Armed with the critical tools of truly stateless rack hardware and a ViRB for fleet management, we can now deliver resources with true public-cloud properties. We begin by booting directly from the cloud, with software, configuration, state, and instructions provided at runtime. Resources are then presented and consumed like a public cloud, while servers remain interchangeable: able to fail, return, upgrade, or be reassigned in real time within a federated resource pool. This model further spans acceleration, storage, and networking.

In this architecture, the hardware estate is defined by resources–not topology, fixed configurations, or “pipes” that must be tuned and maintained. Lifecycle management moves beneath the abstraction. To make this reliable, we boot stateless hardware with a guaranteed, known-good state.