One week / one topic: Chips no more

What happened?

For the last fifteen months, owning AI-linked hardware has been an outstanding investment.

Chipmakers, memory producers, optical networks and infrastructure suppliers have all benefited from a seemingly insatiable demand for computing power.

More recently, however, cracks have emerged in that narrative.

Look closely, and in fact there have been several clues that something different might be afoot:

• Samsung reported record profits, and yet the stock sold off 10%

• Memory stocks suffered some of their worst relative performance of the year

• Hyperscalers and software stocks held up much better

What’s going on?

I don’t think investors are actually questioning AI itself or its ability to add value, but they are much more focused on where the next phase of value creation will occur.

If the AI theme is moving away from an infrastructure scarcity story and towards an application and distribution story, then AI ‘picks and shovels’ stocks should indeed be less appealing.

Just look at Nvidia – the OG of the AI trade – sitting at 10-year valuation lows.

Markets are questioning ‘What do these businesses earn three years from now?” given that, historically, bottlenecks eventually disappear.

Importantly – since the performance of open models remains within close distance of the flagship ones – markets should indeed be exploring the question of what happens if AI becomes 10x cheaper over the next 5 years.

If that’s the direction of travel then, winners are not the producers of the cheapest compute – but rather whoever owns customer relationships, data, distribution and software ecosystems.

Throw in ‘AI sovereignty’ concerns – as most companies and countries are increasingly focused on protecting their data and IP – and again platform owners look better positioned than the producers of more compute.

We have spoken before about the inevitability of grappling with the AI theme given the immense capex, equity index concentration and investor focus at play…

But what should we make of these latest developments?

Our observations

• Fundamentals: The AI investment cycle remains extraordinarily strong. If anything, capex commitments keep going up.

  As the demand story remains intact, the key debate is therefore not whether AI wins, but who wins.

• Price action: Arguably, this all increasingly resembles the behavior of a maturing investment theme, with the ‘picks and shovels’ trade now being quite crowded.

  Dispersion of returns also remains high, reflecting investor focus on picking winners as opposed to a less discerning ‘AI mania’ stance.

• Investor beliefs: For a long time, investors equated scarce compute with pricing power – and rerated ‘bottleneck stocks’ accordingly.

  Time for the 21st century version of toll operators – aka hyperscalers – to regain their spot in the sun?

So what?

For all is limitations, historical analogies can offer insight into the virtually unchanging core elements of human nature.

With that in mind, the railroad boom seems to offer a useful analogy here since, at first, it was incredibly profitable to own the scarce assets of the time: rails, steel and locomotives.

Eventually, however, supply increased and ultimately the lasting value was accrued to the businesses that controlled commerce flowing across the rails.

The AI equivalent, of course, is the following:

• Phase 1 / construction boom: GPUs, memory, networking

• Phase 2 / lasting value creation: Hyperscalers, software platforms, AI-enabled productivity tools.

The railroad boom was transformative and deeply reshaped the economy, and AI holds the threat promise potential to have an even larger impact.

Perhaps, when the technology matures, hyperscalers will eventually announce a decrease in AI capex, which could portend a shift back to returning cash to shareholders in the form of buybacks – but from an even stronger position than before the AI boom started.

In the meantime, we all remain strapped to the AI rocket – whether willingly or not.

Wish us all luck…

Mood music: Massive Attack – Angel

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.