The AI conundrum that every investor needs to know about

DeepSeek seems like a long time ago now. It was just over a year ago that the introduction of China’s artificial intelligence (AI) champion was being talked about as a potential ‘Sputnik Moment’ when the US feared it had lost its technological edge to a communist state¹. 

More important was the impact on the stockmarket, wiping $1trn off the tech-heavy Nasdaq index in a matter of days². Nvidia alone fell by close to 40% between 23 January and 4 April 2025³, with the ‘Liberation Day’ tariff announcements giving it another kick while it was down⁴. The view at the time was that DeepSeek, which was developed at a much lower cost than its US competitors, would undermine any new investment in AI⁵. 

We felt this was the wrong way of looking at it and we believe we have been proved right. Since then, expected AI capital expenditure for 2028 has increased from $250bn to $500bn⁶. We think it will be closer to $1trn. Since bottoming out on 4 April, Nvidia is up by more than 100%⁷.

Big tech exceeds expectations

Exceeding expectations has been a common theme since we first invested in Nvidia in 2023. This is one of the reasons why we are sceptical about claims of another Dotcom Bubble. There are many others. 

If you look at them individually, every one of the Magnificent Seven (Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla) tech companies has a set of characteristics that makes them attractive to investors: they have consistently delivered some of the highest profit growth in the market⁸. They have dominant positions in the areas in which they operate. And for all the talk of excessive valuations, they have average free cashflow yields (the money left over after all liabilities have been met, expressed as a percentage of the company’s total value) of 16%⁹. 

The third quarter of 2025 was a strong one for the tech giants, with many of them beating expectations¹⁰. Yet rather than surge on the good news, some of them actually fell, making them cheaper as they grew faster¹¹. This is not what ‘irrational exuberance’ looks like.

No ‘traditional’ bubble in AI

Yet here we would sound a note of caution. We do not think the US stockmarket has entered a bubble of the type that peaked around the turn of the millennium. Instead, we think the bubble – if there is one – will take a different form. 

The third quarter of 2025 was a strong one for the tech giants, with many of them beating expectations. Yet rather than surge on the good news, some of them actually fell, making them cheaper as they grew faster

In our view, AI presents a threat as well as an opportunity to the business models of the tech giants: whereas most of these currently operate in effective monopolies, AI means every category in which they compete is now up for grabs, whether that be e-commerce transactions, human software interactions, digital ad sales or even social media posts.

We believe greater competition will bring down prices for customers, while the cost of delivering these services will rise as the providers are forced to invest more in AI.

As a result, if there is any sort of bubble among the tech giants, we think it will be in the cashflows they generate: we expect these will eventually fall back from their elevated levels towards the stockmarket average, with valuations potentially going the same way. 

The ‘picks and shovels’ approach to AI

Once investors have taken this on board, they can begin to think about how to make money from the AI boom – because we think there is still plenty to be made. 

Inevitably, there will be winners and losers among the tech giants, but we don’t yet know who they will be. We want to retain an open mind so that when the evidence presents itself, we can act accordingly.

One thing we are confident about, however, is that the billions of dollars already allocated to AI are going to be spent, and then some. 

The hyperscalers (the providers of massive cloud computing services) are terrified of being left behind, and we think one of the main determinants of who wins and loses will be who has secured enough GPUs (graphics processing units). 

Therefore, that old cliché about the ‘picks and shovels’ approach to making money from a bubble first leads us to Nvidia – we think the world is going to need every GPU it can produce for at least the next three years.

AI’s reliance on energy 

Another potential bottleneck we have identified for these businesses is securing enough energy for their data centres. While they appear to be pinning their hopes on nuclear power, we don’t think this will be the answer for at least the next five years. In our view, it will likely be a mixture of natural gas, solar and fuel cells. 

The hyperscalers are terrified of being left behind, and one of the main determinants of who wins and loses will be who has secured enough GPUs

But the beneficiaries go all the way down to boring old regulated utilities. We are used to these growing at a fairly unexciting 5% a year, but they have now raised their forecasts to 9%¹² – a figure that is pretty much nailed on following negotiations with the utility commission.

AI’s boost for blue-collar workers 

Another consideration with the data centres is the small issue of who is actually going to build them. Somebody has to flatten the land, lay the concrete, connect the power and air conditioning and supply all the necessary equipment.  We think the supply of skilled labour could be as much of a bottleneck as the supply of energy or microchips – if you have gone to the great expense of building a data centre, you don’t want it sitting there doing nothing for a month because you haven’t connected the power. 

We recently met the chief financial officer of a company that is benefiting from the build-out of AI centres. Previously, the fortunes of his business had been closely tied to economic growth, so even in the good times a downturn was never too far away. Decades of this experience do not typically produce a chief financial officer prone to hyperbole. Therefore, we were struck by his observation that because of the expected size of the AI build-out, he did not expect a blue-collar recession in his lifetime.

There is plenty of excitement around AI and its potential to transform and disrupt even the most advanced industries. But this time around, we think taking a ‘picks and shovels’ approach to a stockmarket boom could well mean investing in the people that actually use picks and shovels.