From parlour trick to paradigm shift
Cast your mind back to late 2022. ChatGPT had just burst onto the scene and, for most people, it was little more than a curiosity – a clever party trick that could draft a passable email. The large language models (LLMs) of that era hallucinated freely, invented citations to academic papers that did not exist and struggled with the kind of basic arithmetic a ten-year-old could handle without breaking a sweat. On the visual side, AI-generated video was laughably bad: people with three legs, five-fingered hands sprouting a sixth digit, faces that melted mid-frame. Nobody was losing sleep over the competitive threat.
That was barely three years ago.
The development trajectory since then has not been linear. It has been exponential and it shows no sign of decelerating. Today’s frontier models reason through multi-step problems, write production-grade software, produce photorealistic video on demand and generate synthetic voices indistinguishable from the real thing. We are fast approaching a world in which it will be genuinely difficult to discern what is real and what is fabricated on social media. The implications of this, both societal and financial, are profound.
When the ‘safe’ career choice stops being safe
Five years ago, if you had asked what the best and brightest students should study, computer science would have sat comfortably at the top of the list. It was the degree of the future – a guaranteed ticket to a well-compensated career at a major technology firm. Today, that calculus is shifting beneath our feet. AI coding assistants are already writing, debugging, and deploying software with increasing autonomy. Entry-level programming roles, the traditional on-ramp into the industry, are being hollowed out. The irony is stark: the very discipline that built these models may be among the first to be meaningfully disrupted by them. This is not a prediction about a distant future. It is happening now, in real time, and the rate of change is accelerating.
Historical precedent and the tempo of disruption
Investors would do well to study the tempo of past technological transitions. The shift from horses to automobiles played out over roughly three decades. The printing press took generations to fully displace the scribe. Even the migration from newspapers to online classifieds unfolded over the better part of twenty years, giving incumbents time to adapt or at least time to deny the threat before capitulating.
AI is different. The replacement cycle is being compressed from decades into years and in many cases months. The infrastructure is already deployed. The marginal cost of an additional AI agent is close to zero. And unlike a printing press, which required capital, logistics and a literate population, an LLM can be accessed by anyone with a browser and an internet connection. There is no historical precedent for a general-purpose technology diffusing this quickly; even adoption of the internet was slower.
What this means for investors
The market has already begun repricing businesses that sit in the blast radius. Consider Teleperformance, once a compounder beloved by European quality investors. When it became clear that AI posed a genuine threat to the outsourced call centre model, the stock derated aggressively. It now trades on a sub-4x price-to-earnings multiple – a valuation that is either a screaming bargain or a reflection of a business whose future cash flows are rapidly evaporating.
Buying the dip in companies impaired by AI because they look “cheap” on trailing metrics is a dangerous game and likely a losing bet. Traditional discounted cash flow models assume a terminal value stretching out thirty or more years into the future. For businesses directly threatened by AI, that terminal horizon may need to be pulled forward to five years or perhaps less.
The contagion is spreading
The disruption is no longer confined to obvious targets. Just as the internet first decimated newspaper classified revenues before spreading to reshape retail, travel and financial services, the contagion from AI is broadening. Travel platforms, SaaS businesses and consulting multinationals are all beginning to feel the tremors. If an AI agent can plan a holiday, configure enterprise software or produce a strategy deck, the question becomes: what exactly are customers paying the incumbent for?
There are likely sectors being indiscriminately sold off today that will, in time, prove to be excellent opportunities. The key is separating businesses that are genuinely impaired from those that are being tarred with the same brush. Credit rating agencies, companies like S&P Global and Moody’s, were long considered among the highest-moat businesses in the market, protected by regulatory entrenchment and proprietary data sets. Whether those data moats are deep enough to withstand the AI onslaught remains an open question. If frontier models can ingest and synthesise the same underlying financial data, the value of a proprietary credit opinion may diminish faster than the market expects. But there is a credible counterargument. If an AI-first world renders LLMs not competitors to S&P Global's data sets but dependents of them, paying for the right to access, validate and redistribute that trusted information through their own products, then today's derating may prove to be less a warning sign and more a generational buying opportunity.
Physical over digital: the case for real assets
In this environment, there is a compelling case for a rerating in real, tangible assets – businesses and properties anchored in the physical world that are unlikely to be replicated or displaced by a neural network. Farmland still needs to be tended. Ports still need to move containers. Power stations still need to generate electricity. The heuristic is straightforward: physical over digital. When digital services can be replicated at near-zero marginal cost by an AI, the scarcity premium shifts decisively toward assets that cannot be virtualised.
Closing thoughts
We have never, in any prior technological epoch, witnessed change at this pace and scale. The printing press transformed civilisation over centuries. The automobile reshaped it over decades. Artificial intelligence is reshaping it over years. The world is not ready and neither are the valuation frameworks that most investors are still relying on. The playbook for the next decade will look nothing like the last. Terminal values will be truncated. Moats that appeared impregnable will be breached. Investors who recognise this shift early will be well positioned. Those who mistake a melting ice cube for a bargain will not.