Who makes neoclouds for AI data centers

A neocloud rents GPU compute and nothing else. It does not run a diversified cloud business with storage, databases and enterprise software attached; it buys accelerators, puts them somewhere with power and cooling, and sells time on them. That focus is the whole model, and it is why these companies appear in a catalog of physical parts — they are among the largest buyers of everything in the layers below. The financing structure is the part worth understanding, because it is unusual and it shapes behaviour. Much of this capacity is funded against the hardware and the contracts attached to it rather than against a corporate balance sheet, which means an accelerator fleet is collateral and a signed customer commitment is what makes the debt work. That makes these operators fast — they can commit to capacity a conventional balance sheet would not support — and it ties them tightly to the residual value of a depreciating asset and to the creditworthiness of a small number of customers. The group is also less homogeneous than the label implies: it includes companies that build and own facilities, companies that lease wholesale capacity, chip designers selling access to their own silicon, and former crypto miners who already held power contracts and converted. What to look for is which of those a given company is, whether it owns or leases its buildings, how concentrated its revenue is, and where its power actually comes from.