When the Fault
Reaches the Grid

You can lend against an asset built to decay

The company at the center is CoreWeave, which rents Nvidia chips to AI companies. To buy those chips, it borrows enormous sums, and the collateral is the chips themselves. That alone deserves a second look, because a high-end graphics processor loses much of its economic value within a few years as the next generation arrives. You are lending against an asset engineered to rot.

The circularity compounds it. In January 2026, Nvidia invested $2 billion directly into CoreWeave at $87.20 per share. CoreWeave, in turn, uses borrowed money to buy more Nvidia chips. The chipmaker funds the customer; the customer buys the chips; the chips become the collateral. Nvidia's chief executive has called the "circular" characterization ridiculous — his point being that these sums are small relative to the trillions the buildout ultimately requires. That is a fair caveat. It is also not a reassurance.

Michael Burry — who called the 2008 housing collapse — has spent the past several months arguing that hyperscalers depreciate this hardware over five or six years when its real economic life is closer to two or three. He estimates roughly $176 billion in understated depreciation across the sector through 2028. His framing is deliberate: not Enron, but Cisco — not fraud, but a valuation and earnings story that unwinds when the assumptions catch up to the hardware. Fellow short-seller Jim Chanos went further on CoreWeave specifically, noting that even stretching the chips to a ten-year life leaves the company barely profitable.

None of that is a BFS forecast. It is the bear case, made by named voices, and you do not have to accept the most aggressive version to see the structural point: the collateral under this debt is fragile in a way that mortgages in 2007 were not. A house does not become obsolete on a release schedule.

How a contained problem gets onto the grid

A fragile asset and an unprofitable borrower are, on their own, a private matter between a lender and a company. Investors who want no part of it can simply decline to own the stock. What turned subprime into a global event in 2008 was not the bad loans themselves. It was the packaging and the rating — the machinery that sliced the debt, stamped it safe, and routed it into money-market funds, pension portfolios, and bank balance sheets that were supposed to be boring. The bad loans were the spark. The packaging and the rating were the detonator.

That same wiring is now being built for AI infrastructure, in three moves over roughly ten weeks:

Then, on June 11, another $3.5 billion in senior notes — dollar and euro tranches — with the euro order book reportedly oversubscribed several times over. The exposure is no longer trapped inside a tech stock you can choose not to own. It is being converted into bonds and routed toward portfolios held by people who have never typed a single prompt.

The reactor and the grid

In The Financial Outage, we mapped systemic risk onto the failure modes of an aging reactor. The grid extends the same logic. A modern power grid is engineered so that a fault in one place is detected and isolated before it can propagate. The protective relays exist for exactly one reason: to keep a local problem local. When they fail to trip — or when they are calibrated on bad assumptions — the fault travels, and a contained event becomes a cascade.

The honest version is sharper than the scare

The viral version of this story says junk got stamped AAA and slipped into your pension. The record does not quite say that — and the truth is arguably more unsettling. Here is what carries which rating:

So the part that earned the investment-grade clearance is the private, Meta-anchored facility that does not trade. The part actually built to circulate on secondary markets is rated junk. That distinction matters — and it sharpens the real question rather than dissolving it.

Because the demand for the junk-rated, tradeable paper is not weak. It is ferocious. The euro book ran multiples oversubscribed. Investors are lining up to fund a buildout whose collateral decays on a schedule, whose anchor customers burn cash, and whose backlog rests on multi-year promises from companies that are themselves mostly unprofitable. The alarming part is not that someone is hiding the risk. It is that the risk is in plain sight, rated accordingly, and the money is sprinting toward it anyway.