My point is that these hyperscalers need X traditional data centers and Y AI data centers. If Y goes to 0 that doesn't mean they eat everything they've spent on those data centers. They get converted to serve non-AI loads while other older ones get abandoned or leases not renewed and so on.

In that scenario the GPUs are a loss. But really at this point inference is so profitable it's hard to imagine these data centers not being able to at least cover operations & rent.