How much should lenders charge hyperscalers?

If AI doesn’t wipe out humanity, it could make a lot of people rich. Buyers of hyperscaler stocks hope these people will include them. Of course, nobody who lends to hyperscalers thinks doing so will 10x their money. But maybe their bonds offer better rates than other companies of comparable riskiness?

A new paper on AI debt by Allianz’s chief investment officer Ludovic Subran, and his colleagues Alexander Hirt, Ziqi Ye and Maximilian Bong-Maurer suggests not, or at least, mostly not.

Here’s how hyperscaler US dollar bonds are marked versus corporate bonds of similar ratings:

So it looks like the market is pricing Oracle and SpaceX as quality junk, and maybe Meta in line with BBB credit, and the rest of them as sort-of-in-line with the average AAA-AA-rated US corporate issuer. Is this cautious enough?

On recognized debt, the market’s calm is justified . . . [but] . . . [t]he real risk lurks below the waterline: off-balance-sheet debt lifts the debt burden by nearly 150% on average and pulls the model-implied credit quality down 1-2 notches.

On recognized debt, the market’s calm is justified . . . [but] . . . [t]he real risk lurks below the waterline: off-balance-sheet debt lifts the debt burden by nearly 150% on average and pulls the model-implied credit quality down 1-2 notches.

Allianz’s view is pretty bearishly clear. But let’s look at their workings. Because their workings as to what the right price might be to carry tail risk of the AI build-out are quite interesting. They come not from boring old credit analysis, but from applying a bit of maths to signals coming from go-go equity markets.

A bit of maths (with no actual maths)…

The maths in question, bond nerds may have already guessed, is Robert Merton’s seminal 1974 paper — On the pricing of corporate debt: the risk structure of interest rates. We threatened to dive down the rabbit hole of this particular paper a while ago but never got around to it. But basically, Merton says “Hey, those corporate bonds you’ve got, they’re a lot like a bundle of risk-free Treasuries and short equity put options whose strike price is way, way, way out of the money, aren’t they?”. And they sort of are.

In the analogy, if the bond issuer survives, you (the bondholder/holder of Treasuries and short equity put options) get paid something that looks not only like a Treasury coupon and principal, but also something that looks like a stream of option premia for equity puts (the right to sell stock at a specified price) that expire out of the money and so are never exercised. 🥳

If the bond issuer goes belly up, we know bondholders don’t get paid and enter into a recovery process. What happens to the holders of the Treasury/option bundle?

With the company in question going bust, the assumption is that their stock will be worthless (even if meme-stock traders disagree). And this in turn means the option (that you have sold short) to offload worthless stock at a specified price becomes very valuable. 😬 Your counterparty forces you to pay out money up to 100 per cent of the value of your Treasury holdings in the imaginary set-up.

Just like bondholders, holders of the Treasury/short option basket are left, if not with nothing, then potentially nothing. 😔

Right, so if we can pretend that a corporate bond is a bundle of Treasuries and short put options, the question becomes how to price the puts. And the value of an equity put option depends both on the distance the price is from the exercise price and the volatility of the stock price. Translating this into credit-relevant language, Merton hypothesised how the probability of default could be inferred from the antsiness of stock jockeys.

…back to the hyperscalers

Using a Merton-inspired model, Allianz calculates the ‘distance to default’ for the hyperscalers — a measure of credit riskiness calculated using equity inputs and balance sheet data — over time. And following work that Moody’s KMV has done mapping distance-to-default to default frequencies, they map the kind of ratings associated with these outputs.

We’ve popped these in the table below, along with ratings from Moody’s and S&P:

Working solely off recognised balance sheet data, Allianz finds that the ratings implied by equity punters’ collective histrionics are higher than ratings assigned by the agencies for Alphabet, Amazon, Microsoft and Nvidia, a touch lower for Meta and a disaster for Oracle and SpaceX.

Once they whisk debtlike uncommenced leases into the equation, they find that Microsoft and Amazon drop into mid-investment-grade territory and that Meta falls into the quality end of junk.

If Allianz is right and the market should increasingly think about some of the hyperscalers as quality or weak junk, what does this actually mean for default probabilities?

Glancing at S&P Global Ratings’ historical data, a mid-double-B rating is associated with a five-year cumulative default rate of 5.75 per cent, and a mid-single-B rating a cumulative default rate of 15.6 per cent. So it’s not like most such companies go belly up.

Still, as the Allianz authors put it: “[t]ech has avoided the iceberg so far, but the navigation is getting trickier.”

Further reading:
— Are credit rating agencies getting fed up with hyperscalers? (FTAV)