The use of standard GPU grids would require a confidential computing approach for “burstable” supercomputing where ever and whenever processing is needed — but with privacy above models and details. The approach must consist of expectations for the proper use of AI, masking crucial spots