My point was that the architecture they are using has already been done multiple times for graph databases using things like RDMA (which has existed in HPC for

My point was that the architecture they are using has already been done multiple times for graph databases using things like RDMA (which has existed in HPC for ages), that is a known quantity. It was less "it's just engineering" and more I've seen similar implementations for a long time so what makes this different? I am interested in this space in part because I spent much of my time in HPC working on graph databases.