Not quite... the number of deployed GPUs is only directly proportional to floor space and power footprint for a given window of time (being roughly the hardware refresh cycle). It's also not entirely clear how AI-as-a-service vs DIY will play out, but the former means the big players need to build out the DC space, not the NextDCs of this world.
(As an aside, this slide is trying to pimp some of the capabilities of Mellanox who were acquired by nvidia a while back... I would strongly question the world order they're trying to paint a picture of here...)