I am conducting research on feasibility of purchasing a GPU, for an organization that has a massive BI service feeding thousands of dashboards. Is it feasible to invest in GPU? Is it possible to use GPU's power in improving query performance? When AI jobs are idle is it possible to use that computation power for BI system? Any advice on resource management? 

On one hand, consider licensing terms, especially when it comes to NVIDIA, because especially consumer GPUs are not (necessarily) allowed to run business applications (see: https://www.nvidia.com/en-us/data-center/buy-grid/).
Ain addition, Kubernetes is supposed to manage workloads across multiple NVIDIA and AMD GPUs (see: https://kubernetes.io/docs/tasks/manage-gpus/scheduling-gpus/)

Investing in a GPU can be a feasible option for improving the query performance of your BI service, especially if you deal with large data sets or complex calculations. GPUs are designed to handle parallel processing, which can significantly improve the performance of certain types of data analysis and machine learning tasks.

In terms of using the GPU's power for your BI system, it may be possible to do so when AI jobs are idle. However, this would depend on the specific setup of your system, the available resources, and the software you are using. It's important to note that GPU usage can vary widely depending on the type of workload and the software being used, so you'll need to carefully evaluate your options before implementing any changes.

When it comes to resource management, it's important to carefully monitor your GPU usage and ensure that resources are allocated effectively. This may involve setting up automated monitoring and alerts, as well as regularly reviewing usage metrics to identify areas for optimization. You may also want to consider using a workload manager or other software tools to help manage and optimize resource usage across different applications and workloads.

Overall, while investing in a GPU can be a good option for improving the performance of your BI system, it's important to carefully evaluate your specific needs and resources before making any decisions. With proper planning and resource management, however, you may be able to significantly improve your BI service's performance and scalability using GPU technology.