Aolani And Rafay Add Governed AI Cloud Layer To GB200 Systems

Aolani and Rafay are deploying NVIDIA DSX OS on GB200 NVL72 infrastructure, Datacenter News reported, giving the project a software control layer for governed customer access rather than a raw GPU-capacity offer.

The collaboration combines Aolani's AI infrastructure with Rafay's orchestration, automation, multi-tenancy and lifecycle-management software.

The companies framed the deployment as an early example of GB200 NVL72 systems packaged as a usable service platform.

AI infrastructure buyers need clusters that can be provisioned, shared, monitored and governed before expensive accelerated systems become useful to developers.

Without that operating layer, model training, testing and inference work can wait on custom integration after the hardware is installed.

Under the arrangement, The report said, customers can request Kubernetes capacity, virtual-machine resources, AI work areas and inference setups from a self-service interface.

The same platform adds policy enforcement, operational visibility and central governance.

That workflow changes how an AI infrastructure operator packages capacity.

Instead of selling access to a scarce server pool and leaving each customer to assemble the operating stack, the provider can expose repeatable environments with the controls needed for shared commercial use.

Aolani is a Singapore-founded AI cloud infrastructure company serving Asian markets.

Rafay sells an operations layer used by cloud operators, telecom providers, enterprises and sovereign AI operators to run distributed compute estates, giving the project both a regional infrastructure owner and a multi-tenant control-plane supplier.

Rafay's role is to simplify infrastructure bring-up and automate lifecycle management, The report said.

For operators, that software layer is where customer access, security policy, resource scheduling and developer tooling either become a managed product or remain a custom integration project.

The deployment covers the path from model-building work to training runs, inference service and later customer-facing AI delivery.

The operational advantage described by Aolani and Rafay rests on making advanced AI capacity usable by multiple customers without rebuilding the control plane for each workload.

The commercial proof remains narrower than the technical claim.

The article did not name launch customers, pricing or a deployment date, so the near-term test is whether GB200 capacity can move from installed infrastructure into paid, repeatable AI services.