Investment Views

The Rising AI Tide

AI can make your head spin. Not literally, although it sometimes feels as though that might be just over the horizon. It is a subject both broad and confusing. Generative AI, for instance, is not the same as AGI; GPU’s (Graphics Processing Units) have taken centre stage from CPU’s (Central Processing Units); Agentic AI (systems that do not just process inputs and generate outputs but make decisions, act and adapt) is already becoming Super-Agentic AI (which basically means getting the disparate agents you’ve created to work in tandem instead of in silos).

It is also rapidly evolving: you can now be taught how to write a book by Agatha Christie, impressive given she died in 1976. More remarkable still are the high quality scans that recently digitally unfurled previously unopenable papyrus scrolls that were carbonised during the eruption of Mount Vesuvius in 79AD. Yet the smartest AI models ever made still cannot beat the brightest teenagers at the International Mathematical Olympiad, which is exactly the sort of thing it should already dominate, given the key skills of logic, speed and a definitively right or wrong answer.

As we gaze into a crystal ball that no-one really has, various schools of thought run the full range of possible outcomes, from approaching a state of Utopia at one end to humanity finally facing the Day of Reckoning at the other. Either extreme seems to require or lead us to the Singularity, the point at which AGI (Artificial General Intelligence) becomes smarter than human intelligence, not just in narrow tasks like playing chess or spotting early signs of disease, but in the kind of general intelligence that means it can learn anything a human can, and more. Combined with recursive self-improvement, technological change becomes uncontrollable and irreversible, rapidly accelerates development and starts to make decisions far beyond our understanding or control. Much like my other half.

A far-right-tail positive outcome might mean, for example, that growth rises exponentially, the cost of physical things diminishes to close to zero and we all have more than enough of everything, leaving oodles of free time since workers have become redundant. That sounds ridiculous, until you read some of the literature*. A far-left-tail negative outcome involves something as innocuous as a “paper clip maximiser” (Nick Bostrom’s well-known analogy) unintentionally destroying humanity when tasked with making as many paper clips as possible. It seeks all resources to make them, eliminates anything in its way (including us), and converts all of earth into paper clip making material. Again, it sounds implausible, but note that the AI does not hate humanity – it is just indifferent to us in the same way we are indifferent to ants when we build a highway. Akin to a small forest fire started to burn brush, but then the wind picks up and changes direction and the fire becomes unstoppable.

Both are examples of a small, well-meaning goal that when combined with exponential power lead to irreversible damage. Try as we might, we cannot just turn it off and put the genie back in the bottle. Larger, real world decentralised systems that do a good job of resisting control include the internet, crypto and file-sharing networks such as BitTorrent. The obvious key difference is their lack of existential risk to humans, but there have been plenty of harmful use cases misaligned with humanity’s goals.

A pervasive “winner takes all” mentality, the notion that all rewards will accumulate to only a handful of successes, leaving runners-up with little if anything of value, looks to be driving parts of the stock market today, Hence the continued domination of some of the IT behemoths. Microsoft recently became the second company to pass $4 trillion in market capitalisation due to strong growth in its cloud computing division, following in the steps of GPU chipmaker Nvidia. Meta also rose as advertising margins expanded because its AI means improved ad targeting and higher ad rates.

Alphabet, Meta and Microsoft added more than $350 billion in market capitalisation in the last week of July alone. In contrast, Amazon and Apple both struggled despite good numbers, because there was less talk about their “superintelligence” story, stoking fears they may lose the AI arms race. Which serves as a reminder that at these valuations sentiment is very fragile, should growth show any signs of slowing whilst costs inexorably rise in light of huge levels of capital expenditure.

Talking of which, Microsoft will spend $120 billion on AI infrastructure such as data centres over the next 12 months, while Meta is spending north of $100 billion on, amongst other things, a data centre the size of Manhattan in Louisiana called Hyperion (named after a Greek Titan, of course. “Meta Data Centre” did not make the cut). It previously demolished a partially constructed data centre to redesign it for today’s chips, which require at least 10 times as much power as the previous generation. Mark Zuckerberg has also offered the brightest AI minds up to $100m as a sign-on bonus. That is not a typo. The next Amazon “All or Nothing” documentary series should perhaps address a topic outside of sport.

This is fine, amidst strong demand for AI computing power and a growing backlog of orders. Yet the euphoria is starting to spread outside of the incumbents with their dominant market positions. Figma, a maker of design software, surged 200% on the first day of trading, giving it a market capitalisation of more than $60 billion, three times the $20 billion that Adobe offered only three years ago. And Mira Murati, former Open AI Chief Technology Officer, has raised $2 billion to give her enterprise a valuation of $10-12 billion despite having no revenues or products. As the FT’s John Plender reminds us, scant details about what it is actually working on has echoes of the 18^th century bubble-era flotation of “a company for carrying on an undertaking of great advantage, but nobody to know what it is”.

Part of our job at Bentley Reid is to ensure that we keep two feet firmly planted on the ground. Yes, AI is here to stay and will significantly change things. On a micro level, we have already used it to carry out some of our more mundane activities such as updating processes and procedures documentation, saving up to 80% of the normal time involved. We have also trialled a Meeting Summary Agent that highlights key points and sends a summary with action items and decisions made, keeping everyone on the same page and again saving lots of time.

On a macro level, we accept it will upend economies, national security, governments and markets. Its importance is reflected by it being one of only two Overarching Megatrends that are the bedrocks of our thinking and that help guide portfolio construction (the other being the far less sexy but equally important Debt Supercycle).

It is a key part of our rationale for having dedicated exposure to particular themes such as Biotech, where AI is accelerating discoveries, reducing costs and increasing their attractiveness as acquisition targets for the multinational pharmaceutical companies; and Quality, where our active fund manager’s two biggest holdings are Meta and Microsoft.

However, we remain cognisant of the fact that right now, crazy valuations, stratospheric pay packets and eye-watering expenditure plans are serving as an AI tide to lift all AI boats. Only when that tide finally goes out will we see who has been swimming naked. So, whilst we are willing to get wet, it might be only up to our waist, with our feet on the bottom and the knowledge that we can clamber back to the beach on our own accord. Far better that, than risking being carried away literally as well as metaphorically.

* The Economist, 24^th July 2025: What if AI made the world’s economic growth explode? And The Economics of Superintelligence, same edition