Earlier this year, ChatGPT disproved an 80-year-old maths conjecture by Hungarian mathematician Paul Erdos called the “planar unit distance problem”, which shows how large language models (LLMs) are breaking into fields once thought off limits to machines.[1]
There are two schools of thought about this. On the one hand, we have the AI “boomers” who believe the technology will be utterly transformative and will benefit a few dominant AI hyperscalers, and on the other, there are those who think the business case for LLMs is built on too much hype and that problems like “hallucinations” haven’t gone away. These “doomers” aren’t predicting doom for the job market but for the investments of AI bulls.
The market oscillates between which view holds the most sway in the minds of investors. While the boomers have now recovered, earlier this year it was the doomers that held greater sway. These prevailing views often ignore the more likely outcome: a middle-ground characterised by a steadier roll-out of AI as a commercially useful technology, with a variety of nimble companies across many sectors standing to benefit – rather than all the gains accruing to a few mega-caps.
Why the bulls might be wrong
We’ve heard some bull cases before, including that Amazon would replace all of retail or PayPal and fintech would replace banks or open-source databases would replace Oracle. All of these predictions failed to come true, and we think there are lessons to be learned from them.
History suggests that technological adoption rarely converges on a single or a few dominant platforms. Rather than all the gains going to a few AI hyperscalers, we believe there is plenty of room for major companies to invest in bespoke solutions underpinned by proprietary data, private models and open-source alternatives.
You don’t need “state of the art” for everyday tasks. A “good enough” AI model will probably prove sufficient for many business needs. Consequently, the field will be open for a variety of providers to craft AI solutions onto current workflows and products, shutting down the possibility that everyone will come to depend on just a few dominant companies.
Then there’s that powerful force called inertia. Large software companies have not historically rolled up all the smaller providers. Even in mature software categories, no single vendor has come close to consolidating the market, and we would expect AI to follow the same pattern.
One reason is that building and integrating new systems is a costly and slow process, dependent on trust between a customer and provider. With this in mind, smaller software providers won’t be pushed out overnight and will have time to adapt to the new world of AI, while benefitting from existing relationships with their customers.
There are other barriers to the AI bull case. The high energy demands of large-scale data centres, rising memory costs and a shortage of specialist talent still pose challenges to its growth. This year, Uber illustrated how easily costs can get in the way after the company blew through its entire AI budget in just the first quarter of 2026 – and is now limiting spend per user.[2]
Data privacy concerns, cash-flow limitations of AI providers and the complexity of deploying these tools across entire organisations could also slow adoption. Finally, there is also regulation to consider. The “boomer” arguments tend to rely on the assumption that the regulatory response by government (at least in the US) will be limited.
However, if the boom case does start to look real: increasing unemployment by replacing workers with AI and lowering consumer spending, then the potential rises for social and political consequences. Today, around 70% of US GDP is consumer spending.[3] If regulatory reforms follow a major labour disruption, these could further curtail AI growth and expansion.
Using the examples of failed predictions regarding Amazon and Oracle, we tend to see that established players have often been more solid than expected.
In the past, venture funds even came up with a rule: a product has to be ten times better than the existing solution to have a shot at displacing the incumbent. This points to the potential for a more measured pace of adoption for AI solutions than the boom narratives suggest.
What the market is telling us
Part of what is driving these divergent narratives is the market’s current behaviour. With the pace of AI innovation making future visibility difficult, investors are largely paying for short-term cash flow, such as memory chips and near-term earnings, rather than long-term value.
The “subscription” software (SaaS) model, which once commanded premium valuations on the assumption of unending recurring revenue, has been largely derated. The market has shifted toward momentum and near-term cash flow, a dynamic amplified by retail investors and options trading platforms.
Warren Buffett has observed that the stock market oscillates between a church and a casino, and by his own assessment, today it looks more like a casino than he has seen in recent history.[4]
When investors develop a clearer understanding of the economics of AI, the market should return to rewarding the high-quality earnings growth that has historically driven strong, durable returns.
The middle ground
AI is an undeniably transformative technology and companies that adapt to it well should stand to benefit significantly. What we question is whether it will lead to the massive market concentration or the destruction of existing incumbents that the boom narrative sometimes foretells.
Instead, we expect meaningful differences in AI adoption and integration to emerge at the company level. In our view, incumbents that can quickly find ways to leverage the new technology to improve efficiency and the customer experience are likely to pull ahead, while those that can’t may fall behind.
This means a bottom-up approach to investing, rather than betting it all on AI hyperscalers or other parts of the AI supply chain. For long-term investors, we believe the more rewarding question is not who’s building AI, but which quality growth companies are best positioned to benefit from it.
[1] The Guardian – OpenAI makes breakthrough on 80-year-old maths problem
[2] TechCrunch – Uber caps employee AI spending after blowing through budget in 4 months
[3] Data as of 31-Apr-2026. “Shares of Gross Domestic Product: Personal Consumption Expenditures.” Federal Reserve Economic Data (FRED), Federal Reserve Bank of St. Louis.
[4] Ma, Jason. “Warren Buffett says markets are like a church with a casino attached, but we’ve never had people in a more gambling mood than now”. Fortune, 02-May-2026.
