Overview

                                                   

Since the performance clock started (1^st December 2025), the fund has under-performed both Bloomberg Small Cap Index Net total return and MSCI World Net total return. Whilst we have made a strong case why the index doesn’t make sense (and I do try to ignore the index), it is ultimately a yardstick on which clients will assess us and I want to have a no excuses approach as I believe the strategy / process should, over time, beat all indices.

Much of our underperformance however, reflects the limited exposure to AI-related equities and technical factors as I will discuss below. I am increasingly forming my view on AI and the fund positioning reflects that. As such I feel comfortable about where the fund is (notwithstanding the disappointing results) and am positive about the upside potential on offer over the next 3-5 years. That being said, I am keen to be challenged and hear a diverse range of thoughts as I believe this is probably the most critical investment question to answer over the next two years.

Detractors: Technical, Not Fundamental

My top 10 detractors since launch represent a 1,600bps headwind. Most were purchased at under 20x earnings with >20% EPS growth forecast over five years, precisely the starting conditions my framework identifies as attractive.

Deconstructing the returns for these detractors, consensus earnings estimates have actually risen 6% since launch, with only one material downgrade (Angi, −50%, a small position and subsequently sold due to a change in thesis). Excluding Angi the weighted upgrade is over 10%! Average share prices, by contrast, have fallen 32%. The de-rating reflects two distinct forces:

• Perceived AI losers (c.900bps): these are stocks that the market has reactively assumed will lose out from AI disruption. Regarding these holdings, I feel it is a “shoot first and ask questions later” approach, where the fears are overly simplistic. Instead, I believe that these companies will become AI beneficiaries; products that are deeply integrated into workflows, mission critical, small percentage of costs and with deep domain knowledge. These are the companies that will use AI to improve their products and add more value to clients with that domain knowledge. As these businesses demonstrate AI acceleration or immunity, I expect a re-rating. Even at unchanged multiples, the earnings growth already embedded implies significant upside.

  
  

• Technical selling (c.700bps): forced sellers (funds closing mainly in UK) and sentiment-driven positioning away from anything outside the AI theme. This should reverse as the technical overhang clears.

Ultimately history shows us that earnings follow share prices so either the stocks need to rise materially or all nine of those companies have big earnings downgrades to come.

The top 10 contributors (vs DMSCN) added 14.2% whilst the detractors cost the fund 16%. As such, the remaining c30 stocks in the fund cost c760bps (25bps each), which highlights the steady headwind that the fund positioning has had (again related to little AI exposure and technical selling from fund closures and selling to purchase AI stocks).

Big thematic drivers tend to be a headwind for my approach given the framework prioritises economically agnostic businesses that have their destiny in their own hands (see the low stock correlation chart in the appendix).

 

It is also worth highlighting the more attractive growth and valuations in small cap (DMSCN index) given where we are in the cycle (see appendix on Small Caps vs Large Cap). The lower ROEs for the small cap index are structural given the higher proportion of loss-making companies, businesses earlier in their lifecycle (so margins haven't yet reached maturity) and higher percentage of declining businesses (hence why they end up in the small cap index).

 

Stepping back, the fund offers very attractive dynamics vs the two yardsticks mentioned above and which supports my positivity going forward. Given the opportunities, I want to be greedy in small cap and have the growth, the quality and the valuation, but importantly within the context of certainty of outcome and downside protection. These starting metrics are consistent with the exceptional upside that the fund is targeting.  

AI thoughts and outlook 

The AI build-out has been extraordinary in scale. Hyperscaler capex has nearly tripled since 2022, reaching approximately $700bn in 2026; equivalent to Belgium’s entire GDP deployed in a single year on AI infrastructure alone. The market has bifurcated sharply: AI-exposed names have surged, others left behind, while software and professional services stocks sold off as the market priced in a winner-takes-all outcome.

As a result of this excitement, the stock market has reacted with force, creating a bifurcated market of “with AI” and “without AI” as the charts below depict with momentum at an all-time high.

Index performance with and without AI

AI related companies on a tear at the expense of everything else

I. Who Actually Benefits? The key question

The dominant narrative is that vast infrastructure investment generates vast returns and has a seductive logic. But, history offers a clear counter-example. The telecommunications sector in the late 1990s executed the most ambitious infrastructure build-out of its generation and destroyed shareholder value on a colossal scale. Real infrastructure was built; genuine utility was created. The economics, however, accrued almost entirely to the consumer, not the builder.

In my view, AI will end up in the same place over time. Once supply and demand balance, it is likely we see multiple players in the market with uncertain differentiation, ensuring very low marginal cost for an additional user. As utilisation drives profitability, companies should (all things being equal) price toward marginal cost, exactly as telecoms did. The vast majority of AI functions do not require the latest model and models are increasingly interchangeable for most tasks. My phone bill today is lower than in 2000 and I receive dramatically more for it. AI inference costs have fallen approximately 10x annually since 2022; the same dynamic is already visible.

 

II.  Valuations Leave Little Margin for Error

AI-related stocks have re-rated aggressively and we have seen many signs of exuberance e.g. the Allbirds shoe company renamed itself Allbirds-AI and rose 10x. Even accepting the growth narrative in full, the mathematics are uncomfortable. AI semiconductor stocks (SOX index) trade at approximately 52x earnings. To justify that multiple requires 16% EPS growth per annum for a full decade from an already elevated earnings base, with approximately 75% of current equity value still sitting in terminal value a decade from now. That implies any investor today would need to take a (positive) view on the sector for 2036 with confidence. We are sceptical that anyone can.

There is also an accounting distortion worth flagging. Many AI infrastructure businesses appear attractive on traditional multiples, but depreciation only begins once assets are placed into service and today’s assets reflect capex committed 12–18 months ago. The current spending surge will flow through as materially higher D&A charges in future periods, compressing earnings precisely when the growth narrative requires them to expand. Telecoms went through the same dynamic; EV/EBITDA became the default lens for that sector because reported earnings were too distorted by depreciation to be meaningful. There is a further wrinkle specific to AI: GPU assets depreciate over 3–5 years and risk obsolescence before they are fully written down, meaning D&A may also understate the true economic cost.

Beyond earnings, the risk profile of the hyperscalers is changing. These were businesses with secure structural monopolies and high returns on capital. They have now entered a capital-intensive race with no clear barriers to entry and structurally lower returns than their core franchises. Several have begun funding the build-out through debt rather than operating cash flow which also changes the risk profile in ways current valuations do not appear to reflect. Their spending power drives demand for small-cap semiconductor and supply chain names, making them a relevant bellwether for the sector.

Finally, the supply chain itself is more complex than the headline capex numbers suggest. Advanced packaging, high-bandwidth memory, custom silicon, specialised cooling and grid connectivity each have their own lead times and bottlenecks. Complex supply chains under demand pressure do not deliver smoothly. They are lumpy, with shortages followed by oversupply as capacity arrives simultaneously. Near-term, this means revenue ramps may be slower and more uneven than consensus assumes. Longer-term, inevitable capacity overshoot creates conditions for sharp pricing corrections.

 

III.  Where Value May Actually Accrue (where I want to be)

If infrastructure providers face structural commoditisation, durable value is more likely to accrue to businesses where deep domain knowledge is married to AI capability, particularly where the product is mission-critical, deeply embedded in workflow and a small share of the customer’s total cost base. A customer relying on specialist software for regulated clinical data or niche financial reporting will not risk migrating to a generic AI alternative to save a modest fraction of costs. The proprietary data, embedded workflow and reputational trust are not replicable by an AI foundation model.

Lower AI prices also tend to expand markets rather than simply redistribute existing spend. If it becomes genuinely easy to reach an effective contact centre agent immediately, call volumes will rise, benefiting the specialist operator with AI capability and domain knowledge, not the generic AI provider. This is why I own NICE Systems: growing double-digit, trading at 8x P/E, with AI now 14% of revenue growing at 60% per annum toward an estimated 30% by 2028.

What we look for in hidden AI winners: 

• Mission-critical and deeply embedded (high switching costs).

• Small % of customer costs (pricing power sustained).

• Market expansion from lower prices (addressable market grows, not contracts).

• Proprietary data advantage (moat a new entrant cannot replicate).

 

On the flip side, the switching costs for AI providers are low and a view reinforced from experts. Enterprise AI technology specialists report that rankings of leading LLM providers have shifted dramatically from year to year and that provider selection is a straightforward value-for-money decision with no hesitation to switch. Brand loyalty is minimal; this market reprices constantly toward the lowest-cost provider.

                                                                                          

Conclusion: patient and rechecking

AI is real, transformative and the infrastructure being built will underpin meaningful productivity gains over the coming decade. What I dispute is that the current equity market distribution of value, heavily concentrated in infrastructure and semiconductor providers, accurately reflects where economic benefit will accrue.

History is clear: the economics of transformative infrastructure cycles nearly always favour end consumers and domain specialists over builders. In my view, the closest analogy is the telecom build-out of 1999–2003; genuinely transformative, with value flowing overwhelmingly to users, not constructors. Set against that, the Goodhart Global Smaller Companies portfolio, with a c.12x forward P/E, forecast double-digit organic growth, 20%+ CFROIs and improving margins, I feel that this is a much more attractive risk/reward than paying for a decade of uninterrupted AI infrastructure execution at 52x earnings. We continue to search for AI-adjacent smaller companies with non-replicable domain expertise, mission-critical products, and valuations reflecting a business rather than a sentiment. They exist. When rationalisation eventually compresses AI valuations indiscriminately, those businesses should start to shine alongside the forgotten idiosyncratic, AI-agnostic companies that fill the portfolio. As a fundholder myself, I remain excited about the portfolio over the next few years.

  

Appendix:             

Sell Discipline: Three Approaches:

A recurring question I receive is around my sell discipline. I now have enough data to show this first hand using the following frameworks within the context of holding onto my winners and letting them run:

Soitec - selling the re-rating

Soitec was purchased as a cyclical recovery play: trough earnings, trough multiples, with AI and silicon photonics optionality. I built the position from 65bps to 2% on weakness (stock fell from €35 to €25). When the AI optionality came through and the stock rose to c.€100, I trimmed meaningfully and sold the remainder at €150 as I could no longer justify the returns within my multi-bagger framework (left hand chart below). The multiple re-rated without the earnings (right hand chart below) and the cyclical upside was largely priced in, especially in relation to other opportunities in my universe.

Tasmea - upgrades and thesis / waypoints to drive position sizing

Tasmea is the contrasting case. Price appreciation from A$4 to A$9.5 has been driven largely by fundamental upgrades and thesis delivery, with minimal re-rating. I trimmed twice, both times a mistake. It is now one of my largest positions and I intend to hold it.

Sanuwave - when the thesis breaks

Sanuwave is a US medical device company with a clinically differentiated wound-healing product and an underpenetrated market. My thesis required >40% revenue growth in 2026 (with all the stars aligning), with high incremental margins amplifying the top-line inflection. Q1 guidance narrowed the full year to 16–25% growth, materially below my thesis and closer to my bear case (a “jam tomorrow” story). I sold with shares down 10% on the day and 30% from purchase. The shares continued to fall especially with 2 further downgrades in recent months.

Low correlation of stocks within the portfolio

Small Cap looks especially well-positioned historically vs Large Cap