What happened

On 7 July, Samsung Electronics released preliminary results for the second quarter. Revenue of ₩171 trillion. Operating profit of ₩89.4 trillion — roughly US$58 billion.

Both are all-time records. Both beat the street's estimate of about ₩85 trillion.

For scale: that single quarter of operating profit tops the previous records of Nvidia (US$53.5 billion) and Apple (US$50.9 billion). By this measure, the best quarter any company has ever printed.

And the stock fell 6.9% that day.

At one point it was down nearly 10%, breaking below the 290,000-won level. SK Hynix fell with it. The two stocks dragged Korea's KOSPI index down 4.9% — enough to trigger a circuit breaker, a forced trading halt, at 1:51 pm.

Why it matters

First, let's clear up the obvious confusion. A stock doesn't move on results. It moves on the story that comes after the results. If good news was already expected, it isn't news.

That's exactly what happened here. Samsung shares had climbed about 10% since mid-June, heading into the report. Profit estimates had jumped from ₩56 trillion in April to nearly ₩90 trillion by late June. The market had already bought the good news in advance. So the announcement only confirmed what everyone knew — no reason to buy more. The old market saying, word for word: buy the rumour, sell the news.

That answers "why sell on that particular day." But Signal Note's interest is the bigger worry underneath. Three things stacked up.

First, this profit came from selling dearer, not selling more. DRAM contract prices jumped about 60% in a single quarter. Samsung didn't ship dramatically more chips — it sold the same chips at much higher prices. Memory booms usually climb two steps: first orders grow (volume), then supply runs short and prices spike (price). The price-driven stage tends to show up near the end of the party, not the beginning. So the better the number, the louder the question: how much longer does this last?

Second, days before the report, Meta threw cold water on the whole sector. On 1 July, Meta unveiled a plan called Meta Compute — selling spare capacity from its own data centres to outside customers. It sounds minor. To the chip market, it hit a nerve.

Until now, big tech firms like Meta were pure buyers — the whales whose relentless purchasing kept chips scarce and prices climbing. Now one of them said: we'll resell what we don't use. That is a signal that a buyer can turn into a seller. On that one remark, US chip stocks — Micron, Intel, AMD — dropped roughly 6–10% overnight. The fear carried straight into Samsung's earnings day.

Third, Apple started moving to buy elsewhere entirely. Around the same time, Bloomberg reported that Apple is negotiating to put chips from China's CXMT and YMTC into iPhones sold in China. By the 8th, follow-up reports said Apple had begun testing CXMT memory and was lobbying the US administration for approval.

Until now, Apple bought memory from exactly three companies: Samsung, SK Hynix, and Micron. But with memory prices surging — enough to force MacBook and iPad price hikes — Apple put even a company on the US Defense Department's blacklist onto its supplier shortlist. Yes, it's limited to China-market devices, and US government approval is a big remaining hurdle. But the direction is the message.

Meta said "we'll resell our spare." Apple said "too expensive — we'll buy elsewhere." Different moves, same target: the pricing power of the three memory makers. The power to set prices only survives while buyers have no alternative — and the world's biggest buyers just started building alternatives.

Meta's reselling and Apple's diversification look like separate stories. But the market read them as one sentence: is the money pouring into AI infrastructure now passing its peak? What pressed the sell button that day wasn't three separate events. It was that one question.

So no, Samsung's stock didn't fall because Samsung failed. Samsung's yesterday — the results — was flawless. The market began doubting the whole industry's tomorrow — the demand. A great quarter couldn't cover that doubt.

In bottleneck language: in a semiconductor boom, the bottleneck never stays in one seat. From GPUs to memory, from memory to packaging, then on to power and cooling, then to networking and inference memory — pricing power moves to wherever the money passes next. Right now, memory sits at the front of that line. What the market read in ₩89.4 trillion wasn't the size of that seat. It was how much longer Samsung gets to sit in it. Meta and Apple just moved the expiry closer.

And every one of these bottlenecks is fed by a single tap: whether the hyperscalers keep spending at today's pace. What the market sold that day wasn't Samsung's earnings — it was a piece of the belief that the money keeps flowing. The market doesn't buy numbers; it buys the direction of numbers. And what it feared might bend wasn't Samsung's profit, but the direction of AI investment itself.

The market doesn't buy results. It buys the sentence after the results.

What to watch

The verdict isn't far away. Over the next month, the answer arrives in pieces, each with a date attached. Put these on the calendar and check one thing per date.

• 16 July — TSMC earnings call. The thermometer that shows AI demand first. Check: comments on AI chip demand and advanced-packaging (CoWoS) expansion. If "demand remains strong" holds, the Meta scare leans toward overreaction.
• 29 July — SK Hynix results. Check: whether the words "sold out" and "long-term supply agreements" still apply to HBM. If they do, pricing power still sits with memory.
• 30 July, 10 am KST — Samsung's full results and conference call. The first time division-level numbers and management's own voice appear. Check: the tone on second-half memory demand, and whether the phrase "capacity expansion" shows up — once that word appears, the bottleneck is starting to unwind. (Anyone can submit questions in advance on Samsung's IR site until the call begins.)
• Late July — Meta and Apple earnings season (dates per each company's IR). Check: whether Meta upgrades Meta Compute from an idea to a business plan, and whether Apple publicly mentions Chinese memory or component-cost pressure. If either becomes real, it's not a news item — it's the demand-supply structure changing.
• End of every month — DRAM and NAND contract price prints. Check: the pace of increase. If the rises start shrinking, this selloff was a signal, not a mood. Price is the bottleneck's thermometer.
• Ongoing — foreign investor flows. Foreign funds have been selling Samsung and SK Hynix since mid-June, dumping nearly ₩2 trillion of Samsung in a single day on the 7th. Foreign ownership is down to around 47% — the lowest in 17 years, since 2009. Check: the day the net selling stops. That day is the real verdict in the "cheap vs. peak" debate.

— Every item on this calendar, though, is a piece of the same question. The real axis is whether the hyperscalers keep spending at today's pace — the AI infrastructure CAPEX cycle. Where the investment language of Meta, Google, Amazon, and Microsoft stands right now — still accelerating, or already speaking the first sentence of a slowdown — deserves its own bottleneck-lens breakdown. That's the next post.

Disclosure: The author holds Samsung Electronics stock mentioned in this post.

For information only. Not investment advice.