Asian semiconductor stocks are crashing. Here's why we're still buying.

July has been brutal to Asian semiconductor stocks.

The Global X Asia Semiconductor ETF (HKEX: 3119) is down 13.5% so far this month (as of 23 July 2026) — and at one point in July, the decline reached as much as 25% from its June peak. SK Hynix, one of the world's most important memory chip makers, fell as much as 15% in a single day — its worst day ever as a listed company. Samsung and TSMC, the two most important names in the entire global chip industry, both dropped sharply in the same window. South Korea's own stock market, heavy with chip names, had three of its worst single-session drops since the 2008 financial crisis — all of them in 2026 alone. Put together, the sector has lost more than a trillion dollars in value in a matter of weeks.

A double-digit monthly decline like that would normally be the kind of number that makes you want to look away. So it is a fair question to ask: is the AI chip boom over?

We don't think so. Before you draw any conclusions from that 13.5%, there's something you need to know: much of this crash had very little to do with semiconductors at all. It was about leverage.

The real trigger was leverage, not lost faith in AI chips

In the build-up to the sell-off, South Korean retail investors' margin loans — money borrowed from brokers to buy more shares than they could otherwise afford — hit a record KRW 38.63 trillion, or roughly USD 26 billion, on June 24, according to the Korea Financial Investment Association. Samsung and SK Hynix alone make up more than half of the entire Korean stock market, so when leverage runs this hot nationally, these two stocks are where much of it inevitably lands.

But the borrowing didn't stop at personal loans. Single-stock leveraged ETFs — funds that promise to deliver twice, sometimes three times, the daily move of a stock like SK Hynix or Samsung — launched in Korea on May 27. Retail investors ploughed a net KRW 14 trillion into them within weeks, seven times what foreign investors put in, according to KB Financial Group. Unlike margin loans, the leverage sits inside the fund itself, stacking a second layer of amplified risk on top of the borrowing already in the market.

When Samsung and SK Hynix dipped even slightly under the weight of all this borrowed money, brokers issued margin calls — demanding investors either put up more cash immediately or have their positions force-sold. Forced selling pushed prices down further, triggering the next round of margin calls, which forced still more selling. According to Goldman Sachs, more than 1.2 million trading accounts in South Korea were hit by margin calls tied to this rally, with 320,000 to 360,000 wiped out entirely — roughly 3% of the country's entire adult population caught in the chain reaction.

Scary stuff. But here is the detail that should reassure you most: South Korea's own financial regulator agreed this was the problem. It has moved to block new leveraged single-stock ETF launches and, from August 5, will triple the minimum cash balance required to trade these products, to KRW 30 million (about USD 20,300) — a direct admission that the products, not the underlying businesses, were the issue.

When a market stops running on borrowed money and starts focusing on fundamentals, that's a healthy correction for long-term investors — not a warning sign. The good news? The fundamentals of Asian semiconductor companies remain rock solid.

Strong results across the semiconductor supply chain

TSMC reported a 77% jump in quarterly profit — a record for the company — and announced a further USD 100 billion investment to build more chip factories in the US, on top of the USD 165 billion it had already committed. Its stock fell anyway.

Samsung reported a profit nearly 20 times higher than the year before. Its stock fell too.

SK Hynix's own chief executive, Kwak Noh-jung, told investors the industry is heading into "the worst year in the industry's history" for chip supply — meaning demand is so strong the company cannot keep up. His stock crashed the hardest of the three.

It didn't stop there. ASML, the Dutch company that makes the machines nearly every chipmaker on earth depends on, beat its own already-strong forecasts and raised full-year guidance for the second time this year, citing surging demand from both logic and memory customers. Texas Instruments, which makes everyday electronics for cars and factories rather than headline AI chips, reported that its data-centre business had doubled in a single year and that industrial demand was up roughly 30% — proof this isn't a story confined to one narrow group of famous names.

The strength doesn't stop at chip companies, either. Days later, Alphabet — Google's parent company — reported revenue up 24% and cloud revenue up 82%, both comfortably ahead of expectations. It also raised its own AI infrastructure spending plans to as much as USD 205 billion for the year, well above what analysts had pencilled in. That number matters directly to this thesis. Google's AI chips are built by TSMC in Taiwan, and Samsung and SK Hynix are among its largest memory suppliers — so a bigger spending plan from Google means more orders for the companies this article is about, even if not every dollar goes toward chips.

Does that sound like a story about weakening demand? It doesn't. It's the story of an entire industry — foundries, equipment makers, memory producers, and the world's biggest AI spenders — all confirming, in their own numbers, that demand hasn't slowed down.

Table 1: Strong results across the semiconductor supply chain

Some investors are calling this a bubble — here's why we disagree

We would be doing you a disservice if we pretended nobody is worried. 

Michael Burry, famous for correctly calling the 2008 financial crisis, has warned about unsettling similarities between today's chip rally to the dot-com peak crash in 2000. GMO's Jeremy Grantham goes further, calling AI 'obviously a bubble' and comparing it to the railroad and internet booms — genuine revolutions that still produced huge speculative excess. These warnings deserve to be taken seriously, not waved away.

But the difference is one of degree: far more of today's spending is backed by signed, contracted revenue than was typical in 2000.

In 2000, a large share of internet-era spending went toward unproven business ideas with little revenue to show for it — companies burning cash on the hope that profits would eventually follow. Today's spending looks different. Unimicron, a Taiwanese company that makes the materials chips are built onto, had to raise USD 1.4 billion this year for one reason: it could no longer fund the orders it already had using its own cash flow. That is not a company hoping demand shows up eventually. That is a company drowning in real, signed orders it can barely keep up with.

Alphabet is in a similar situation. Google's cloud business is now sitting on a backlog of USD 514 billion in contracted work, and management expects to deliver just over half of that within the next two years. That is money already promised to Google, for work it hasn't finished yet — the exact opposite of speculative building.

Samsung, TSMC, and SK Hynix are all making the same bet with their own money. Samsung's semiconductor chief has said the company is moving away from annual and quarterly deals toward supply agreements running three to five years, with customers including AMD, Microsoft, and Google — companies don't sign five-year contracts for a fad. TSMC has said its capital spending over the next three years will be "even more significantly higher" than the previous three, a notable statement from a company famous for spending cautiously. And this isn't a shortage anyone expects to end soon: SK Hynix's own chief executive has said demand will continue to outstrip supply "even beyond 2030."

The clearest evidence, though, is still in prices. Memory chip prices have risen close to ten-fold over the past year — not because anyone is speculating, but because real, physical supply genuinely cannot keep up with real, physical orders. Raising prices while supply is already this tight is a bigger signal than it looks. It means a company isn't just confident demand exceeds supply — it's confident customers have nowhere else to go. TSMC, which faces no real substitute at the cutting edge, has told customers to expect prices up to 10% higher starting next year. ASML — sole supplier of the machines needed to build the most advanced chips — has already gotten some customers to accept a 10% increase of its own.

So where is all of this demand actually coming from? The proof isn't confined to hardware and factories, either — it shows up in how ordinary businesses are putting AI to work. Cybersecurity firm CrowdStrike said its AI-specific security business went from nothing to a USD 50 million sales pipeline in under two quarters, including an eight-figure deal to secure more than 200,000 devices for a major US government agency. Enterprise software company ServiceNow said its AI-linked revenue crossed USD 1 billion this quarter, with the number of customers running AI systems in live, everyday production up ninefold in nine months.

Even outside of tech, the pattern holds. Eli Lilly has committed USD 1 billion over five years to build its own AI drug-discovery lab with Nvidia — real money already spent, not a pilot project. And it's paying off: in a recent industry survey, 44% of healthcare and pharmaceutical companies said AI has already increased their revenue by more than 10%.

We take the bubble question seriously. We are simply not convinced this is one.

We are still buying Asian semiconductors

What happens when share prices are down but fundamentals are unchanged? They become more attractive – and that's exactly what's happening to Asian semiconductors and why we are still buying. We were bullish on them before this sell-off began; we are even more bullish now given the cheaper valuations.

Chart 1: The ETF has just experienced its sharpest pullback in over a year

For investors looking to act on this, we continue to recommend the Global X Asia Semiconductor ETF (HKEX:3119) as the simplest way in. Rather than betting on any single company, it spreads your investment across the region's most important chip businesses — including TSMC, Samsung, and SK Hynix — capturing the AI infrastructure buildout across Taiwan, South Korea, Japan, and China in one position. Based on a fair PE ratio of 18 times applied to 2028 earnings estimates, our target price for the ETF is HKD 334. The ETF last traded at HKD 179 as of 23 July 2026, which we estimate implies upside of more than 86% over that horizon.

Even as we remain convinced the multi-year story is intact, we acknowledge that share price movements have been volatile. However, this is not a reason to avoid Asian semiconductors altogether. Investors uncomfortable with near-term swings should size their positions accordingly or slowly build exposure using regular savings plans.

The businesses behind this month's sell-off did not get weaker. If anything, they told you — in their own results, in their own words — that the opposite is true.

If there is ever a time to buy the dip, it is now.

Declaration:

This research report was prepared with the assistance of artificial intelligence (AI) tools. iFAST Financial Pte Ltd does not rely exclusively on AI for content generation; the content of this report – including all investment theses, ratings, price targets and conclusions – has been independently reviewed and verified by the research analyst(s) to ensure accuracy and professional integrity.

For specific disclosure, at the time of publication of this report, IFPL (via its connected and associated entities) holds a NIL position in the abovementioned securities. The analyst who produced this report holds a position in Global X Asia Semiconductor ETF.