Oct 2026: AI capital cycle roadmap (Part 2)

Download PDF

AI Capital Cycle Roadmap (Part 2)

October 2026

Most investors love what is working today and that led some down an alley of momentum and chasing operational gearing. AI offers amazing productivity advancements, but a short term bubble has been created by the capital chasing it. In contrast, we love companies that invest and innovate for their customers long term benefit. The faster their innovation, the greater today’s investment, resulting in lower near-term profits and share prices. Today Mr Market doesn’t favour our companies and that is a great opportunity.

Revisiting our roadmap

Please take a look at Part 1, in it we tried to put the current AI capital cycle into some historical context. Crucial was opening readers minds to the distinction between the Installation period of a capital cycle and the Deployment period. The latter sees the long term benefits of AI delivered in decades. The former is about the near term capital that chases the idea, creating an investment boom and bust.

We also observed that during the 1998-2002 period, views on both sides (bulls and bears) were far too ingrained. Open mindedness and balance were nowhere to be seen. This is a 2026-2028 investor’s greatest risk, i.e. to be an uber-bull or a perma-bear on all things AI related.

Munger and Risk

We are trying to give ourselves an investing roadmap for the next few years. Observing and navigating cycles is different to calling them. I am minded to behave as Munger did towards Musk c.10 years ago. He neither backed Elon nor bet against him. Instead, just observing the scale of risk Elon was taking on in so many near-death experiences in his businesses. Munger just decided he did not need to play. That is how I feel about parts of the AI investment landscape today (SpaceX, Neoclouds etc.). However, a good understanding of how the environment could evolve may present us with future opportunities.

AI is amazing… and a bubble

“The test of a first-rate intelligence is the ability to hold two opposed ideas in the mind at the same time and still retain the ability to function.” F. Scott Fitzgerald

To have an edge in the next few years, investors need to be able to think as per Fitzgerald:

• We should be bullish on the long term potential that AI offers to society and businesses, but still carefully watch the capital cycle unfold.
• We can admire the functionality that AI delivers but be cautious about whether many of those developing it will see returns sustainable enough for us to invest at current prices.

This suggests a wait and watch tactic rather than a need to jump on the bandwagon.

AI solves long standing economic malaise

I remain excited by the prospect of what AI can deliver to the world. Where some see a threat to security or employment, I see a reason to celebrate. This is because I think AI stands a high chance of solving an ingrained global economic problem: the lack of productivity. Productivity, simply expressed, is output per hour worked. When productivity rises, standards of living rise too. This is good for government incomes, those who rely on them, corporate profits and the stock market.

An observation on AI’s potential productivity was made by Warren and Charlie in their last Berkshire meeting together. They rightly observed that technological changes bringing about productivity boosts can take 20-30 years to fully roll out. As such, small incremental changes in total output and employment occur each year. This, and the new roles every new technology creates, are points missed by the AI fearmongers. As such, I don’t subscribe to the mass unemployment idea. Today there is fear of the AI overlord. This is normal for new technologies… humans feel threatened by rapid change. With all new technologies, the fear comes early and the regulation comes late.

“The likelihood of change taking place depends on the outcome of the struggle between novelty, thirsting for a chance to take its place, and the old, fearful of any threat to the status quo.” Joel Mokyr 1997 (see below)

The contradiction in the AI investment case

This productivity boost highlights a contradiction in the blanket bull case for AI investment. One of the reasons why companies and individuals like AI so much is the time savings/usefulness it has vs. its cost. If its cost were to rise significantly (so that all model/datacentre builders can make an acceptable rate of return on their investment), then that trade off becomes far less interesting. This is not questioning the value AI creates; it is where that value accrues. Email and Excel saved us all huge amounts of time, but there was still only a limited price they could be sold for, even by a quasi-monopoly provider.

Large language models (LLMs) are improving at a very fast rate and competing models are leapfrogging each other in accuracy, functionality and reduced cost. They also all learn from each other. This does not suggest good medium term ROIC’s or high terminal values for many LLM standalone companies.

One company we know well at the forefront of AI deployment has been Block. Their inhouse system (Goose) utilises all models. This is what they said last month:

“We don’t need leading-edge models to do the vast majority of AI work we undertake today”

That said, at Holland we like Claude and are unlikely to turn away from it to save a few pounds. It is our star employee. We currently spend c.£40 a month on it, and it might be saving us £10k a year. Both these figures might double in the next 12 months and then I suspect will plateau.

Why this time could be different

In 1999 a new breed of digital companies emerged to challenge analogue incumbents. Today’s incumbents are very high-quality businesses. All of which have long term outlooks, a commitment to innovation and customer centricity (e.g. Meta/Google/Amazon/TSMC). The high level of conviction such companies have in AI’s long runway of profitable growth is notable.

Whilst huge capital is being spent to scale AI, the wiring needed for its initial rollout (fibre, 5G, cloud) was already in place, thus helping its pace of adoption. As such todays incumbents have many advantages not disadvantages. The growth impact of AI to their core operations, and their ability to flex investments (up or down!), make ROIC’s discussions for such companies more nuanced than for new entrants.

What we look for – Long term compounding

We invest in businesses that can compound our capital at a good rate by having powerful Sustainable Competitive Advantages. These come in different guises, but all have similar traits:

1. They are long term in focus;
2. They offer customers great value for money;
3. They innovate heavily and invest for the long term;
4. They are almost always run by an aligned founder for whom this is their life’s work.

Hamilton Helmer’s 7 Powers framework has stood the test of every market regime we’ve lived through. Whether it’s an Installation or Deployment period, Frenzy or Crash. Power dynamics don’t switch off just because capital is euphoric. But perhaps the Frenzy phase is when long term SCA’s gets harder to distinguish from the temporary sugar rush of operational gearing. This is maybe where we are today.

What we are searching for is not the next hot theme, but the next generational compounder. A business run by an owner manager who will make the correct capital allocation decisions for the long term, even when those decisions look wrong, unpopular, or margin dilutive today. Someone building a moat, not chasing a moment.

These companies will not always be comfortable to own. They may be misunderstood by Mr Market for long stretches, as Amazon and Google once were. Also, they are businesses that navigate themselves through a capital cycle, rather than simply being carried along by one.

Momentum investors’ unseen addiction

The reality of analysing businesses is complex and very few people are interested in the effort required to do so. Instead by investing in what is working today they take a shortcut that seems to work… for a while!

Most investors look to invest with portfolio managers who are performing well today. In turn, those PMs want to be investing in companies that are also performing well today. These biases are perfectly reasonable but expose investors to peak cycle risks (as they did in 2007).

Q: What makes a share price rise fast in a short period of time? A: Operational gearing in a commodity business. Simply put, if you refine oil and make $1 of profit on every barrel you sell at $50, you make a 2% margin. If the price then leaps by $2, your profit and margins treble. This is what has been happening to memory chip businesses like Micron of late. Their profits and margins have leapt, and share prices boomed.

Meanwhile what are our companies doing? Powerful network businesses with wonderful customer offerings, like Mercado Libre, Block or Meta, are investing heavily in future AI products they think their customers will love.

So, at the same time AI commodity businesses are booming, some of the world’s best businesses profits are slowing. Seen through this lens recent year stock price movements are understandable.

Owner managers aggressively invest in pivots

The DNA of our companies means they are good at pivoting, bringing new products and services to their customers quickly. They are wired and happy to invest aggressively today for a stronger market position tomorrow, embracing Charlie Munger’s ethos of deferred gratification. Let’s lay today’s AI opportunity over this.

Each one of these companies: Amazon, Meta, Block, Wise, Nu, SEA, Meli, sees the wonderful things that AI can do to develop their offering. So, which of the following strategic routes do they choose today?

1. Invest a little, see what competitors do, and review in three years….?
2. Invest heavily today, depress margins, and strengthen their market position further.

In every single one of these companies, they are choosing option b – seeking to widen their moat. Crucially, as mostly lower margin businesses, investing today disproportionately hurts their short term profitability and that makes them unpopular with Mr Market.

The hated investment phase

I have been looking at share prices for almost 40 years and one of the greatest constants I see is that markets always dislike investment phases. This is because uncertainty is rising, risk is seen to be rising, and earnings forecasts are falling.

That is what makes the Frenzy phase of a capital upcycle so unusual. All such normal market concerns about an uncertain period of investment are just brushed aside…for a while. During such phases the promoters find new metrics to use. So, they point to mobile subscriber growth or website hits (1999), or recently token usage.

Where are we in the cycle

In Part 1, we highlighted the Installation period and the Deployment period of the capital cycle, noting that between the two often comes a recession or crash.

Spring/Summer 2026 felt like the second half of the Installation period, known as the Frenzy. This is when capital aggressively chases the new opportunity. What we can’t know however, is how long this period will last. Notably, in recent weeks markets are now starting to ask more ROIC type questions. They are also starting to question how some future projects will get funded. The cycle is ticking on.

“The turning point from Installation to Deployment is a crucial crossroads, usually a serious recession, involving a recomposition of the whole system…” Source: Technological Revolutions and Financial Capital, Carlota Perez

2026 is a capital cycle at work in real time. It has nothing to do with the long term benefit this new technology will bring to society; it is about the scale of supply-side capital now chasing it, and about that supply one day meeting an unknowable medium-term demand curve. This is history rhyming.

“In virtually all financial innovations and investment fads, Wall Street creates additional supply until it equals and then exceeds market demand. The profit motivation of Wall Street firms and the intense competition among them render any other outcome unlikely.” Seth Klarman, Margin of Safety, 1991

“We knew we needed more funding. We also knew that the boom was going to end…. That March 2000 funding bought us the time we needed to make PayPal a success. Just as we closed the deal the bubble popped.” Source: Peter Thiel, Zero to One, 2014

Being knowledgeable about past cycles, and the non-linear growth they produce, might be a powerful tool for investors one day soon.

How to invest…carefully!

Markets today are betting that SpaceX, Nvidia, OpenAI and Anthropic will be the AI winners. Also, that AI/semiconductor profit growth will keep rising in a linear fashion.

Markets also know that Google, Facebook and Amazon were born out of the web age, but they assume tomorrow’s dominant players will be newcomers, as they were in the 2000s. That is not illogical thinking, but it might be wrong. Digital vs. analogue was like oil and water, and the transition was almost impossible for any business that was dominant pre-c.1997 to make. AI, however, is a digital tool and while it is a threat to some businesses, it is a boost to others that are already digitally empowered and ingrained.

The generation of businesses that were left behind in 1999 were nothing like the market leaders we have today. Today’s leaders are run by visionaries with a huge focus on innovation and a desire to quickly adopt whatever technology will benefit their customers. They are also very happy to spend heavily, either when customer demand requires it (Amazon) or when they think a pivot is coming and they want to be ready for it (Google, Meta, Microsoft). The spending such companies have been making over the last two years, that worried markets, is a sign of how fast they wish to evolve in an AI world, and of their early commitment to do so. The lazy analogue leaders of 1997 did not do this. Many of these companies look attractive to us.

So much change… or so little

“The atom bomb will change everything except how men think” Albert Einstein

The share price changes of many companies in the last year indicate the seismic change Mr Market thinks AI is going to bring to all parts of the global economy. We see this change too, but we think some things will stay surprisingly the same. Mostly the desire for value for money and great customer service. Next and Walmart are two examples of pre-1998 winners who did adapt brilliantly to the digital world. Both did so by continuing to excel in offering great value for money and a good customer experience. We think future winners will have similar traits (Amazon, Nu, Wise + SEA are examples).

If customers and advertisers continue to love the experience, innovation and value they get from using say Instagram or Nu, they will not look elsewhere. The same is true across a great many of the companies we favour. TSMC stands out today as the supplier of choice to the world’s semiconductor designers. Yes, it has technical dominance, but it also has market power. This was derived from working hard to give customers value for money and great service for c.30 years.

In this context, Amazon stands out as both a present and a future winner. No entity globally has focused on customer (consumer and B2B) value and service like Amazon. Whilst obvious to any retail user of the business, the jewel in their crown today is AWS. AWS has developed an unparalleled relationship of digital trust with many of the largest and most complex organisations on earth. As AI usage scales, they will be trusted more than anyone to implement it powerfully across enterprises in a safe and cost-effective manner (see Andy Jassy/our view on this here).

AI investing danger zone

Today we see great value in founder run, innovative, customer focused, powerful network businesses. In contrast we think there are a few AI danger zones.

The digital age brought cheaper broadband, unlimited music streaming and brutal price transparency to a great many products and services (Thank you Amazon). We think AI will help bring that transparency to a few remaining legacy industries. We think AI agents will/are coming for the fat profit margins made by those who enjoy a little customer apathy even in the face of poor service. The big vulnerable pools look to be banking, cable/telecoms and some utilities. These are places where some customers today remain on aging contracts that allow incumbent groups to make higher than justified margins, even in the face of a poor customer experience.

This is largely due to the hassle of switching. If AI agents make this process easier, customers will switch to more keenly priced contracts. Today you have some loyalty to Netflix or Amazon due to the price you pay for the value you get. But do you have loyalty to Sky who ratchet up prices at contract expiry, or utilities who do not answer the phone? We thought not!

The banking pool is particularly interesting. The whole global sector relies on NIM^[1] for its core profitability. If more people seek out better interest rates on current account type balances then core banking profitability will come under sustained pressure. Courtesy of customer champions like Wise and Nu, this is already happening today. AI we think could supercharge this process. This is profit-pool the sector will find hard to replace. (#my iceberg is melting)

Deflation and Economic Cycles

Today inflation and higher interest rates are the theme. Oil, wars (trade and real) and welfare have all played a part. However, we observe that both AI technology itself and the capital that is chasing it today are deflationary. Economic crystal ball gazing is not our thing, but the learnings of 1999 are worth understanding. After the TMT bubble popped in 2000, the resulting economic contraction saw FED rates reduce from 6.5% in 2000 to 1.75% in 2002. Arguably this created the housing boom that followed. During this 2002-2008 period the deflationary effects of digitisation (and globalisation) helped sustain strong economic growth at low inflation rates. This is no forecast, but a scenario worth being aware of. Cycles are powerful and often have knock on consequences – the AI capital cycle will be no different.

I’ve seen this movie before

The longer this AI cycle runs the more it looks like a replay of 1999. Then, like now, the stock market was highly bifurcated with great companies like Diageo or Unilever trading at low multiples.

Recently I listened to Paul Tudor Jones who also sees today as a 1999 replay. He was at a conference and asked the audience to put their hands up if they were investing professionally in 1999. Only 20% of people did.

History rhymes when no one is left to remember it. The people asking the difficult questions of this cycle are those with experience of past ones. I started in 1987 and saw 1999 close-up as a sell side Telco analyst. That doesn’t make me right, but it does give me some context.

Whilst this AI cycle could affect the economic cycle, this is not 2008. Although some of the financing look as spurious! (see this great piece by Ares Management). 2008 taught investors that when a correction comes it can be dramatic and almost existential. 1999 learnings are very different, however. It saw a huge rotation back into underpriced quality companies.

We are positioning the portfolio for all scenarios but that one is looking plausible. I will sign off as per Top Gun. We are finding great companies run by great managers at great prices. A target rich environment indeed.

Thanks for reading and best wishes

Andrew + Team

Related reading

Holland AI capital cycle roadmap – part 1

Technological Revolutions and Financial Capital written by Carlota Perez hugely helped my capital cycle thinking

Mike Burry’s Substack

Stripe’s 2025 annual letter had some good thoughts on Agentic commerce starting on page 9

Ares AI funding insights. Excellent piece on the opaque funding of AI data-centres (recommended by Mike Burry)

Joel Mokyr piece on why new technology adoptions can be fast or slow (recommended by Collison Brothers)

The Directors and employees of Holland Advisors may have a beneficial interest in some of the companies mentioned in this report via holdings in a fund that they also act as managers to.

Disclaimer

This document does not consist of investment research as it has not been prepared in accordance with UK legal requirements designed to promote the independence of investment research. Therefore even if it contains a research recommendation it should be treated as a marketing communication and as such will be fair, clear and not misleading in line with Financial Conduct Authority rules. Holland Advisors is authorised and regulated by the Financial Conduct Authority. This presentation is intended for institutional investors and high net worth experienced investors who understand the risks involved with the investment being promoted within this document. This communication should not be distributed to anyone other than the intended recipients and should not be relied upon by retail clients (as defined by Financial Conduct Authority). This communication is being supplied to you solely for your information and may not be reproduced, re-distributed or passed to any other person or published in whole or in part for any purpose. This communication is provided for information purposes only and should not be regarded as an offer or solicitation to buy or sell any security or other financial instrument. Any opinions cited in this communication are subject to change without notice. This communication is not a personal recommendation to you. Holland Advisors takes all reasonable care to ensure that the information is accurate and complete; however no warranty, representation, or undertaking is given that it is free from inaccuracies or omissions. This communication is based on and contains current public information, data, opinions, estimates and projections obtained from sources we believe to be reliable. Past performance is not necessarily a guide to future performance. The content of this communication may have been disclosed to the issuer(s) prior to dissemination in order to verify its factual accuracy. Investments in general involve some degree of risk therefore Prospective Investors should be aware that the value of any investment may rise and fall and you may get back less than you invested. Value and income may be adversely affected by exchange rates, interest rates and other factors. The investment discussed in this communication may not be eligible for sale in some states or countries and may not be suitable for all investors. If you are unsure about the suitability of this investment given your financial objectives, resources and risk appetite, please contact your financial advisor before taking any further action. This document is for informational purposes only and should not be regarded as an offer or solicitation to buy the securities or other instruments mentioned in it. Holland Advisors and/or its officers, directors and employees may have or take positions in securities or derivatives mentioned in this document (or in any related investment) and may from time to time dispose of any such securities (or instrument). Holland Advisors manage conflicts of interest in regard to this communication internally via their compliance procedures.

1. Net Interest Margin – The difference between what banks pay in interest on deposits and what they charge on loans. ↑