Insights
By:
Are we in an AI bubble?
Are we in an AI bubble?
I don't believe we're in an AI bubble, for a few reasons.
The main one is demand, which is very strong and expected to remain so for years to come. The AI market is expected grow at a compound rate of more than 36% per annum for the next five years, reaching US$1.68tn by 2031.[1]
Another is that a bubble is typically something that gets blown up disproportionately without people noticing the risks. But everyone is talking about an AI bubble. Generally in financial markets, when everybody is talking about the risks, they’re already in the price.
Third, when a bubble bursts, there’s usually nothing much left. But AI is not going to disappear. The dotcom bubble was largely about the internet. At that point in time, it wasn’t creating much value – we just sent emails instead of letters, which was a small productivity gain. Today there is a lot of value being created in the process of AI adoption, particularly in terms of the infrastructure build-out. That alone is having a material impact on US GDP.
That’s not to say there aren’t risks. But this isn’t a bubble in the general sense of the word.
How concerned are you about high valuations?
Companies like Microsoft and Alphabet trade on relatively high valuations but the risks in those businesses are very small. I am far more uncertain about the companies, listed and unlisted, that now trade or place shares at very lofty valuations, at much higher risk.
I'd like to be a fly on the wall when the CFO of a company taking a stake in one of these businesses presents the investment to the board. How did you get to that valuation? They are taking large risks given the uncertainty of the potential returns.
Then there are the debt financing amounts, which can be enormous. If there are delays to infrastructure projects or problems connecting to electricity grids, we could see bottlenecks. These could make the highly indebted companies very vulnerable, as they may not be able to cover their interest costs. They won’t be able to take on more debt.
This is where I think there is a distinction to be made between a general bubble and the risks associated with specific businesses. Some of those higher risk businesses may fail. But the operationally stronger companies in the sector will benefit from less competition, even if sentiment in the short term is likely to suffer.
What about competition from China?
China has surplus power capacity in a way the US does not and that gives Chinese companies an advantage because they can access cheaper electricity. China also offers subsidies for using more power-intensive domestic chips. But the risk from China to US AI companies is limited, in our view, because China is developing AI capabilities for its own domestic purposes. It’s not really an export story.
What we may well see is some Chinese companies, like EV manufacturers, using AI to become even more productive and competitive. That would be bad news for US car manufacturers selling in China, but it’s not going to be something that really affects the big US tech companies.
How are you playing the AI theme in GBI?
We prefer to invest in businesses with strong balance sheets that are well diversified with different income streams. Companies like Alphabet, Amazon and Microsoft all make money from AI and non-AI activities. They have very profitable cloud businesses and very strong free cash flow. They can largely self-fund their investments so they don’t face the capital intensity risk associated with AI.
Broadcom, another one of our portfolio holdings, makes half its revenue from custom chip design, but the other half from software. We have other holdings too, like Amphenol, which have elements of AI to them, but they are not pureplay AI businesses. These types of companies give us exposure to the growth in AI without going ‘all-in’. We don’t want to be overly exposed to any one single theme.
What do you expect to happen in AI in 2026?
Investors will be looking to see whether hyperscaler capex is going to start tapering off. It has to at some stage. A year from now, I think it’s probable that spending in some areas will start to decrease, particularly as the focus is likely to move from training AI models to inferencing - where chatbots make decisions and generate results.
That might impact some businesses negatively. But it should benefit a company like Broadcom, which accounts for around 75% of the market for custom chips, which are well suited to inferencing. We feel very optimistic about its prospects for this reason. We also remain very positive about Alphabet, our biggest holding, which has shown itself to be a leader in AI, with a very strong economic and technological moat.
1 Source: Statista https://www.statista.com/outlook/tmo/artificial-intelligence/worldwide
