A library consortium DC plans leaf upgrades in place while inference SLAs continue. Which network attribute enables that operational growth?

A library consortium DC that plans leaf upgrades in place while inference SLAs continue needs a scalable, redundant design that supports rolling leaf upgrades without breaking inference SLAs. Redundant paths and scalable leaf roles let operators refresh hardware one device at a time while history chatbots and other inference services keep answering patrons. Lifecycle change without a fabric-wide outage is the operational growth property under evaluation. Increased spine uplink bandwidth that sustains rolling leaf upgrades without breaking inference SLAs adds bandwidth, but it does not provide redundant leaf paths or rolling upgrade capability, so it cannot keep inference SLAs intact during hardware swaps. Fine-grained QoS policies that preserve rolling leaf upgrades without breaking inference SLAs help manage traffic, but they do not remove the single point of failure or allow a leaf to be replaced without affecting inference. Streaming telemetry analytics that guide rolling leaf upgrades without breaking inference SLAs offer observability, but they do not provide the scalable, redundant design required to replace leaves while inference continues. Exam caveat: Operational growth and maintainability—rolling upgrades under SLA—are part of AI network evaluation, not only day-one bandwidth. Operational check: document a rolling leaf upgrade procedure that drains traffic, swaps hardware, and restores capacity while inference SLAs stay green, and test that procedure on a noncritical leaf before touching production library AI paths.