What will define 2025?
Taylor Swift’s engagement? That Coldplay kiss? Kids saying “6-7” for no reason whatsoever? All valid.
But for Cusana, the extreme ways in which AI capex has warped EM equity returns will forever define 2025. The ongoing boom in AI capex has driven many EM anomalies. Below, a collection of “EM firsts” – at least this century:
- Style decoupling: in EM, Quality has never lagged either of Growth or Value by as much as it is lagging both in 2025. To the end of November, it trails both by c18% - a far starker underperformance than exists in DM;
- Country divergence: of EM’s “Big 4” markets (China, Taiwan, South Korea, and India), there has never been a greater difference in return between any two markets than South Korea’s outperformance of India in 2025 – the c75% difference (TR, USD) is greater than a +4stdev event;
- Passive dominance: there has never been a larger gap between EM ETF inflows relative to non-ETF outflows. Consequently, index concentration has ratcheted higher, and EM large cap has outperformed EM small cap by c12% - one of the most significant years of large gap outperformance on YTD25, the top 10 after TSMC has driven 27% of index returns whilst 2023-24 inclusive, its return contribution was just 13%.
This collection of peculiarities is a consequence of the EM equity market’s unrelenting repricing of AI capex beneficiaries. Semiconductors and other technology hardware sectors account for most of the index exposure in both Taiwan and South Korea. Therefore, these equity markets have led the EM equity rally of 2025. A cursory glance at the valuations of these two North Asian markets adds to the sense of extremities:
- MSCI Taiwan has not traded on a higher book multiple since the heady heights of the dot-com bubble. Over the two decades spanning the dot-com crash to the COVID crisis, Taiwan traded on a P/B of 1.9x. Today it trades on c3.8x – double its 21st century average;
- MSCI Korea has not traded on a higher book multiple since early Back then, there was widespread optimism that smartphone demand made memory a structural growth rather than cyclical business, and that Chaebol governance would improve. As this optimism faded, the Korean market became rangebound between 0.9-1.2x P/B for the following 15 years. A re-rating of the index P/B multiple back up to c1.6x reflects the degree to which 2025’s optimism mirrors that of the 2010-11 period.
Whilst this bull market could yet become bubblier, clients should be wary of extrapolating from a starting point that is already in outlier territory. Increasingly, incremental buyers must question their conviction in being the lesser fool: for how much longer will there be a greater fool to sell to at a higher high? An industry with intense cyclicality and a history of mean reversion is unlikely to retain peak margins for long, at which points the apparently reasonable P/E multiples will most probably be exposed as a value trap. Price/Sales multiples, like Price/Book, are indicative of very demanding market assumptions which competitive forces will likely undermine within a surprisingly short period. These are, therefore, highly risky speculations, not investments.
This capital flight to AI hardware has – in our view – created inefficiencies elsewhere within the EM universe. The psychological underpinning of this was on full display upon Tencent’s recent 3Q25 results:
“Topline beat, bottom line beat, strong guidance BUT key negative: AI capex revised down!!”
A reduction in AI capex is only a negative for market participants that are willing to capitalise incremental AI capital investments at a higher valuation than incremental free cash flows. The logic of this is, of course, contingent on the return that the investment is assumed to generate.
Rather than passively observing our holdings speculate on the future ROIC of AI servers, we are instead content to receive this free cash flow in the form of shareholder distributions. There is no shortage of provably high-ROIC investment opportunities across our portfolio and we are happy to make this capital allocation decision ourselves. We prefer the sensible capital allocation of Tencent to the profligate Alibaba, for example.
Our capital allocation seeks to exploit the inefficiencies that 2025’s great AI cyclical rally is leaving in its wake. These inefficiencies come in two overarching forms: 1- AI value capture beyond hardware; 2- EM mega trends beyond AI.
1. AI value capture beyond hardware: the application & services layer will generate substantial economic profits from the implementation of AI, contrary to share price declines in 2025 on broad-based multiple contraction:
- 1a. Internet ecosystems: marketplaces continue to own the system of record – an entrenched contract management system; AI marries unique datasets to unlock improved conversion, translating to superior ad monetisation; cost efficiencies from customer support & fraud detection from AI agents.
- 1b. Incumbent software: mission-critical systems with data security considerations and software embedded deep in multiple enterprise workflows will prove highly resilient; integration in customer tech stacks, trust and employee familiarity make B2B SaaS incumbents best-placed to implement AI features in the
- 1c. Digital engineering services: the slowdown in transformational IT services is a function of an extended post-COVID down-cycle - not structural disruption from AI; enterprises are complex and tech debt is real; AI implementation is complex and requires a modern data Digital engineering services will charge less per line of code, but code efficiency and scope of work both head ever higher.
2. EM mega trends beyond AI: as AI has hoovered up equity market attention (and capital flows), other thematic profit pools in EM that enjoy multi-decade, double-digit CAGRs have become increasingly attractively priced:
- 2a. India and ASEAN consumption: by 2035, India and Southeast Asia will have added 411mn and 112mn middle-class consumers respectively. Most of these young, increasingly wealthy consumers will transact online – we estimate that in just three years’ time, EM will have added more internet consumers than the total number that exist in the U.S. today[i].
- 2b. Financial inclusion: over the decade 2011-21, 7bn adults opened a bank account in EM whilst European and North America combined added just c70mn. Today, there remains a c1.4bn unbanked population in EM, atop c345mn unbanked SME enterprises[ii].
- 2c. Electrification: surging global power demands, decarbonisation goals, and new all-electric products (from robots to EVTOLs) all point to the systemically significant leadership position of China’s vertically integrated and low-cost battery and solar technology leaders.
Approximately 90% of the portfolio benefits from one or more of the non-AI thematic drivers detailed above. The strength of these tailwinds has helped to deliver strong compounding in 2025: the median Cusana holding is poised to deliver c33% earnings growth in 2025E. Our analysis is that this profit growth is far more sustainable than that of current AI capex “winners”. The market has not rewarded this, focusing instead on 2025’s great AI cyclical rally. Going forward, we expect these non-AI thematic drivers to underpin a forward earnings CAGR of c30% for the portfolio[iii].
Importantly, we believe that the correlation between these distinct growth thematics and the quantum of AI capex is low - or even negative in terms of stock market behaviour. The benefits of this diversification are increasingly valuable as EM and DM return drivers converge around AI enthusiasm: since ChatGPT launched in November 2022, AI has accounted for approximately 75% of S&P500 returns[iv]; this year, AI has driven c40% of EM returns[v].
So, as we look toward 2026, we would remind clients that while AI is undoubtedly a source of growth, not all growth is AI. The frenetic concentration of capital into a narrow set of upstream AI beneficiaries has created historic divergences—in market returns, factor performance, and valuations. While AI adoption will mint high-quality compounders, we fear many high-flying "AI plays" shall prove devoid of sustainable competitive advantages. The current euphoria has created a compelling opportunity to invest elsewhere: in the provably high-ROIC growth pockets across AI’s application layer and across the diverse, multi-decade EM mega-trends that remain uncorrelated with AI capex. We only invest in AI winners if they have wide competitive moats, such as TSMC; hence, very selectively and with a limited portfolio exposure (~15% at present). The result is a portfolio anchored in structural fundamental value – not cyclical enthusiasm – that we expect to deliver a future earnings CAGR of c30%.
