One week / one topic: The cost of magic

What happened?

For most of the AI boom, the big question has been whether demand would ultimately justify the extraordinary amounts of money being spent.

As of late, it is becoming ever clearer that demand is not the immediate problem:

• OpenAI expects annualised revenue of $70bn+ by year-end, from roughly $50bn at the end of September, driven largely by its enterprise business.

• Anthropic, meanwhile, reported $16.8bn of third-quarter revenue, around twelve times the level of a year earlier, and positive free cash flow.

…and yes, there are some questions about the accounting treatment here – but the direction of travel is difficult to dispute: commercial adoption continues at extraordinary speed.

That appetite is also visible in the continued strength of compute rental prices.

Unsurprisingly, the money chasing the opportunity remains enormous.

For example, Blackstone has built a $200bn data-center portfolio – up from $55bn in June 2024 – alongside investments spanning chips, power generation and AI companies themselves.

Its experience with the acquisition of QTS Realty Trust is striking: capacity that Blackstone initially expected to support roughly a decade of growth was exhausted within just 18 months, following the launch of ChatGPT.

So far, so bullish.

And yet, over the last few days, some of the companies enabling that expansion have been hammered, with CoreWeave – a specialized AI cloud provider – as the most obvious example.

Its five-year credit-default swaps reached almost 1,200bp this week – at one point implying a greater than 60% probability of default within five years – while one of its 2031 bonds was yielding 14.5%.

Other debt associated with data-center projects has weakened too, so something apparently does not add up…

If AI demand remains this strong, technological progress this rapid and investment this enormous, why is the market suddenly worried about the companies building the infrastructure to supply it?

Yes, demand for AI might be effectively unlimited… but capital might not be, and markets are starting to ask questions.

Our observations

• Fundamentals: Underlying demand remains extremely strong and competition is intensifying at the technological frontier. (Hello, Meta!)

  But satisfying that demand requires an extraordinary amount of physical capital, with the largest technology companies heading towards $1tn of annual capital spending between them.

• Price action: Markets have begun differentiating between companies that already own scarce compute and can finance investment from very strong balance sheets, and those that require continuous access to external capital.

  The recent sell-off across leveraged AI infrastructure names is a clear warning that you can’t assume capital will always be available cheaply…

• Investor beliefs: Until recently, more AI investment was mostly treated as good news – but this is beginning to be challenged.

  Investors increasingly want to know not simply how much is being spent, but who can earn an acceptable return on that spending. The answer is unlikely to be everyone.

So what?

Perhaps investors are starting to conclude that too much capital has chased the same theme – increasingly financed through debt – and the cracks appearing in the weaker parts of the ecosystem are the beginning of something bigger.

However, less bullish on the broader ecosystem does not necessarily mean less constructive on its strongest participants.

Also, there is another possible interpretation.

What if the problem with the AI boom is not that demand is too weak, but that demand is simply becoming too expensive to satisfy?

After all – no matter the pixie dust – even AI magic ultimately faces physical constraints.

Compute requires chips > Chips require data centers > Data centers require electricity, land, cooling, transmission… and, above all, very large amounts of capital.

(Political will, too – but that’s for another day)

The market might simply be recognising that not everybody will be able to bridge that mismatch on acceptable terms. Transformative technology, powerful demand and good investment returns are ultimately three different things.

If the early AI trade rewarded almost everyone selling picks and shovels, the next phase may be less forgiving.

Advantage should increasingly accrue to companies with strong balance sheets, internally-generated cash flow, established distribution and large installed user bases.

Surprise, surprise – it might be the usual suspects once again, in the form of some of the largest US Tech companies.

Recently, while they have continued to demonstrate strong earnings, technological progress and a growing ability to monetise their investment, their valuations have de-rated substantially relative to their own recent history.

Since there are no points awarded for style in this business, I am looking at our US technology positions with an increasingly constructive bias.

Stay tuned.

Mood music: Queen – A Kind of Magic

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.