One week / one topic: Bills come due

What happened?

Last week, we focused on who would finance the AI build-out.

This week, results season asked a harder question: who will actually make money from it?

Three hyperscalers gave very different answers:

• Microsoft beat expectations, especially on cloud revenue growth. Its shares rose about 3% after hours, helped by a smaller-than-expected rise in capex, but then gave back most gains.

• Alphabet had its worst day ever on Thursday (-13%) as search advertising missed forecasts and grew at its slowest pace in five quarters. Free cash flow was negative for a second quarter running, which has never happened since the company went public.

• Amazon fell about 7.5% before Friday’s open after capex topped $77bn and online sales grew just 2%. Over the past months, its free cash outflow has reached $42bn over the past 12 months.

Almost overnight, it feels, the market has stopped paying for capex as a sign of ambition.

Now it wants a return on it, and soon.

Yes, most people now recognize that AI is (1) transformative and (2) here to stay. Can you imagine a scenario where we use less AI in ten years' time?

But even if demand keeps growing strongly for years, the build-out of physical infrastructure is on a historic scale. And the less of it a company can fund from its own cash flow, the more it depends on other people's willingness to lend.

The sequence is clear. Retained earnings no longer cover the build-out, so funding has moved to bond markets, then to private credit and asset-backed structures. Now hyperscalers are issuing equity as well.

It's no surprise, then, that lenders are charging more for the risk. At least 18 data-center financings in junk bonds and leveraged loans now carry double-digit yields.

This is not just a question of pricing the risk, that is, working out the appropriate yield for these securities.

The deeper question is whether there is enough capacity to absorb all this debt.

Estimates vary, but the scale is simply breathtaking.

There are already signs of strain…

Some banks say they are approaching their lending limits for the sector, and some insurers have reached theirs on data-centre exposure. Several AI infrastructure IPOs have also stumbled.

To be clear, this is not a credit crunch: hyperscalers are still raising tens of billions.

But after a period of eerily unflappable calm in the face of meaningful shocks (remember Iran?), stress that was confined to a few names is now spreading across credit markets.

Could this be the start of something more systemic? And if so, what should we do?

Our observations

• Fundamentals: Demand isn't the problem, then… it’s simply (?) the availability of enough cash.

  If we really need to find ~ $2trn from IG bonds, then the risk now reaches the core of most bond portfolios, not just high yield.

• Price action: Investors don’t seem to have lost their heads and have remained discerning, somewhat frustratingly.

  With the whole complex under such constant scrutiny, is it even realistic to hope for overdone reactions?

• Investor beliefs: With the debate now about financing rather than demand, AI borrowers increasingly compete for the same pool of capital as everyone else.

  ‘Crowding out’ is real, but that may be where opportunities arise.

So what?

AI is a bit like a 3D object: look at its outline from different angles and you see different shapes…

To technologists, it is a revolution. To equity investors, it is an earnings story. To bondholders, it is a $5.5trn funding gap.

But no. It's more like a 10-dimensional object.

Demand, capex, cash flow, funding costs, credit capacity, valuations, competition, energy, policy and even employment keep changing at the same time. Each one shifts the others.

That is why the debate feels so slippery. We seemingly settle one question only for another one to immediately open up in its stead.

But however you turn it, we keep coming back to the same question: who can turn all this spending into durable free cash flow and returns above the cost of capital?

This week, the market remained relatively unperturbed about the company that could show a return (Microsoft), and punished the ones that couldn't (Alphabet and Amazon).

As the bill moves from shareholders to bondholders, I suspect that test will only get stricter.

For us, that means favoring companies that can fund their ambitions from their own cash flow, thinking hard about where to sit in the capital structure, and watching whether crowding out spreads to non-AI borrowers.

Sooner or later, bills come due – and we are increasingly asked to figure out who can pay them, and who cannot…

Mood music: Keziah Jones – Sinnerman

By popular demand, here is the One week / One topic playlist

The information provided should not be considered a recommendation to purchase or sell any particular security.