Are neoclouds actually good businesses, or just beneficiaries of compute scarcity?

Trying to understad the economics of CoreWeave, Nebius and IREN beyond the obvious "AI demand is exploding" since Reuters recently compared the neocloud buildout to the alternative telecom companies of the late 1990s.

The internet bulls were ultimately right. Internet usage exploded, and the world needed massive amounts of fiber. But that didn't necessarily make the companies financing all that infrastructure good investments. Something similar could potentially happen with AI compute.

Neoclouds are spending enormous amounts of money securing GPUs, power and data center capacity while compute remains scarce. CoreWeave, for example, had $72B in total liabilities at the end of June, according to Reuters, up 56% since the beginning of the year.

Right now the demand is clearly there. The question I'm struggling with is what these businesses look like once supply catches up.

If GPU availability improves and compute pricing becomes more competitive, what creates the moat?