AI systems need data all along the spectrum of unstructured, structured, and multi-modal.  The protocols by which these diverse types of data are both acquired and delivered are as varied as the data types themselves.  At the same time data volumes and latency requirements grow ever stronger which demands solutions which scale down and up first – then out.  In other words we need maximum efficiency, we can’t resort to remote procedure calls for every operation, and we need to support hundreds if not thousands of different components or tools in the same virtual machine.