Every dollar spent training or running an AI model traces back to a data center somewhere, and a growing share of those data centers were built with borrowed money. One of the AI industry’s most persistent critics says that debt is being priced as if nothing can go wrong.
What he said
A data center, in this context, is a large facility packed with the specialized computer chips that train and run AI models. Building one costs billions of dollars: land, power connections, cooling systems, and the chips themselves. Ed Zitron argues, in a newsletter post on Where’s Your Ed At, that the companies borrowing to build these facilities are taking on risk that lenders haven’t priced in: “the companies building them don’t even have any experience building AI data centers, securing power, […]”
His central number: “The finance industry has fed somewhere between $100 billion and $150 billion of debt” into AI data centers outside the major cloud companies (such as Amazon, Microsoft, and Google), “all under the assumption that ‘everything will be alright.’” He describes a self-reinforcing cycle: “the more debt it raises, the more expensive the debt becomes, and the more money spent on AI capex, the more expensive that capex becomes.” Capex, short for capital expenditure, means money spent building physical infrastructure like data centers, as opposed to day-to-day operating costs.
Who he is
Zitron is the CEO of the media relations firm EZPR and host of the Better Offline podcast. He has spent the past several years writing about the technology industry, with a specific and sustained focus on what he considers unsustainable spending and hype in the generative AI boom. His newsletter examines AI infrastructure spending, supplier relationships, and the demand needed to support them.
What he gets right, and where it’s incomplete
Zitron writes that “every data center project is its own unique monster,” pointing to geography, power availability, and cooling. Those constraints change construction costs and schedules. Pricing projects as interchangeable can miss the conditions that determine whether each facility opens on time and within budget.
The aggregate debt estimate cannot identify which provider will struggle. Assess a specific supplier through its financing, power access, construction schedule, and customer demand. A large industry-wide debt total does not establish that any particular operator is insolvent.
Why it’s notable
Most AI coverage focuses on model releases and product launches, not on how the physical infrastructure underneath them gets paid for. Zitron’s post is a reminder that the AI industry’s growth has a financing layer, and that layer runs on debt held by lenders and investors who aren’t AI companies themselves. If the assumption that “everything will be alright” turns out wrong at scale, the fallout wouldn’t be limited to AI labs. It would hit whoever holds that debt.
What it means for builders
Zitron’s warning is about whether debt-backed providers can sustain the compute builders rely on, not about an imminent subscription cutoff. If you depend on a specific AI infrastructure provider, especially a newer one you haven’t vetted, examine its financing, operating commitments, and ability to meet demand. The distinction between a major cloud and a smaller operator does not establish whether either uses debt or is financially secure. The practical move isn’t to switch providers reflexively. Identify which dependencies would be hardest to replace if a supplier hit financial trouble, and plan how to export data or change providers where that matters.
Updated October 13, 2026: Zitron’s $100 billion to $150 billion estimate excludes hyperscalers; that exclusion does not mean major cloud providers avoid debt financing. His comments about inexperienced builders refer to many cases, rather than every developer.
End of article