By Robin Parbrook, portfolio manager, Schroder Asia Total Return
July is usually a quiet month in Asia as holiday season begins and markets wait for August’s financial results. This year was different. Extreme volatility swept through the region, with Korea and artificial intelligence (AI) related technology stocks at the epicentre.
Korea had been showing dangerous signs of a retail investor driven bubble, which started to unravel quite spectacularly in July. Much of the damage came from a new breed of leveraged exchange-traded funds (ETFs) on popular AI stocks such as Samsung and SK Hynix – products built to magnify whatever the underlying shares do that day, in either direction. Retail investors had piled in fast and, when sentiment turned, the leverage worked against them even faster.
See also: Beyond AI: Three overlooked Asian stocks demonstrating long-term growth
The most dramatic capitulation came on 29-30 July, when a highly leveraged hedge fund called Situational Awareness effectively collapsed under the weight of its AI infrastructure bets. Citadel, one of the world’s largest hedge funds, stepped in to take on its remaining positions, but not before the Korean market had fallen 35% for the month. It then rebounded 18% on 31 July. Taiwan followed a similar, if less extreme, pattern while other Asian markets rallied close to 10% on hopes of Middle East de-escalation and falling oil prices.
For us, July was a reminder that when expectations around AI wobble, markets tied most tightly to the AI infrastructure cycle can move violently. And that brings us to the two canaries in the coal mine investors should be watching closely.
Canary 1: The rise of Chinese open-weight models
We have been cautious for some time about the long-term assumptions embedded in AI capex forecasts and recent developments have added to our concerns. Chief among them is the rise of Chinese models such as Kimi K3 and Alibaba’s latest Qwen release, which are now nearly as capable as leading US models.
Because these models are ‘open-weight’ – meaning their underlying code is freely available for anyone to use and build on – and far cheaper to run, their usage is exploding and vastly outgrowing US models, which prompts increasing scrutiny over the assumed path to profitability for the likes of OpenAI and Anthropic.
Canary 2: The financing structure behind the AI boom
The second warning sign is about how the growth of AI is being paid for. The so-called ‘hyperscalers’ – the big tech companies building AI’s underlying infrastructure – have committed to enormous data centre build-outs, on the assumption that the AI labs renting that computing power will eventually be able to pay them back.
Those future payments already add up to more than $2trn, and much of the build-out itself has been funded through corporate debt. In effect, the hyperscalers are extending credit to the AI labs on a bet that demand for computing power keeps growing and that this usage eventually delivers enough revenue to service the debt.
Some observers are drawing uncomfortable parallels with the run-up to the global financial crisis – markets appear to be treating an uncertain promise of future payment as though it were secured income, rather than pricing in the possibility that it never materialises. The risks may not be quite so systemic, but the underlying pattern looks familiar.
What this means for the ATR portfolio
Despite these concerns, AI capabilities continue to surprise and compute demand is growing rapidly. Our tech analysts remain constructive on the long-term outlook, but we think some of the more bullish capex assumptions are vulnerable, which prompted us to gradually trim Taiwanese technology positions earlier this year after strong outperformance.
See also: Artificial intelligence: The environment’s ally
At the same time, we have added to more defensive, higher yielding stocks in Hong Kong, and consolidated our China internet holdings into companies where we believe the threat of AI disruption is less material. The net result is that, for the first time we can remember, the portfolio is overweight the China and Hong Kong cluster.
In short, while the AI canaries are still singing, we are listening carefully and stand ready to take further action on the portfolio should it be required.
