Market commentary: 1st July to 30th September 2025
AI BOOM
In June 2025 a start-up called Thinking Machines Lab raised $2bn at a $10bn valuation – the largest seed round in history. The company elected not to declare what it was working on.
[source: FT]
In 1720 a company was launched “for carrying out an undertaking of great advantage, but nobody to know what it is”.
[source: Charles Mackay’s Extraordinary Popular Delusions and the Madness of Crowds]
The latter example was possibly apocryphal but, if so, only a moderate embellishment of some of the madness that took place during the South Sea Bubble, one of the first investment “bubbles”.
This quarter we have no choice but to focus on AI. It has secured, by some measures, the most frenzied investment in history and has led already to explosive increases in profitability for some firms. Stock markets have been driven higher by this spending, both as a reaction to the profitability of the providers of the “picks and shovels” of this boom (Nvidia, Oracle etc.), but also hope that the unprecedented investment (Apple, Alphabet (Google), Meta (facebook) etc.) will lead to value creation for the end customer and therefore for the investors.
For investors, this has led to profound uncertainties. How will AI affect us? Will the effects be positive? For whom? When will it appear? How fast will it grow? Should we be scared? Can we profit from this? How do we assess the risks of investing?
We do not profess to know the answers to all of these, or indeed to many other questions. But we do have insights into how investors react to different types of uncertainty and that is the focus.
Uncertainty is often broken down into two sorts: aleatoric and epistemic uncertainty. Aleatoric uncertainty is named after the Latin for dice or game of chance. If you toss a coin a thousand times, the results do not reduce the uncertainty of the next toss. Epistemic uncertainty arises from a lack of information or knowledge. While we currently have very limited data on the impact of AI, over time the uncertainty should reduce.
As a starting position, the total AI capital expenditure (CAPEX) in the US is expected to exceed $500bn in 2026 and 2027. The Apollo program – to land humans on the moon – consumed about $300bn (inflation adjusted) in total across the 1960s and 1970s.
[souce: Citi Research]
The spending on AI is the equivalent of an entire Apollo project every ten months.
In 2025 US consumers will spend around $12bn on AI services. That’s approximately the GDP of Somalia or North Macedonia.
The disconnect between investment and revenues is gargantuan and it is without precedent. It may be the grandest of follies that historians pick over for centuries; “how could they not have suspected….” etc. It may change the world. It may be both.
Irrespective of its future, the AI spending boom has changed the present. Our analysis suggests that over half of US GDP growth is now derived from this CAPEX explosion. Some commentators have contrasted this boom with the dotcom era and have found differences that they believe supports the market reaction. Their claim is that this investment is being spent on physical assets such as silicon chips and data centres, rather than valuations rising based on mere speculation. There may be some truth to that but the ultimate value of one of these chips remains highly uncertain, so this observation doesn’t take us much further on.
Benjamin Graham once observed that markets, in the short term, are voting machines (they respond to the narratives and stories of the day) but in the long term are weighing machines; whatever the short-term valuation attached to a business, it will eventually correspond to its profitability. This will apply to the AI boom in due course. For now it is hype, eventually it will be maths; how much money will be made as a result of this investment? As things stand today, the answer to this question is unknowable. The lesson from history is that the transformation from innovations tends to be exaggerated in the short term but underestimated in the long term. The early winners in the dotcom boom did not last the race and the eventual winners mostly emerged after the boom dissipated.
Our investment position is to recognise both the transformational potential of AI, but also that much of the current investment boom will be wasted and will lead to huge overcapacity and losses. On a micro level, we look for companies that will be beneficiaries of the technology, but avoid companies that are burning through cash in the expectation of future profitability. On a macro level, the boom in investment will be driving US GDP growth for the next few years. This will be supportive of valuations and growth, but the early losers will start to emerge as the scale of the malinvestment becomes clearer.
Cautious optimism.