AI is spending large amounts of funds for AI deployments, and AI is a bottleneck to completing AI jobs. Juniper wants to push this back to being compute bound.

Each GPU has its own 400G NIC! I cannot even fathom that amount of networking.

GPU fabric Rail-optimized Design:

https://cloudlabs.apstra.com in order to find and do the demonstrations that Juniper showed. You do not need any hardware or connections to run through the demonstrations.