And not just the vertical integration and the supply chain control, which are important. (As the shortages of Mellanox ConnectX adapters right now is causing all kinds of server delays….) But the fact that Nvidia is a big user of supercomputing, understands all of the headaches and builds the first system for itself and runs it before putting it out for sale means it has built the skills to do this. Is HPE or Dell or Lenovo or Atos a big user of supercomputing in the same way? I am not contending that Nvidia will want to be a prime contractor. I am thinking it will be forced into it.

Back when Omni-path was the latest greatest, the switch here broke (during a vendor upgrade) leaving thousands of cores connected to each other only by means of the administrative Ethernet. Because the replacement switch was astonishingly out of stock this situation persisted for more than 6 months.

My guess is most of the small HPC clusters similarly rely on vendor service and warranties for big-ticket replacement parts rather than redundantly keeping their own on site.

Do Lenovo, Dell, HPE or even IBM have the needed level of control over their supply chains to provide the same practical delivery results as Nvidia? What about smaller companies such as Penguin?

I presume there are spares sitting in most HPC and AI centers for this contingency, but it is a good point to make that what happens when you “chaos monkey” the network. I don’t know the answer, but it sounds like a good story to chase….