Weekly Digest: The pendulum swings on AI sentiment

Last week, I used a butterfly metaphor to describe how seemingly small events in one corner of the globe can ripple across the world. This week it’s swans – broader indices and balanced portfolios are gliding along serenely, while beneath the surface there’s a violent rotation happening across and even within sectors.

Much of this has to do with developments relating to generative artificial intelligence (AI) and investors’ response. Remarkably, February’s market gyrations have not been driven by President Trump. Still, past policy initiatives are still making waves and I’m sure he won’t be dormant for long.

For much of the last decade or so, US growth stocks, with large-cap technology companies to the fore, have been leading equity markets higher. What started with the migration to cloud-based software services, provided by the likes of Microsoft and Salesforce, developed into the dominance of a handful of ‘platform’ companies, for example Amazon, Apple, Alphabet and Meta. And just as that phase of growth might have been ebbing, the launch of ChatGPT in November 2022 brought in a whole new tide. Some have characterised this as a ‘fight to the death’ to be the dominant provider of large language models (LLMs). And possibly the first to develop artificial general intelligence, which enthusiasts believe will surpass all human cognitive capabilities.

An ebbing tide of AI returns?

Initially, this tide was one deemed strong enough to float all boats. Hyperscalers (the companies behind most of the big data centres) would benefit from increased demand for the output of LLMs. The companies that provided the technology hardware, such as semiconductor chips, would see a huge increase in demand too (with Nvidia the primary winner). Those that built the infrastructure would also see a massive boost, as would companies providing energy. Downstream of this, businesses that could harness AI would become more productive, leading to higher profits and more economic growth.

This sunny scenario has been slowly clouding over for as long as a year, but uncertainty has really increased in the last few weeks. It’s evident on at least three separate fronts.

The first big worry that’s been growing since last year is the amount of capital expenditure (capex) on data centres planned by the hyperscalers: Amazon, Alphabet, Meta, Microsoft and Oracle. The current $660bn aggregate capex forecast for this group in 2026 has grown by $120bn in just the last three weeks. That’s equivalent to 78% of their annual capex as recently as 2023 (figure 1). No wonder people’s brains are exploding trying to work out the implications. (Maybe an AI agent will have the answer!)

Figure 1: Hyperscaler capex estimate ($bn)

Going back to a familiar theme, the big question is what returns this spending will generate. To match the very strong returns on capital that the hyperscalers enjoy from their existing businesses would require buyers of LLM-derived services to pay a lot more than they are paying today.  For now, all of them apart from Oracle are funding this capex with cash flow from existing operations, but only just. Some debt has been raised too, although it remains more than manageable, relative to the scale of these companies. On this level, we don’t see direct parallels to the debt-fuelled boom and bust cycle of the late 1990s to early 2000s.

Bad vibes for software services

Older readers may remember a time when anyone with a Haynes manual and a basic toolkit could’ve done most of their car maintenance at home. But most still chose to trust a professional mechanic. This is analogous to what’s happening to the next sector under the cosh – software services. These are predominantly companies that provide these services to other businesses, but also to individual consumers. The worry here is that customers will be able to knock out their own solutions by doing a bit of ‘vibe coding’ (you tell the LLM what you want to do and it writes the code for you). Alternatively, they might be able to buy other ‘off the shelf’ products for less from AI providers. That sounds plausible, but notionally free open-source software has been available for years and made minimal inroads into the mass market. Software has to be verified, integrated and regularly updated. As with the annual car service, businesses may feel safer with professionals doing this.

The leading software companies were previously expected to generate strong and growing cashflow well past a visible investment horizon (say five years). That future pot of money contributed to the company’s ‘terminal value’ which, in turn, represented a big chunk of its present value. (Present value also takes in more immediate profits and cash flow.) Software companies have been viciously derated as investors have started to worry that this terminal value could be substantially smaller, even as current earnings remain strong. With current earnings unchanged, prices relative to forecasts for the next 12 months (the forward price-to-earnings ratio, or PE) have been cut in half in some cases, or worse (see figure 2 for the sector in general).

Figure 2: Valuations drop; growth still strong for software services

Last week, the fear spread into other areas, including wealth management. This is thanks to the launch of an AI-driven product that can, it seems, look at all your income and expenditure and tax and other personal details and deliver a fully formed financial plan.

Good value, or a value trap?

This is making some companies look very cheap on the surface, especially on metrics like five-year historical average PEs. But we should be very wary of stepping into value traps. Those with longer memories will recall the demise of 
sectors such as directories (e.g. Yellow Pages) and regional newspapers, which sported cashflow yields (the ratio of cashflows to share prices) in the double digits, but ended up being worth very little after online services went mainstream.

Finally, we have what might be termed ‘second-order effects’. AI was supposed to boost productivity for everyone, but what was less accounted for was the fact that a lot of people would lose their jobs because of it. As the AI tide that was lifting all boats starts to ebb, fears have surfaced that office demand will fall, for example, hitting real estate companies. One study suggested a loss of office space equivalent to a dozen Gherkins (the iconic City of London building). Then there is the knock-on effect on services such as business travel and catering.  

AI sentiment swing brings opportunity

This all feels as though it follows the Pendulum Theory popularised by veteran investor Howard Marks. Investors got too carried away with pricing in a winning environment for everyone and now they are seeing bogeymen in almost every corner. Or to use the tide metaphor, they are envisaging lots of perfectly seaworthy boats being left high and dry. 
 

While there will undoubtedly be losers, there will also be survivors and even thrivers. The current uncertainty offers an opportunity to evaluate who will fall into each category. We may have foregone some of the upside in the past couple of years because of our reluctance to overcommit portfolios to any single theme or trend. But that leaves us in a better position to preserve wealth during more uncertain periods. And there are also previously neglected sectors that are benefiting from the rotation within equity markets. More on that next week.