Nebul adopts Novus for a 100 TB/s sovereign AI cloud

NetApp said on November 10, 2026 that Nebul was the first neocloud, a specialized artificial intelligence (AI) cloud, to adopt Novus, a generally available architecture targeting up to 100 terabytes per second of aggregate throughput.

Nebul is a European AI-native provider. The deployment tests whether storage can keep graphics processing units (GPUs) supplied with data while preserving customer ownership, jurisdiction and access controls. NetApp says conventional storage architectures can push GPU utilization below 30 percent when data arrives too slowly.

What does the Nebul deployment establish?\n

Nebul will use Novus in its sovereign Intelligent Cloud. The provider describes that platform as combining accelerated infrastructure, enterprise data services and production AI capabilities. It also says the cloud preserves customer ownership and control.

Beyond the Novus deployment, NetApp and Nebul are exploring ways to accelerate sovereign-AI uptake by combining Nebul’s services with the NetApp Platform.

For Nebul’s multi-tenant environment, meaning infrastructure shared by multiple customers, NetApp says Novus can scale performance, capacity and concurrency independently. All three dimensions grow beneath one namespace, a shared logical view of stored data.

Nebul founder and CEO Arnold Juffer said: “With this architecture, we can scale our AI Factory to serve growing customer demand without sacrificing the performance, security or compliance they depend on.”

How does Novus separate metadata from data?

Novus separates metadata, the information that maps file locations, from the movement of data itself. The separation lets the system add control capacity for file-location requests without using the same controllers for large data transfers.

NetApp’s technical explanation describes a client requesting a layout from Novus Data Director. The client then communicates directly with ONTAP-based storage, with ONTAP serving as the underlying data-services platform.

Data Director is an independently scalable, software-defined control layer running on standard x86 processors. The metadata plane, which manages file-location information, and the data plane, which moves stored content, can scale separately.

This separation addresses a conflict between small-file metadata operations and large checkpoint writes. Those operations can otherwise share controllers, so Novus assigns them to independently scalable layers.

• Standard access: Novus documentation lists standard in-kernel NFSv4.2 (network file system), pNFS Flex Files (parallel file access) and nconnect (multi-connection file access), plus NVIDIA GPUDirect Storage (direct transfer between GPUs and storage).
• Existing clients: NetApp says operators do not need custom client software across thousands of GPU servers. The documented protocols and GPUDirect path work with standard clients.
• Scale-out: NetApp says additional ONTAP clusters add throughput without remounts, or client reconnection, or data fragmentation. Every GPU continues to see the same namespace across that expanded infrastructure.

Why do governance controls matter for regulated AI?

NetApp says the ONTAP foundation carries the data-governance side of the announcement. The design should let organizations extend existing data estates into sovereign cloud environments. It aims to preserve consistent security, protection and governance, including current jurisdiction and access policies.

Those controls place performance and governance in the same architecture. As Nebul adds tenants and GPUs, it must keep jurisdiction, access and security policies consistent across shared infrastructure.

Willem Hendrickx, NetApp’s senior vice president and general manager for EMEA (Europe, the Middle East and Africa) and LATAM (Latin America), identified government, healthcare and biotech as regulated industries. In those sectors, providers must address compliance, security and multi-tenancy.

These requirements explain why performance scaling and data controls must evolve together in Nebul’s deployment. More tenants and GPUs increase the operational load. Regulated customers still require consistent controls across shared infrastructure.

What shipped at the September launch?

NetApp introduced Novus on September 29, 2026, making the architecture orderable that day. The initial release combined Data Director software on qualified Supermicro infrastructure with ONTAP data services on AFF A90 systems.

NetApp described a path toward software-defined deployments over time. The company said the transition would retain the same namespace, operating model and data foundation. By November 10, it described Novus as generally available.

The timing gives Nebul an early commercial test of an architecture customers could order from launch day. It also tests NetApp’s claim that a single namespace can scale across more infrastructure.

Why does storage throughput shape GPU economics?

NetApp’s economic argument is that GPUs waiting on storage still depreciate without generating a return. At the September 29 introduction, the company said utilization can fall below 30 percent under traditional architectures. That happens when storage cannot supply data quickly enough, limiting the economics of GPU clusters.

NetApp technical fellow Arindam Banerjee illustrated the bandwidth requirement with one GPU demanding as much as 2 gigabytes per second. He also described 50,000 GPUs requiring that rate simultaneously. The example tests whether shared storage can sustain fleet-wide demand.

His comparison places conventional storage arrays at 40–80 gigabytes per second. Reaching Novus’s aggregate target with arrays in that range would require more than 100 systems, according to the technical explanation.

Each additional system would add its own namespace, failure domain (a boundary for isolating failures) and management overhead. Novus instead uses one namespace across additional ONTAP clusters, which NetApp presents as a way to scale throughput without multiplying those administrative boundaries.

That comparison makes the namespace an economic control as well as a technical feature. NetApp’s case is that aggregate throughput can rise without adding an administrative boundary for every conventional storage system.

What must Nebul prove in production?\n

For Nebul, the adoption will test whether independently scaling metadata and data can support more tenants and GPUs. The test also covers whether data fragments across storage systems as the environment expands. Jurisdiction, access and governance controls must remain consistent throughout.

NetApp says Novus can independently scale performance, capacity and concurrency beneath one namespace as demand grows. Nebul’s operational challenge is to expand shared AI infrastructure without weakening the isolation and control that regulated customers require.

Novus now moves from specification to a measurable operating test. Nebul’s ability to add tenants and GPU capacity without fragmenting data or weakening regulated controls will determine whether the architecture delivers the economics NetApp claims.