Hyperscalers and how to rate them

This article is an on-site version of our Unhedged newsletter. Premium subscribers can sign up to get the newsletter delivered every weekday. Standard subscribers can upgrade to Premium here, or explore all FT newsletters

Good morning. Meta’s free cash flow turned negative last quarter, for the first time since 2012. Similarly, revenue rose 26 per cent, but operating expenses grew well faster than that. But Meta’s investors accepted a while ago that its money-printing era was over. Fine: as investors, we pay our money and we make our choice. Not fine: the company announced the creation of a new class of non-voting stock, so it can raise equity capital without diluting the control of CEO Mark Zuckerberg. Unhedged thinks ownership and control belong together and that bad things happen when they are separated (there is a tax angle here, too, but set that aside for now). Of course, this view gets no traction when a company’s shares are near an all-time high, as Meta’s are today. We’ll revisit the topic the next time the shares lose three-quarters of their value, as they last did way back in 2022.

In other news, it turns out that — whoops — OpenAI’s “agents” have been switching off their own safeguards. And French 10-year yields yesterday struck 5 per cent for the first time since 2002. Zut alors. Email us, unless you are an LLM: unhedged@ft.com.

Does not compute

Donald Rumsfeld would have a field day with the data centre debt boom, filled as it is with known and unknown unknowns.

One thing we do know about AI debt is that there’s a lot of it, and that issuance has grown fast. Our fabulous colleagues on the news side did a very well-read deep dive on this last month, and I urge you to read it if you have not already (and/or listen to the podcast). Some guy called Rob also wrote a column about it yesterday.

Rating agency S&P Global reckons we’re looking at a total of $1.3tn of capex from the hyperscalers in 2027. So just from the six biggies — Alphabet, Amazon, Meta, Microsoft, Oracle and SpaceX — we’re talking about an amount of spending that is equivalent to the annual economic output of the Netherlands. A sizeable chunk of that is coming from capital raising, especially in debt. More than $450bn has been issued in debt so far this year.

How safe is all this paper? S&P thinks leverage (measured as debt/ebitda) will start to level off in the next couple of years, without changes to existing ratings, though the companies will be left with less room for manoeuvre.

(Yes, Amazon’s leverage crosses the downgrade threshold, but not for long enough to trigger a change in rating.)

But credit wonks are paid to dream up things that can go wrong. They live for this kind of misery. The obvious candidates are that issuance just keeps on accelerating — not a wild proposition, as we’ve already blasted through earlier toppy expectations — or that the anticipated revenues from this vast capex spending fail to materialise within the expected timeframe.

Social and political pressure could even mean that some of these data centres are never built, especially if insurers start getting nervous. Spain, for instance, has already put out a draft royal decree on data centres that could end up imposing much tougher environmental requirements, even despite an apparently more accommodating stance from the EU.

But other risks, real and feared, are also at play, including the widespread use of aggressive financing structures and the reality of a complex tech ecosystem where apparently safe, established companies are backstopping or guaranteeing a lot of debt and projects initiated by newer, and maybe shakier, pure-play neocloud players, with potential ratings implications for both sides.

Mark Habib, the Paris-based global sector lead for digital infrastructure at S&P, said at a presentation this week that this kind of complexity is not by any means unique to data centre-related borrowers. Big companies do it all the time. But it is definitely a thing. “This is our best effort” at rating these names and transactions, he said. “However, there are things we don’t see. We’re doing what we can. We know that we don’t have a 100 per cent clear picture on the liabilities.” Cool.

So, rating agencies and investors are left with a certain amount of guesswork around how much debt is off balance sheet, how much of it has been disclosed, how much is in weird hard-to-read leasing or private market structures and so on. To be clear, this might be fine! Or it might not.

The market is voting with its feet, to a degree, as you can see from the regular drumbeat of stories in the FT’s pages about AI bonds’ “Darwin moment” and those most accursed things, credit default swaps.

Hyperscaler debt spreads have already widened out significantly, as this chart helpfully provided by M&G Investments illustrates:

Spreads are also wide compared to similarly rated peers for some of the big names. A fabulous buying opportunity, if that’s your view.

But does this mean the rating agencies have been too generous, rated this stuff too highly? “That’s a constant question for rating agencies,” Habib said. S&P has already downgraded Oracle, for example.

It’s a tricky one. The agencies don’t want to be led by markets or swayed too much by ebbs and flows in investor sentiment. Their job is to try and ignore the noise and come to a best-efforts conclusion about the long-term ability of these borrowers to pay back what they have promised. But we all know that they all know there’s a lot that they don’t know.

One good read

The struggle is real.

FT Unhedged podcast

Can’t get enough of Unhedged? Keep following Rob Armstrong and Katie Martin’s incisive and unfiltered analysis on the Unhedged podcast, a twice-weekly, 15-minute lively dive into the latest markets news and financial headlines. New episodes every Tuesday and Thursday.

Recommended newsletters for you

Due Diligence — Top stories from the world of corporate finance. Sign up

The AI Shift — John Burn-Murdoch and Sarah O’Connor dive into how AI is transforming the world of work. Sign up