AI has moved from research lab to the centre of global capital allocation faster than almost any technology before it. Goldman Sachs Research projects worldwide AI-related investment will exceed $1 trillion in 2026, with about $581 billion of that in the US alone. At 2-5% of GDP, this peak intensity is comparable to prior general-purpose technology buildouts — railways, electrification, the internet — though compressed into a far shorter timeframe.
Artificial Intelligence: Opportunity, Risk, and a Question Bigger Than Markets
By Andrew Lawford
This article is published on: 16th September 2026
Where the opportunity lies
The clearest opportunity is infrastructure: chipmakers, power and grid companies, and data-centre suppliers behind AI’s physical build-out. A second lies in software and services companies genuinely embedding AI and showing measurable productivity gains, rather than simply attaching the label. A third is second-order: healthcare, logistics, and financial services benefiting from AI-driven efficiency without being “AI plays” themselves. For cross-border clients, diversified thematic exposure, rather than concentrated single-stock bets on US mega-caps, is usually the more sensible route in.
Where the risk lies
The most immediate risk is circularity: a meaningful share of AI capex is financed by the same handful of companies that supply and buy from one another, so returns are more sensitive than headline growth suggests to any slowdown in adoption or rise in the cost of capital.
Concentration compounds this — a small number of companies drive a disproportionate share of both AI capex and recent index gains, so even diversified portfolios may carry more AI-specific risk than clients realise. And the return on this capital is genuinely uncertain: even Goldman’s own analysis calls AI spending a “key source of uncertainty for macro markets,” without forecasting whether it will pay off.
The existential question
Clients increasingly ask directly: could AI pose a genuine risk to humanity? Honesty about uncertainty matters more here than a confident answer either way, because expert opinion is genuinely split. A recent academic survey found two coherent camps — those viewing advanced AI as a powerful but controllable tool, and those viewing sufficiently autonomous systems as a harder-to-control risk. Disagreement correlated less with general AI expertise than with familiarity with specific technical concepts, such as the tendency of highly capable, goal-directed systems to develop self-preservation sub-goals; researchers less familiar with such ideas were markedly less concerned. Still, 78% of surveyed experts agreed that technical AI researchers should be concerned about catastrophic risks — real agreement on direction, even amid disagreement on scale and timing, which most estimates place in decades rather than years.
A balanced conclusion
AI will likely remain a defining investment theme this decade, and exposure to it is reasonable. But the same technology is the subject of genuine, unresolved expert disagreement about its longer-term risks to society. The sensible approach: disciplined, diversified exposure to the theme, awareness of concentration already embedded in portfolios, and honest acknowledgment that both AI’s financial and societal trajectory remain more uncertain than its loudest advocates or critics currently claim.
Perhaps it will come as no surprise (or perhaps I’m just hoping that readers recognised the above for what it was!) when I say that everything written before this current paragraph was generated by AI (Claude in this case), based on the following prompt I had given it (please note that Claude knows what work I do and that I am based in Italy):
“I need you to write me an article that I will distribute to my clients. The topic is AI and its opportunities and risks from an investment standpoint, as well as a balanced discussion of whether or not there are existential risks to humanity over the medium to long term. Max 1000 words”
It took about 2 minutes to generate the response, which I subsequently asked it to shorten to no more than 500 words and which I have not edited at all. To me it seems like a competent, if somewhat uninspiring commentary on the current AI debate and its consequences for investors. I’m not sure that it has really given me any great insights beyond what one would glean from reading a modest amount of financial journalism and from being generally interested in the subject. In fact, I find myself wondering if these sorts of AI-generated documents “are simply the copy of one thousand summaries” (solo la copia di mille riassunti) as Samuele Bersani put it in his song Giudizi Universali.
Needless to say that the topic of AI must be taken into consideration when investing, yet we must acknowledge the limits of our understanding in terms of how all of this plays out. I think it is wise to exclude the potential for an existential crisis of humanity from our calculations, mainly because if that scenario does unfold, rather like a global nuclear war, we will have more important things to worry about than the value of our investment accounts. On a more moderate view, there will be beneficiaries and victims of AI that we cannot even begin to imagine, and we must maintain a fully diversified approach and understand how much exposure we have to certain types of investment.
One of the first things I do when evaluating a portfolio is try to get an understanding of the underlying investment exposures in order to make it clear what risks clients are exposed to: it remains a useful discussion to have before contemplating any investment changes, above all considering the concentration in a small number of companies that certain stock indices have (these points are well-made by Claude above).
Finally, should I just forget about AI in my day-to-day work?
That seems foolish, rather like the accountant I remember stories of from my youth who forbade the use of calculators in his practice because he was convinced that his employees’ brains would atrophy as a result. AI is enormously helpful in analysis and presentation, in making calculations and especially in investigating and flagging potential errors. Yet it is imperitive to understand any AI output before adding my own evaluation and conclusions, and the exacting task of finding the correct investment structures and navigating the potentially difficult topics of estate planning and tax efficiency seem, at least for the moment, to be beyond what AI can competently deal with. So, welcome to the era of the AI-enabled adviser. Believe me when I say that I am highly motivated to be one, because otherwise I may find myself facing that most uncomfortable of questions: what exactly do you need me for?