We are witnessing an unprecedented “mortgage crisis” in AI, and OpenAI’s preemptive request for a bailout is a clear sign of the top.

by

They turned mortgages into time bombs in 2008 and now they are doing the same trick with AI, just with bigger numbers and darker corners. When the funding chain snaps this time, it will not just be a few banks choking, it will be the whole credit system gasping at once. Most AI companies are running on borrowed money, using loans to fund GPUs, data centers, and cloud leases instead of real profits. With AI capex now in the trillions and exceeding 1% of GDP, if those firms fail to turn profitable, the fallout could rival or even surpass the global financial crisis as those debts turn into bad loans across the banking system. OpenAI asking for a preemptive bailout is the clearest tell that the stress is no longer contained. That’s the moment you know the cracks are starting to spread across the whole system.

This next debt crisis will be NDFI + AI. Of course AI is a subset of the NDFI bubble as well as affecting EVERY other debt market.

That’s what JP Morgan wrote on Monday that AI will tap out every credit market on the planet:
https://finance.yahoo.com/news/ai-5-trillion-data-center-165808446.html
“Others on Wall Street have expressed concern about the complex private debt instruments hyperscalers are using to keep AI funding off their balance sheets.”

So far, credit spreads are tight indicating low concern in the bond market, but that can change overnight:

Here is the where the Fed historically has stepped in to provide liquidity.

NOTE: There has been a debt crisis every four to five years for the past two decades:

Parallels between the 2008 Credit Crisis and Today’s AI Valuations

Structural Echoes

1. Off-balance-sheet leverage.
2006–08: banks warehoused mortgage risk in SIVs.
2025: hyperscalers are funding AI data centres through SPVs and JVs that sit off the parent balance sheet.
Meta’s $27 bn joint venture with Blue Owl is the current archetype.

2. Risk transfer engineering.
Pre-crisis: CDOs and CDSs transferred exposure to investors.
Now: banks like Deutsche Bank are structuring short AI equity baskets and synthetic-risk transfers to hedge the very data-centre loans they originate. Classic “originate-and-distribute.”

3. Insiders hedging their own theme.
When the lenders hedge the asset they’re promoting, it’s late cycle. Deutsche’s internal discussions of short AI trades are that tell.

4. Celebrity shorts as sentiment markers.
Michael Burry, of The Big Short, has disclosed large put positions on Nvidia and Palantir, publicly warning of AI-bubble dynamics.

5. Credit flood feeding the build-out.
As mortgage securitisation once did, AI infrastructure is now being fed by a deluge of investment-grade and private-credit issuance. Meta’s off-BS bond programme drew record demand.

6. Circular financing and concentration risk.
2008’s feedback loop: originators → securitisers → dealers → investors.
Today: hyperscaler → GPU vendor → SPV → private credit → back to hyperscaler. A tight, reflexive loop with leverage buried in the middle.

7. Securitisation footprints re-emerging.
Discussions of data-centre ABS and “AAA-rated” structures mirror the plumbing phase before the mortgage bubble peaked.

8. Regulatory and disclosure opacity.
GFC: hidden leverage via conduits.
Now: “AI-washing” fines and SEC scrutiny show that reported metrics and financing footprints are already stretching transparency rules.

9. Macro mismatch risk.
AI infrastructure is a long-duration, capex-heavy bet financed in a short-confidence cycle. If utilisation or pricing lags, the unwind dynamic will resemble housing credit compression.

Annnd. 10. Now AI companies want Fed backstops!

Meta is hiding $30 billion in AI infrastructure debt off its balance sheet using special purpose vehicles, echoing the financial engineering that triggered Enron’s collapse and the 2008 mortgage crisis. Morgan Stanley estimates tech firms will need $800 billion from private credit in off-balance-sheet deals by 2028. UBS notes AI debt building at $100 billion per quarter “raises eyebrows for anyone that has seen credit cycles.”

The Structure
Off-balance-sheet debt through SPVs or joint ventures is becoming the standard for AI data center deals. Morgan Stanley structured Meta’s $30 billion in an SPV tied to Blue Owl Capital, making it easier to raise another $30 billion in corporate bonds. Musk’s xAI is pursuing a $20 billion SPV deal where its only exposure is paying rent on Nvidia chips via a 5-year lease. Google backstops crypto miners’ data center debt, recording them as credit derivatives.

My Take
This is 2008-style financial engineering repackaged for AI. The key difference from the dot-com era is growth was financed with equity then. Now there’s rapid capex growth driven by debt kept off balance sheet. When chips estimated to last five to six years may be obsolete in three, and companies structure deals where their only exposure is short-term leases, that’s hidden leverage creating the opacity that preceded past crises. Meta keeping $30 billion off its balance sheet while UBS warns about $100 billion quarterly AI debt buildup shows the pattern I’ve been documenting where leverage accumulates outside traditional visibility.