The cloud is rented. The debt is real.

The AI boom has landlords, and some of them are borrowing heavily. So-called neoclouds rent specialized computing capacity to customers that need it quickly. That can be a useful business: buying and operating advanced hardware is expensive, and not every customer wants to own a data centre. The BBC's latest report puts the financing behind those less familiar companies under the spotlight. There is a concrete balance-sheet anchor. CoreWeave's June 30 filing lists about $35.1 billion of debt after discounts and issuance costs. That is a dated accounting measure, not a live November debt counter, and it excludes separately reported lease liabilities. Its second-quarter results also show substantial revenue and a large contracted backlog. Those are reasons to examine the business rather than dismiss the entire sector as imaginary. The tension is timing. Hardware, construction and financing require commitments before the eventual customer payments arrive. A large order book can support borrowing, but it is not the same thing as cash in the bank. Delivery obligations, customer concentration and the price of refinancing can determine whether impressive demand becomes a durable business. Our reading is narrower than another prediction that all AI will crash. Useful technology can sit on a fragile funding structure. A provider can also be acquired without failing, or grow into its obligations. The next evidence to watch is cash conversion, service delivery and financing terms, not simply whether the machines are busy.