Who Is Actually Paying for the AI Boom

In partnership with

Of the $24 billion currently financing AI infrastructure, $12 billion of it does not appear on any AI company's balance sheet.

GPU rental rates have dropped 75% in some markets over the last 18 months.

Here is how the structure works.

An AI lab needs tens of billions to build a supercomputer. Taking that debt onto its own books would make the company look insolvent. So instead, a shell company is created, called a Special Purpose Vehicle. The SPV borrows the money, buys the Nvidia GPUs, and leases them back to the AI lab. The lab pays rent. The lab carries no debt. The investors in the SPV own the hardware.

When the hardware becomes worthless, which happens in three to four years as models demand newer chips, the AI lab walks away clean. The SPV investors are left holding rapidly depreciating equipment with no recovery mechanism. The debt is secured by the physical chips. When the chips are obsolete, the security is gone.

xAI's Colossus 2 supercomputer was built this way. Twenty billion dollars total. Nvidia invested two billion in equity. The remaining $12.5 billion was financed through debt held by smaller investors and private credit funds. None of it shows on xAI's corporate balance sheet.

Last week Viktor wrote a brief, built a landing page, and opened a pull request.

Last week, Viktor wrote a campaign brief, built a landing page, opened a pull request, generated a board-ready PDF from live Stripe data, and sent a follow-up email to a churned customer. All from Slack. Same colleague that also pulls your reports and monitors your dashboards. 5,700+ teams. 3,000+ integrations.

Start free. $100 in credits →

xAI is not an exception. SPVs are described by analysts as "heavily used" across the AI infrastructure market. Meta took on $72 billion in private debt to finance its Hyperion data center in Louisiana. Oracle added $18 billion in fresh debt for AI infrastructure. The difference is those figures appear directly on corporate books. The SPV debt, across the industry, does not.

The people holding the SPV debt are being sold a specific pitch: private credit yielding 6% above inflation, backed by contracts with major tech companies. That pitch is going to pension funds. The same funds that hold retirement savings for teachers, nurses, and civil servants in cities that have never heard of a GPU.

The Bank of England issued a formal warning on this in recent months, noting that AI stocks are reaching valuations last seen at the peak of the dot-com bubble. Researchers David Kirsch and Brent Goldfarb, who spent years analyzing 58 historical technological innovations, identified four conditions that reliably precede a market crash: high narratability, extreme technological uncertainty, pure-play companies with no revenue diversification, and an influx of novice investors. They have stated publicly that the current AI infrastructure boom meets all four.

Daniel Priestley, who has been tracking the infrastructure spending cycle, puts a specific number on the risk. He predicts a significant financial correction by 2029, driven by the gap between what was spent building AI compute and what the compute actually generates in return.

Gladly Connect Live '26. May 4–6 in Atlanta.

The room you want to be in. This is where CX leaders are tackling the hard AI questions and sharing what's actually working. For CX and ecommerce leaders. Atlanta, May 4–6. Space is limited — secure your spot now.

Register now

The thing worth sitting with is not the prediction. Predictions can be wrong. The thing worth sitting with is the structure.

The companies building AI are legally insulated from the downside of their own infrastructure bets. The risk has been packaged, rated, and sold to investors who were told it was backed by the biggest names in tech. Technically that is true. The lease payments come from those companies. But the hardware risk, the depreciation risk, the obsolescence risk, does not.

Someone is holding it. Most of them don't know the full picture of what they're holding.

That is the story. Not whether AI is overhyped. Whether the financial architecture underneath it has been designed to protect the builders and expose everyone else.

Your Analytics Stack Is One Database Too Many

Pipelines, backfills, sync lag, data drift… that's the cost of splitting your stack. Tiger Cloud extends Postgres, fully managed, so analytics run on live data. No second system. Stay on Postgres. Scale on Postgres.Try Tiger Cloud free.

Try Tiger Cloud free

404 Found covers AI developments from a European Insider, three times a week. Next issue: Wednesday. April 29 2026.