We love LLMs but we're also buying "AI Losers"

28 October 2025

• Existing LLMs are great, for limited use cases. We use Quartr to maximise our stock research efficiency and provide an example in this post.


• The current market narrative seems to accept that LLMs – aided by the scaling up of infrastructure to support them - will swiftly evolve into more powerful ‘Artificial General Intelligence’ (AGI). We note compelling counterarguments from credible voices in AI, such as Gary Marcus.


• The market has pre-ordained several “AI losers”, for whom we think the threat from current LLMs is overstated and mispriced.


• We have established in a position in outsourced customer services provider, Teleperformance. It is now the cheapest stock in the European market (excluding some auto makers).


• Teleperformance has disappointed on growth post-COVID, but will not die the imminent death for which it appears priced.

LLMs (large language models) - which are enhanced pattern-recognition tools - do some wonderful things. We use Quartr as the AI-powered backbone of our stock research process. It is simply a very, very good productivity tool that has transformed our efficiency compared to the days of visiting individual company websites, punching numbers into spreadsheets, frantically typing notes on earnings calls and then trying to make sense of this information through activities such as peer group analysis. This workflow has been condensed to a series of prompts, which allows us to "turn over more stones" and have detailed interactions with more management teams. Tom (referring to himself in the third person) was an early investor in Quartr and serves in a non-executive capacity for the company, which of course explains this shameless plug for the company. But, more importantly, it is to provide evidence that we are not AI-luddites. Look at some of the advanced fundamental research we conduct using Quartr:

Quartr's LLM allows us to then dig further into the nuance:

Ryanair

Note that last bit of stunning insight uncovered by the LLM's advanced pattern recognition capabilities:

What is it about January that makes Michael O'Leary want to swear? Or is it just coincidence? What isn't coincidence is that the most profanity-liberal CEO in the stock market holds a top position in our strategy. Authenticity, integrity, shareholder alignment. We'll start to worry if he begins to sound like a McKinsey drone running a large, low growth conglomerate, talking about "go-forward strategy and value proposition". SELL.

Now, with the preface that we are LLM-lovers, we do see a growing number of concerning signs that the market has completely lost its mind over AI and its implications for society.

The market narrative requires LLMs to swiftly evolve into something much more advanced

There are credible voices arguing that LLMs are not the route to ‘Artificial General Intelligence’, in contrast to the promises made by the biggest voices in AI (no doubt to justify the eye-watering investments and increasingly creative funding arrangements). These concerns seem to be dismissed by a market intent on bifurcating companies into winners and losers. One of the convenient arguments here seems to be that because more people are questioning whether we are in an AI bubble, then we’re probably not in an AI bubble. But, if we are not going to see a swift and largely seamless transition from LLMs to more advanced forms of AI, then the market’s segmentation of companies into those with a promising AI future and those without one, is probably incorrect and contains opportunities for stockpickers to exploit these inefficiencies.

Prominent LLM-sceptic Gary Marcus and some other big brains recently gave talks at The Royal Society in London. They’re well worth a watch and contain some compelling arguments on the limitations of LLMs, that should give investors pause for thought:

The Grand AGI Delusion - by Gary Marcus - Marcus on AI

Only finance dweebs would enjoy the future AI society the market seems to have priced

For now, the music keeps playing. Stocks are going up by tens to hundreds of billions in market cap in a single day on the back of partnership announcements with OpenAI, a business which still only prints around USD 13 billion* in revenue and whose main source of cash flow is regular equity rounds. Amid the orgy of announcements, OpenAI founder Sam Altman justified it on the basis that more compute capacity was essential so that humanity (or he) doesn't have to choose between curing cancer and educating every child on the planet. This is the level of bombast the market is lapping up, all while LLMs still hallucinate, don't give the same answer twice and think there might be a seasonal pattern in the cadence of C-suite profanity, even though we're now three years into the ChatGPT era. Finance bros and an army of AI-grifters clog up the digital airwaves from LinkedIn to Substack, predicting an imminent transition to a society in which AI agents will run our lives, helpfully telling us when to eat, drink and defecate, buying our clothes, booking our holidays and buying birthday presents for family members, at no point pausing to question whether human beings would submit to such passivity (some might, to be fair). This version of the agent-augmented human would almost certainly have no friends. There is an irony in the market's scorched-earth assessment of AI's rout of the labour force and entire business models: denying humans self-actualisation through the traditional routes of wealth accumulation and professional status would presumably elevate the importance of self-expression and cultural capital, and free up time to pursue it, which seems very much at odds with being infantilised and commoditised on an AI life-glug.

We're not doomers and we're not betting the portfolio on the bubble bursting

It's important we don't become the new dot-com doomers, many of whom lost their investment jobs before they were proven correct. And there are some important distinctions between the AI buildout and the dot-com bubble relating to balance sheets; the hyperscalers actually have them. The music can keep playing for some time yet, especially if there are enough believers in the LLM-to-AGI transition. Furthermore, AI is just one driver of structurally growing electrification trends, which is our preferred exposure (through the likes of Siemens Energy, Prysmian, SSE and SPIE, which it must be said, are somewhat consensual positions for European equity funds).

But we are buying some of the pre-ordained 'losers' of the brave new world

There does, however, appear to be significant optionality in AI-loser-land. We've been doing a lot of work to trawl through the businesses which have already been written off as irrelevant in the apparently impending de-human-ification of the economy, but which could well adopt and deploy LLMs in their businesses and stay relevant and profitable. We think French-listed customer services outsourcing provider Teleperformance encapsulates the extreme narratives and valuation dislocations that are now the norm in a market dominated by short-term, momentum-driven capital. The business had a great COVID, growing the top line at 10%* and more at its peak, but has since faced a period of normalisation which forced consensus (the average of sell-side estimates) to moderate earnings expectations from its previous 'brilliant forever' trajectory. This earnings downgrade cycle has coincided with (or perhaps caused) existential angst over the role of human customer services operators in the AI era. The result: a stock that trades on about 4.5x net income, close to 25%* free cash flow yield, for a business that, granted, is no longer putting up double-digit revenue growth rates, but hasn't declined since ChatGPT was released. Its balance sheet appears sound and its bonds are BBB rated and yield 2.5-3.5%* on various maturities. Is the equity market conflating an earnings estimates downgrade cycle - part macro driven - with AI disintermediation? We think it may well be.

Perhaps one of the reasons an anti-human interpretation of AI has taken hold in markets, of which Teleperformance has been on the receiving end, relates to the tendency of finance nerds to view companies through spreadsheets and percentages rather than as representations of human actions and incentives. If you are the Chief Customer Officer of a large corporation (with a strong LinkedIn presence), and your three-year outsourced customer services contract with Teleperformance is coming up for renewal, are you really going to go full O'Leary and say “fxck it”, ripping out Teleperformance in favour of the capital intensity and project risk associated with implementing a fancy LLM-powered technology that is unproven, sometimes unstable and may be redundant in a few years anyway? (Klarna's failed experiment with AI-only customer services is an important, but seemingly ignored, case study). Or, do you let Teleperformance give you a unit price discount in return for more volume (always a lever the outsourcers pull), with a promise to progressively integrate the best AI alongside its humans and deliver you enhanced KPIs over the next three years? You get to look good to your superiors, having delivered cost savings and better performance metrics at low risk. Your backside is covered. You keep clipping that corporate coupon. With peace of mind secured, you can return to narrating your own awe-inspiring personal journey on LinkedIn…

We think if Teleperformance can draw a line under the three to four year earnings estimates downgrade cycle, the AI-dread will ease, some will start to once again suggest it’s a good company, maybe even an ‘AI winner’, and the stock will probably no longer trade as the cheapest in the entire European market, excluding the auto manufacturers (which we don't own). We may be proven spectacularly wrong in time, if LLMs rapidly advance into something far more powerful. Let’s see...

Tom O’Hara, Jamie Ross and David Barker manage European Equities strategies at GAM Investments. You can find out more information on the team and the strategies they are responsible for here.