Since the start of 2025, market leadership has shifted. Emerging market (EM) equities have delivered roughly double the returns of developed markets through July 2026, outstripping indices including the Nasdaq. However, these gains have been narrow, driven largely by information technology on the back of AI hardware demand. This became particularly acute in the second quarter of 2026, when the sector rose by some 73% – the only GICS sector to outperform the broad EM index.
This performance pattern was also seen at the country level, with tech-heavy South Korea and Taiwan leading index returns, along with other geographies linked to the AI theme, as market sentiment remained positive. Unsurprisingly, questions around the durability of the recent technology rally have grown, with concerns over stretched valuations triggering a period of heightened volatility in a number of leading technology names in July. Companies related to the AI theme that had driven markets reported strong headline revenue and profit growth, yet still saw their shares pull back, reflecting increasingly demanding consensus expectations.
The engine for the technology sector has been hyperscaler capital expenditure, which is up from roughly US$110 billion in 2020 to a projected US$1 trillion by 2027, with a fourfold increase expected to come since 2024 alone. That capex flows into AI accelerator chips and high bandwidth memory (HBM) orders faster than fabrication plants can expand, since new capacity typically takes two to three years from decision to output. Each gigabyte of HBM also consumes roughly three times the wafer capacity of standard dynamic random-access memory (DRAM). Leading manufacturers such as SK Hynix and Samsung Electronics have diverted existing lines away from conventional memory, squeezing supply on both fronts and driving DRAM contract prices up sharply over the past year, as illustrated by Figure 1.
Figure 1: Supply-demand dynamics are reflected in DRAM pricing
Your chart will be shown here
Source: Company accounts, Bloomberg, Ninety One. June 2026.
Your chart will be shown here
Source: Bloomberg, DRAM spot price DDR5 1Gx16, Ninety One. 30 June 2026.
We see the memory cycle as structurally different from prior upcycles. An increasing proportion of memory demand, particularly in strategic and advanced-memory applications, is now locking in future capacity through longer-term customer agreements, which are setting price floors above prior-cycle peak margins. In effect, a historically commodity, spot-priced business is becoming a contracted supply chain, improving revenue and margin visibility more directly than in prior cycles. Supply cannot respond quickly to this demand, with capacity additions likely to remain constrained into 2028. Together, this points to tighter-for-longer memory fundamentals, improving through-cycle returns at the leading memory suppliers. In our view, this is the more durable story than near-term volatility.
Consensus estimates have yet to catch up; despite the sector's recent de-rating, they continue to underappreciate the durability of growth and profitability at the leading memory suppliers, in our view.
The recent earnings season reinforced our view that this is a fundamentals story, not a pure momentum one.
SK Hynix1 delivered record revenue and margins but fell on the day regardless, driven by average DRAM pricing that came in below elevated consensus expectations. We believe the shortfall was a matter of timing rather than demand, as higher-value shipments were pushed into the second half of the year as production shifts toward the next generation of AI chips. Management pointed to a growing number of long-term supply agreements now locked in, with more in progress, and raised capital expenditure to fund this accelerated capacity. The stock trades on a depressed multiple of near-term earnings, which in our view understates the visibility now embedded in those long-term contracts.
Samsung Electronics posted a sharp recovery in memory profitability from a year earlier, when export-control-related write-downs had pushed its semiconductor margins to near breakeven. The company shipped next-generation HBM samples ahead of its rivals and guided to an uplift in the shipments of its current generation, evidence that the HBM positioning gap that has weighed on the stock for two years is closing. TSMC's results followed a familiar trend, with an increase to its full-year growth outlook and capex. However, similar to SK Hynix, the share prices of Samsung and TSMC reacted negatively following these releases, which we read as investors taking shorter-term profits after a period of elevated expectations.
We remain of the view the quality of earnings behind leading memory players has improved, with multi-year supply agreements accompanied by price floors set above the previous cycle's peak, which should make future earnings less sensitive to short-term pricing swings.
The strong performance of South Korean equities this year has been accompanied by a marked increase in volatility. The KOSPI Index had more than doubled before the recent pullback, with SK Hynix and Samsung Electronics – together accounting for around 60% of the index2 – responsible for a significant portion of the market moves.
The increased use of leveraged exchange-traded funds (ETFs)3 by retail investors in Korea, alongside rising margin financing, has added to the concentration dynamic by contributing to more crowded and leveraged positioning. This has amplified the market's sensitivity to shifts in sentiment, increasing the risk of forced selling and exacerbating volatility. Daily KOSPI moves of 5–10% have therefore not been unusual over recent months.
This volatility is also reflected in the betas of the key names. While TSMC's beta, a measure of sensitivity relative to the market, is essentially unchanged across trailing look back windows, betas of Korea-based Samsung Electronics and SK Hynix have notably increased over more recent periods, as shown in Figure 2. Risk models built on historical relationships may understate current levels of risk, as the underlying volatility and correlation characteristics of these stocks have shifted from longer-term norms. This raises the possibility that realised portfolio risk could run materially above what standard forecasts imply.
Figure 2: Recent volatility has impacted the betas of key names in the EM tech space
Your chart will be shown here
Source: Ninety One, Bloomberg as of July 2026. Beta vs MSCI EM calculated over 1, 2 and 3-year windows.
This gap between modelled and realised risk has been a key consideration in how we manage our portfolios. We remain disciplined to a consistent and repeatable process, with clear guardrails in how we manage position sizing and portfolio construction. Therefore, where a sharp rerating raises the risk that a position contributes without a corresponding shift in the investment case, the process requires that this be addressed.
In this instance, our response has been three-fold: proactively managing risk and position sizing4; diversifying our technology exposure across the supply chain ecosystem to address the concentration risk; and balancing capital allocation with names less correlated to the global semiconductor trade.
Another aspect of us proactively managing risk is to remain fully abreast of any potential issues that could undermine our current view. With that in mind, we see three catalysts that could unwind the technology trade:
This would reduce the revenue opportunity for semiconductor suppliers and challenge the pace of AI infrastructure investment. This includes related considerations on financing: while hyperscalers currently fund capex largely from internal cash flow, a move towards debt financing given intensified spending could sharpen the debate on whether management teams become increasingly willing to lever balance sheets, and whether capital markets are prepared to underwrite that additional investment. Market sentiment towards this outcome could have a second derivative effect on EM names that have been supported by capex spending, even before expenditure slows.
The emergence of credible Chinese competition in memory manufacturing could put pressure on pricing and weaken the supply discipline supporting current earnings expectations for the leading Korean producers.
Increasingly capable models are being developed without equivalent reliance on leading-edge chips, which could challenge the assumption that AI advancement requires the scale, pace and technological intensity of current infrastructure expansion. However, recent developments have looked more like evidence of Chinese catch-up than a step-change reduction in compute intensity. On that basis, the implications for the AI infrastructure supply chain appear less negative than initially feared.
In our view, the risk to earnings is more likely to emerge from 2027 onwards, as semiconductor EPS revisions have continued to move higher and hyperscaler capex expectations remain firm. We continue to monitor the trajectory of hyperscaler capital expenditure, inventory levels across semiconductor companies, and revenue growth at major AI application providers for a read-through on the sustainability of AI infrastructure demand and the extent to which the current investment cycle is translating into fundamental evidence for the infrastructure players.
Concentration risk, while reduced through the recent de-rating and de-leveraging, has not disappeared entirely. A small number of companies now account for a large share of the Emerging Markets and Asia opportunity set, backed by corporate fundamentals superior to those of the state-dominated enterprises of the past, but nonetheless leave the asset class more exposed to idiosyncratic developments in a handful of names.
We believe that selling out of the AI/semiconductor complex could mean giving up a structurally underappreciated multi-year earnings story. Our investment framework5 distinguishes between the stock-specific risk, which we see as favourable (tight supply, rising profitability, multi-year visibility), and market-behaviour risk (retail/ETF-driven volatility) which is real but addressable through the portfolio construction described above. In this way, we believe it is possible to remain invested in a structural theme that still looks early rather than exhausted.
Share prices in the most exposed names already reflect a degree of scepticism around the durability of current profitability and discount much of this risk through depressed multiples. In line with our bottom-up, style-agnostic approach, we continue to be selective in our capital allocation. Our conviction remains fundamentally underwritten, with the flexibility to adjust as risk-reward dynamics evolve among potential winners and losers, ensuring our portfolio is prepared to capture the current EM multi-year ascent.
1. No representation is being made that any investment will or is likely to achieve profits or losses similar to those achieved in the past, or that significant losses will be avoided.
This is not a buy, sell or hold recommendation for any particular security.
For further information on specific portfolio names, please see the Important information section.
2. Source: Bloomberg as of 30 June 2026.
3. Korea Economic Institute of America (KEI) analysis shows average daily trading turnover in ETFs increased from KRW6.6 trillion in December 2025 to KRW34 trillion in June 2026 (US$23 billion).
4. The portfolio may change significantly over a short space of time.
5. For further information on the investment process, please see the Important information section.
General risks. The value of investments, and any income generated from them, can fall as well as rise. Costs and charges will reduce the current and future value of investments. Past performance does not predict future returns. Investment objectives may not necessarily be achieved; losses may be made. Target returns are hypothetical returns and do not represent actual performance. Actual returns may differ significantly. Environmental, social or governance related risk events or factors, if they occur, could cause a negative impact on the value of investments.
Specific risks. Geographic/Sector: Investments may be primarily concentrated in specific countries, geographical regions and/or industry sectors. This may result in wider fluctuations in the value of the portfolio compared to more broadly invested portfolios. Emerging market: These markets carry a higher risk of financial loss than more developed markets as they may have less developed legal, political, economic or other systems. Equity investment: The value of equities (e.g. shares) and equity-related investments may vary according to company profits and future prospects as well as more general market factors. In the event of a company default (e.g. insolvency), the owners of their equity rank last in terms of any financial payment from that company.
