An administrator is integrating an application with an endpoint and finds that the application is...

**Comprehensive and Detailed Explanation From Exact Extract of Nutanix Artificial Intelligence documents knowledge without any external URL or links:**

In Nutanix Enterprise AI (NAI), high latency during inference often indicates that the endpoint is under heavy load or that the compute resources allocated to a single instance are saturated. To resolve this and ensure lower latency for connected applications, the administrator should **increase the number of endpoint instances** (also referred to as scaling out replicas).

By increasing the instance count in the endpoint configuration, the platform distributes the incoming inference requests across multiple pods (and potentially multiple GPUs or nodes). This parallelization prevents requests from queuing up behind a single busy instance, thereby reducing the overall response time (latency) for the application.

***