← Back to Library

Bubble Trouble 2

Deep Dives

Explore related topics with these Wikipedia articles, rewritten for enjoyable reading:

  • Collateralized debt obligation 12 min read

    The article draws a structural parallel between 2007-era housing-backed debt and today's GPU-collateralized financing by companies like CoreWeave. Understanding how CDOs worked, their role in the 2008 crisis, and the assumptions they relied upon illuminates the risks the author is highlighting in AI infrastructure financing.

“In the marketplace there’s all kinds of incentives right now, and rightfully so. What do you expect an independent lab that is sort of trying to raise money to do? They have to put some numbers out there such that they can actually go raise money so that they can pay their bills for compute and what have you.”Satya Nadella, CEO, Microsoft, November 2025

A few weeks ago, we harked back to 1907 to hunt down a blueprint for what is going on in markets today. The following week, we looked at 2000. This week, we turn our attention to 2008.

This isn’t something I set out to do. As much as there are clear differences between the current environment and that of 1999/2000, the differences with 2007/2008 are even more stark. Yet over the past few months, I’ve watched as a new asset class has emerged. Back in 2023, neocloud company CoreWeave raised a $2.3 billion debt facility collateralized by Nvidia AI chips. It has since gone public, raising a further $14 billion in debt and equity this year alone. Alongside it, other companies have issued debt similarly backed by AI chips (GPUs) and data center infrastructure. AI-related companies have collectively issued around $170 billion of US-dollar denominated credit so far this year according to Goldman Sachs – more than the prior three years combined and, based on their diminishing cash ratios, will continue to tap debt markets for capital.

But just as the financing model of 2007 hinged on a standardised set of assumptions around home price inflation, the current model rests on assumptions around the useful life of those chips – assumptions that are currently being challenged. CoreWeave estimates a six-year useful life for its computing equipment; other borrowers project less. The difference between six years and four years on a depreciation schedule underpinning a piece of GPU collateral is not far off the difference between +2% and zero home price inflation on a piece of housing collateral.

And just as the launch of the ABX subprime index in January 2006 put a spotlight on a previously opaque part of the market – and was in many ways the coordination point for the unwind that then unfolded – the Silicon Data H100 Rental Index, launched in May 2025, adds transparency to the compute market by tracking the hourly cost of renting a GPU. ...

Read full article on Net Interest →