NORBERT BIEDRZYCKI

The risk in AI isn't demand. It's the capital structure.

$35bn of debt. $5bn of equity. Around $2.5bn of revenue a quarter.

That isn’t the profile of a tech company. It’s the profile of a leveraged buyout, except the asset isn’t a factory but graphics processors.

That is what the balance sheet of CoreWeave, one of the largest neoclouds, looks like today. The model is simple: buy as many of the latest GPUs as possible on credit and rent their capacity to other companies by the minute.

The problem is that three risks stack up at once. Debt that has to be serviced every quarter. Hardware that loses value faster than almost any machine in industry. And contracts with hyperscalers that carry hard exit clauses if capacity isn’t delivered on time. A grid connection delayed by a year or so doesn’t simply push revenue back here. It shortens the window to repay the debt.

Then there is the demand side, which I see up close. In the companies I work with, including private equity portfolio companies, rapidly growing token consumption and AI costs have forced discipline. Every use case now goes through an ROI review. Where it pays off better, companies hire experienced developers instead of burning tokens. Demand for compute is real, but it isn’t blind to price.

Gartner estimates that by 2030, 80% of neoclouds will no longer operate independently. Through acquisitions, partnerships or bankruptcies.

The risk in AI infrastructure doesn’t sit in demand. It sits in the capital structure.

That’s why, when I analyse assets like these, I start with three questions, not with the technology. What share of the order book is conditional? When will the grid connection realistically be in place? How does the pace of hardware depreciation compare with the debt repayment schedule?

This cycle won’t be won by whoever has the most GPUs. It will be won by those who didn’t have to borrow to get them.

Read next

  1. AI won’t run out of customers. It will run out of power and time to repay its debt

    Many people read the sell-off in data center stocks as the first crack in the AI bubble. I look at the same numbers and see something else. The demand is there, and it is substantial. The trouble lies elsewhere: in the construction itself, in access

  2. The token bill is the new payroll

    For decades, people were the largest cost line in a services or technology company. Budgets were planned in headcount, margins were calculated from rates, and company value was estimated, among other things, from revenue per employee.

  3. The AI bubble is the wrong questionLinkedIn

    Microsoft has just pulled out of a deal to build a data centre in Loughton, near London. Not for lack of demand, but for lack of power: the grid connection there won't be ready until the 2030s. And it's not the only case.