Platform architecture Nebius AI
Nebius AI is natively designed to provide a single pool of unified hardware resources. Nebius AI has no dedicated physical management servers, and many of its services are deployed on virtual machines similar to end-user VMs.
This approach ensures easy scalability of the platform: at minimal cost, you can automatically add infrastructure capacity ranging from a single server to an entire data center running thousands of machines.
Nebius AI is built around the storage layer that stores data and system information for all services. The storage layer uses a fault-tolerant Distributed SQL DBMS, which provides secure and efficient data storage, manages heavy data loads, and ensures horizontal scaling as Nebius AI expands. The dedicated replication mechanism of the DBMS covers all hardware storage capacities used by Nebius AI.
Compute Cloud is built on top of the storage. It allows you to manage virtual machines used both by external users and internal platform components. KVM functions as the hypervisor, with QEMU as the emulator. VirtIO drivers are used for device virtualization. Scheduler is an important component to enable hardware infrastructure allocation for your VMs. This is what determines which physical server to deploy a VM on.
This layer also includes other Infrastructure services.
The next level includes the platform services (Platform as a Service, PaaS). These include Data Storage & Analytics and other.
By using Identity and Access Management (IAM) and Resource Manager, you can enable role-based access control for your resources: each user gets a role that permits certain operations. For example, the editor role allows the user to create, delete, and edit resources, but disables resource access control.