The Rhyme: Samsung's Record Profit Crowns Memory's Peak & Its 1995 Echo

"History doesn't repeat… but it rhymes." — Mark Twain

Samsung Electronics is expected to report roughly 85.5 trillion won in operating profit for the second quarter today, an 18-fold jump from a year ago and the largest quarterly profit any private-sector tech company has ever posted. The number, driven by memory chip prices that have risen sixfold in a year, would top Nvidia's record of $53.5 billion and push Samsung past Nvidia as the most profitable tech firm on the planet. Meanwhile, SK Hynix is three days from listing on the Nasdaq in a $28.21 billion offering that would be the biggest first-time share sale by a foreign company in American history, eclipsing Alibaba's $21.8 billion in 2014. South Korea's president just unveiled a $576 billion national chip investment plan. The word everyone is using is "supercycle."

Samsung Q2 op. profit est. ~85.5T won (~$55.9B) | SK Hynix Nasdaq IPO: $28.21B on July 10 | DRAM ASP +45-55% QoQ (Q1)

The last time memory chips made a country this rich, this fast, the calendar said 1995. And the people calling it a supercycle back then used the same word.

In the summer of 1995, a 16-megabit DRAM chip cost more than three times what any reasonable price model said it should. Demand was coming from everywhere at once. Microsoft had just shipped Windows 95, and every new PC needed four to eight megabytes of RAM instead of the one or two that had been standard. Unit shipments were growing at double-digit rates. The factories in Icheon and Giheung were running flat out, and it still was not enough. Samsung's semiconductor division was printing money so fast that the company's net profit jumped 165 percent for the year, a figure so large it barely fit in the annual report.

Korean memory had grown 8.3 times over from 1991 to 1995. Semiconductors accounted for 13.4 percent of the country's total exports, and the government was calling it the greatest boom in Korean industrial history. The chaebols took the compliment and doubled down. Samsung broke ground on new lines. Hyundai Electronics, the company that would later become SK Hynix, announced its own expansion. Across the industry, roughly fifty new fab construction plans were drawn up during 1995 and 1996 alone. Capital expenditure blew past 30 percent of total semiconductor production revenue, a level that in hindsight looks less like confidence and more like a fever.

The spot market for DRAM had been tight for two straight years, since 1993, and gross margins for leading suppliers sat well above 50 percent. Micron Technology in Boise was having the best year of its life, with revenue growing 81 percent. Analysts who covered the sector kept raising estimates. The phrase they liked was "structural shortage." The old boom-and-bust cycle, they said, was a thing of the past because PC demand would never stop growing.

Nobody noticed that the fifty new fabs were all being built to produce the same chips for the same customers. The ramp-up schedule called for those fabs to come online in late 1996 and early 1997, roughly eighteen months after the orders were placed. By the time the concrete dried, the shortage would be a memory. The price per megabit of DRAM, which stood above $3 at its peak, would begin a descent that didn't stop until it hit sixteen cents three years later. But in the summer of 1995, with profits doubling and stock prices climbing and a president praising the semiconductor industry as the engine of a modern Korea, none of that was visible yet. The only thing visible was the money.

A new platform creates real demand. Memory makers post record profits. A country bets its economy on the cycle continuing. Everyone agrees the old rules no longer apply. The pattern is the same. The only question is the timing.

The rhyme is tight, but it is not exact. Four things have changed since the last time memory was king.

Here is what happened after the 1995 peak. DRAM prices, which had been rising for two straight years, began to soften in late 1995 as the first wave of new fab capacity came online. The decline started slowly. A few percent here, a few percent there, enough for analysts to call it a "normalization." Then it accelerated. By the end of 1996, DRAM prices had fallen 51 percent. Micron's operating income, which had more than doubled the year before, shrank 28 percent. DRAM prices fell another 75 percent during Micron's fiscal 1996, then another 40 percent in fiscal 1997. From peak to trough, the cost per megabit went from over $3 to sixteen cents.

The damage went beyond spreadsheets. Korea's chaebols had funded their expansion with debt, and the top thirty conglomerates were running average debt-to-equity ratios above 500 percent. When semiconductor prices collapsed, the cash flow that was supposed to service those loans evaporated. Hanbo Steel went first, in January 1997. Sammi Steel followed. Then Kia Motors. By November 1997, the won was in free fall, the IMF was writing a $58.4 billion bailout package, and Korea's semiconductor miracle was being cited as one of the root causes of the crisis. Samsung survived, barely, by cutting costs and pushing into the next generation of chips while competitors bled out. But its stock lost more than 70 percent of its value from peak to trough.

The pattern is not that record profits cause a crash. The pattern is that record profits cause record investment, and record investment creates the supply that kills the price that justified the investment in the first place. The lag is typically eighteen to twenty-four months, the time it takes to build a fab and ramp production. Today, Samsung and SK Hynix together account for 60 percent of the KOSPI, up from 40 percent just two years ago. Goldman Sachs has warned that if the weighting rises another percentage point, index rules could force $2 billion in foreign outflows. The concentration is higher than anything in 1995, applied to a market cap that is orders of magnitude larger.

Smart money in late 1995 did not short memory stocks on the day of record earnings. That would have been early by a year. What smart money did was watch the capex announcements and count the fabs under construction, because those fabs were the supply that would arrive in eighteen months. The ones who timed the exit did it when fab utilization rates started ticking down from 100 percent, usually two or three quarters after peak pricing.

Every memory cycle in the last thirty years has ended the same way: record profits, record capex, then a supply wave that crashes prices 50 percent or more within two years. This time may genuinely be different because AI demand scales differently than PC demand, but no one making that argument today can tell you exactly how much new HBM capacity is coming online in 2027 and 2028. The edge is not in the direction. It is in watching the fab construction timelines and the utilization data, because those numbers have called the turn every single time.

The DRAM price-fixing lawsuit filed against Samsung, SK Hynix, and Micron alleges the three companies are restricting consumer memory supply to keep prices high while feeding AI customers first. Samsung paid a $300 million fine in 2005 for fixing DRAM prices between 1999 and 2002. If history is any guide, the legal overhang won't matter until profits start falling, and then it will matter very much.