue of common AI applications.

What Is in the Rack

Each Supermicro compute node will hold eight Nvidia Blackwell Ultra GPUs. A separate Nutanix management cluster based on the NX-8170-G10 provides the control plane, so a management failure does not interrupt GPU workloads. A Nutanix Unified Storage cluster built on the NX-8150-G10 handles data ingestion, checkpointing, context caching and model serving. Networking uses Nvidia Spectrum-4 switches, ConnectX-7 and ConnectX-8 SuperNICs, and BlueField-3 data processing units.

The companies said customers will be able to add GPU capacity as demand grows while keeping control of infrastructure management and running costs.

Supermicro chief revenue officer Matthew Thauberger said the pre-validated, rack-scale design was meant to give organisations "predictable access to high-density GPU compute" and to avoid the usual lead-time delays. Thomas Cornely, executive vice president of product management at Nutanix, said infrastructure complexity and hardware procurement often slow companies trying to move AI from experiments into production. He said the system would let them "control the cost per token as AI adoption scales."

Regional Pitch

Nutanix is presenting the system as a fit for organisations in the Middle East and Africa that are increasing AI spending and want more control over inference costs and data. Sush Kajaria, EMEA alliance sales director at Nutanix, said inference costs, access to GPUs and the work of integrating several technologies can hold back AI projects in the region.

The announcement reflects a wider change in enterprise AI spending. As companies move from training and pilots to running models every day, the cost of each inference becomes the main concern. Matt Kimball, vice president and principal analyst at Moor Insights & Strategy, said hardware, memory, networking, model optimisation and orchestration must be managed together to scale inference efficiently.

For Gulf buyers, the private, on-premises approach also fits data residency and sovereignty requirements that discourage sending sensitive workloads to overseas clouds. The offering will compete with integrated AI systems from other server makers and with the GPU capacity hyperscalers are building in the region. Pricing was not disclosed.

The partnership comes as Supermicro's business grows quickly. The company reported this week that quarterly revenue had nearly quadrupled to $19.9 billion on demand for AI servers.