Ketu Desai
By Ketu Desai
Margin calls, positioning, de-grossing, options, rebalancing, and technicals dominated the market action in July. The epicenter of the action was Korea. Before the last day of the month, the KOSPI was on pace for its biggest monthly drop ever, larger than in 2008. Korean investors entered the month highly levered both Korean stocks and US tech stocks. According to Goldman Sachs, 1.2 million Korean brokerage accounts were hit with margin calls. Approximately 320,000-360,000 accounts were fully liquidated. Approximately 3.4% of the Korean population got margin called compared to 1-5% of US retail accounts during the 2008 crisis.
July was a reminder of how important understanding market structure is for investors. Much of the volatility was exacerbated by market structure. Multi-manager platforms (pod-shops), levered hedge funds, quant funds, options traders, and retail now dominate. Coming into the month this cohort involved very long AI infrastructure stocks and short certain large cap tech, software, and equal-weight S&P names. The Korean margin call created volatility in AI infrastructure names.
Many AI infrastructure names were down 20-40% in just a couple of weeks. This caused a massive de-grossing / de-leveraging event for certain levered hedge funds, pod-shops, quant funds, retail investors, and CTAs. The current market structure allows for quick gains, as we have seen with AI infrastructure names, but will take away gains just as quickly. What kind of investor you are is incredibly important so that emotional decisions are avoided during these periods of extreme volatility. Understanding what these investors are long and short is also critical.
Creating a balanced portfolio will help ride out volatility. Fixed income won’t hedge a 20-40% downturn in a couple of weeks. Understanding why they are short and going to cover in a de-grossing event will help mitigate losses in the long run. Position sizing will also be critical. Finally, having conviction in fundamentals can go a long way and let you play offence during these periods.
This de-grossing / de-leveraging event is an opportunity for long-term investors in AI infrastructure names. Positioning has been cleaned up with the largest selling on record according to Goldman Sachs. We are going through a generational value transfer from hyperscalers and software to AI infrastructure and hardware. CapEx estimates for 2027 continue to move up, now expected to be north of $1 trillion.
The software companies are now confirming the transfer of value to AI infrastructure and hardware. IBM preannounced earnings saying, “In the last few weeks of June, we saw clients shift their quarterly CapEx spend toward servers, storage, and memory purchases to secure supply-constrained infrastructure ahead of expected price increases. This dynamic impacted client buying patterns. While we anticipated some supply chain related impact in our expectations, we did not anticipate the magnitude of the capex reprioritization.”
The fundamentals for AI infrastructure remain excellent. South Korea’s semiconductor exports climbed by 181% YoY in July. Taiwan’s June export orders from the US rose nearly 84% year-on-year, the fastest pace on record. DRAM prices continue to accelerate upward as there is essentially no GPU availability. Taiwan Semiconductor also raised CapEx and prices. SK Hynix’s Chairman said that the memory shortage will worsen next year and could last past 2030. King Slide, a Nvidia AI server rail supplier, posted accelerating revenue growth of 220.5% year-over-year for June.
Foxconn, a large maker of AI servers, reported its biggest year-over-year revenue growth (+52.11%) for the month of June. GE Vernova’s order growth hit 88%. ASML said on its earnings call, “ASML’s 2Q reinforces our view that the AI-driven semi investment cycle remains in its early stages, with demand strength across both advanced logic and memory, and meaningful 2027 and 2028 orders already in hand.” The release of Chinese open-source model Kimi K3 will likely accelerate the AI war between both the US and China and frontier models and open source. That leads to even more demand for compute and AI infrastructure.
This is not a boom-and-bust cycle driven by PC upgrades or consumer electronics. This demand for compute is driven by digital agents that are working 24/7, 365. We’re just at the beginning of recursive self-improvement, despite not reaching the point of autonomous vehicles in mass or robotics, which will require more than 10x the compute power. With the correction this month, you’re able to buy infrastructure names below both their historical average and market multiple.
Looking forward, the market will focus on the latest in geopolitics, earnings, economic data, and the Fed’s Jackson Hole meeting.
