To The Moon

The $1 Trillion Question: Did Wall Street Misread the Memory Crash?

If there is one lesson I keep relearning in semiconductor investing, it is that the market has the memory of a goldfish until it suddenly remembers it is dealing with the memory industry.

That is precisely what has happened to SK Hynix.

Only weeks after completing the largest first-time US listing ever by a foreign company, with a Nasdaq ADR valued at roughly US$28–29 billion, the company found itself caught in one of the sharpest sentiment reversals of the AI era. A stock that had climbed roughly 220% before its July peak abruptly lost around 35%, joining a wider sell-off across $Micron Technology(MU)$, Samsung and $SanDisk Corp.(SNDK)$ as investors abandoned what had been one of 2026's favourite trades.

The remarkable part is not the decline itself. Semiconductor shares have always been dramatic. The remarkable part is how little the underlying fundamentals actually changed.

When One Research Note Moves Billions

The spark was almost absurdly small compared with the reaction.

A Korea Investment & Securities note projected second-quarter operating profit about 8% below consensus. That alone sent $SK hynix(SKHY)$ shares tumbling 15.4% in Seoul—the worst single trading session in the company's history.

Yet the same report still implied operating profit would increase roughly 556% year-on-year.

That disconnect fascinates me. Markets like to believe they price information. More often, they price narratives. Once investors decided the AI memory boom might be peaking, an 8% earnings miss suddenly became evidence that the entire investment thesis had collapsed.

Those are very different conclusions.

The Financial Story Doesn't Look Broken

Looking purely at the numbers, I struggle to reconcile the panic with the business itself.

Trailing twelve-month revenue has surged 145% to KRW 189.2 trillion. Net income has exploded more than 460%, while free cash flow has reached almost KRW 92 trillion. Operating margins have expanded to roughly 68%, with EBITDA margins approaching 76%.

Those are extraordinary figures for any industrial company, never mind one traditionally viewed as highly cyclical.

The valuation also looks surprisingly restrained. Even after years of AI enthusiasm, the shares trade on a price-to-earnings ratio of about 8.4 times, with forward earnings falling closer to four times. Wall Street's average target still sits roughly 65% above today's share price.

Equally striking is the balance sheet. Cash has swelled while leverage remains remarkably modest, leaving SK Hynix entering this uncertain period from a position of considerable financial strength. Previous memory downturns often forced manufacturers to protect their balance sheets by cutting investment or preserving liquidity. This time, the company has far greater flexibility to absorb volatility without compromising long-term capacity expansion.

None of that guarantees upside, of course. Cheap stocks sometimes deserve to be cheap.

But businesses generating this level of profitability rarely receive bargain valuations unless investors believe those earnings are about to disappear.

That is the central debate.

Fundamentals improved while sentiment moved in the opposite direction

The Timing Trap Investors May Be Missing

The most interesting argument in SK Hynix's favour receives surprisingly little attention.

High Bandwidth Memory is not sold like ordinary DRAM. Contracts are often negotiated 12 to 36 months before delivery, meaning reported earnings can lag actual demand by several quarters.

Independent research has suggested July's panic may simply reflect accounting timing rather than weakening AI demand. Korean customs data reportedly showed memory exports reaching record highs even while reported profits appeared softer than expected.

If that interpretation proves correct, investors may have mistaken a reporting lag for a collapse in demand.

That distinction matters enormously because markets have an unfortunate habit of reacting to the income statement while ignoring where products are actually going.

Competition Has Become a Margin Fight

The AI race is no longer about building the fastest chips. Increasingly, it is about who captures the economics surrounding them.

Samsung remains the industry's manufacturing heavyweight and is investing aggressively to narrow the HBM technology gap. Micron, meanwhile, has executed exceptionally well over the past year, steadily improving both yields and customer confidence with hyperscale buyers. Neither company can be dismissed. The real question is not whether SK Hynix has credible competitors—it clearly does—but whether either can expand capacity quickly enough to erode the scarcity premium that currently supports HBM pricing.

SK Hynix currently commands roughly 56% of global HBM revenue and reportedly has much of its 2026 supply already committed. Its customer list includes the biggest names in AI infrastructure, including $NVIDIA(NVDA)$, Google and $Microsoft(MSFT)$, with reports also linking it to OpenAI's supply chain.

That creates genuine scarcity.

However, scarcity only remains valuable until buyers regain negotiating power.

Goldman Sachs recently downgraded the shares, arguing hyperscale customers may begin squeezing HBM pricing through 2026 and 2027. If pricing weakens while production expands, today's spectacular margins could prove difficult to sustain.

Meanwhile, China's CXMT has raised substantial capital through an estimated US$8.6 billion IPO, reviving concerns that Beijing-backed DRAM capacity could eventually reshape industry economics.

Memory investing has always looked wonderfully simple until someone builds another factory.

Separating Noise From Signal

Another overlooked aspect is the broader market backdrop.

SK Hynix did not collapse in isolation.

The decline coincided with a broader de-risking across growth equities as geopolitical tensions surrounding the Strait of Hormuz, rising oil prices and a rotation towards defensive sectors pushed investors away from richly owned technology names.

Nearly every AI infrastructure stock suffered.

By some estimates, more than US$1.5 trillion in market value disappeared across the global semiconductor sector in little over a month. That matters because it reframes SK Hynix's decline. Investors weren't simply reassessing one company's earnings outlook—they were simultaneously repricing the entire AI infrastructure trade. Sometimes the tide really does lower every boat.

That does not mean SK Hynix is immune from company-specific risks, but it does suggest some of the selling reflected portfolio positioning rather than deteriorating fundamentals.

Investors often confuse correlation with causation during market panics.

Sentiment reversed faster than the underlying business

The Real Question Isn't Demand

SK Hynix has become less a memory stock than a referendum on the economics of artificial intelligence.

The bullish argument remains straightforward. Scarcity is still real, profitability remains exceptional and valuation continues to look surprisingly undemanding.

The bearish argument is equally coherent. Memory has always been cyclical, pricing power eventually fades and additional capacity—whether from established rivals or emerging Chinese producers—has a habit of arriving faster than investors expect.

Perhaps the most interesting outcome is that both sides can currently point to evidence supporting their case. The debate is no longer whether AI demand exists; it is whether that demand can remain profitable enough to justify today's extraordinary margins.

The market has now delivered its first verdict. Investors who feared July's sell-off signalled collapsing AI demand instead watched the shares rebound sharply following earnings. That doesn't settle the debate. It simply shifts the question from whether demand exists to whether today's extraordinary margins can endure.

That is a far harder question, and one unlikely to be answered by a single quarter. Wall Street spent July trying to resolve it in one violent sell-off. The more interesting answer will probably emerge over the next several years.

Markets deliver verdicts before evidence finishes arriving

@TigerStars @Daily_Discussion @Tiger_comments @Tiger_SG @Tiger_Earnings @TigerClub @TigerWire