THE HIDDEN LEVERAGE OF HYPERSCALERS

Whenever someone worries about how the hyperactive data centre investments of the AI hyperscalers are becoming increasingly financed by debt, optimists are quick to scoff.

They point out that these companies still have core businesses that generate a ton of cash, relatively little debt and strong credit ratings. Even if the data centres don’t produce the expected financial returns, it’s more of an issue for equity investors than lenders.

It’s a fair point. Despite the recent bond issuance splurge, hyperscalers have an average net leverage ratio of just 0.5 times, compared to 0.8 times for the technology sector as a whole, and 1.8 times for (non-financial) US companies in general, according to Morgan Stanley. Most eye-catchingly, they have more cash than debt.

However . . . 

You may have noticed that the hyperscalers have become increasingly inventive in how they raise the money required for a titanic series of data centres being built around the world.

The first example of what this looks like was the record-breaking bond sale for Meta’s “Hyperion” data centre in Louisiana, which Alphaville explored in depth here.

In short, instead of just issuing bonds off its own balance sheet, Meta formed a joint venture with Blue Owl called Beignet that would develop and own Hyperion. Meta owns just 20 per cent of Beignet, but made a rock-hard commitment to lease Hyperion for at least 20 years. That guarantee allowed Beignet to issue an amortising $27bn bond, but this debt doesn’t actually appear as debt on Meta’s balance sheet, even if it is on the hook for the payments.

Anyway, a lot of the other hyperscalers seemed to think that this was a tremendous idea, and have since explored their own increasingly creative ways of raising a lot of money while limiting the optical impact on their balance sheets. And various lease structures vaguely along the lines of Beignet are the favoured way of doing so.

But just how meaningful are these non-debt financial obligations? Fortunately, Goldman Sachs’ analysts have gone through all the fine print for us, and totted up a massive $1.5tn of lease commitments, of which about $1tn doesn’t appear in the financial statements of the hyperscalers:

Credit investors have also been increasingly focused on the lease commitments of the hyperscaler sector. Using the most recent data from the ongoing earnings season, we have tracked $1.5 trillion in aggregate lease commitments. Of this amount, $1.0 trillion is associated with leases which have not yet commenced. Because of the nuances of lease accounting under US GAAP, these ‘uncommenced’ lease commitments are not reflected in the financial statements, even though they will result in future lease payments (or other obligations) in upcoming years.

This is a hefty increase from total lease commitments of about $200bn just five years ago, and an even larger estimate than Goldman came up with just last month (when it estimated $1.2tn of leases, of which $750bn hadn’t commenced yet).

Why don’t these leases appear like normal financial liabilities, given that in many cases they are irrevocably guaranteed by the hyperscalers and therefore constitute a hard dollar commitment? To understand why, we have to (briefly!) dive into Generally Accepted Accounting Principles.

According to GAAP ASC 842, lease payment obligations should only appear on a balance sheet once the lease actually starts, not when the agreement is entered into, when a right-of-use asset also appears on the other side of the balance sheet.

Legally binding but non-commenced lease obligations are put in the footnotes — which allowed Goldman Sachs to count their overall size. But many investors may be oblivious to their extent, not least because Fitch and Moody’s also don’t count leases in their own metrics until they’re started (S&P takes a more conservative view and includes them, when it considers the leases as “debt-like in substance”).

As Goldman’s analysts noted:

From a credit perspective, this treatment can understate leverage and future liquidity needs as these obligations are eventually recognized and contractual payments come due.

Quite.

Then there are promises to buy computing power, chips, equipment and electricity. These purchase commitments do not appear as conventional debt on a balance sheet either, but are in aggregate becoming hefty financial obligations that will at some point have to be paid for.

Helpfully, Morgan Stanley’s credit analysts have looked at these promises and have counted another $982bn of purchase commitments across Alphabet, Microsoft, Amazon, Nvidia and Oracle at the end of the first quarter.

© Morgan Stanley

Of course, all this will be fine if the massive data centre investments pay off.

Hyperscaler bonds have sold off a little lately — the pioneering Beignet bond is at pixel time trading at a yield of 6.95 per cent, up from a low of 5.65 per cent when it was issued last autumn — but this is no end-of-days sell-off. Some of it is just because rising Treasury yields are pushing all yields higher.

But at some point those capex returns do kinda need to materialise, surely?

Further reading:
— A closer look at the record-smashing ‘Hyperion’ corporate bond sale (FTAV)