"History doesn't repeat… but it rhymes." — Mark Twain
◉ THE PRESENT
ChangXin Memory Technologies debuted on the Shanghai STAR Market yesterday and the stock didn't rise — it detonated. A 466 percent first-day gain vaulted CXMT past Industrial and Commercial Bank of China to become the most valuable listed company on the mainland, worth $488 billion. Memory stocks bled: Micron fell 5 percent, SanDisk cratered 12 percent, SK Hynix shed 9 percent ahead of its Q2 earnings today. The market is doing the math on what happens when a state-backed challenger with $8.6 billion in fresh capital starts building fabs in a sector with a long history of destroying itself through oversupply. This math has been done before.
CXMT mkt cap $488B | CXMT debut +466% | IPO raised $8.6B | MU −5% | SNDK −12% | SMH −3% | CXMT DRAM share 8% | MU peak-to-trough −31% since June 25
◉ THE ECHO — OCTOBER 23, 2014
A dying man walks into a room in Tokyo and bets four hundred million dollars on a chip.
Lee Byung-chul was seventy-three years old and terminally ill. He had built Samsung from a noodle-trading company into one of Korea's largest conglomerates, but on this February morning in Tokyo he announced that Samsung would enter the semiconductor business. Not peripherals. DRAM — the same memory chips that Japan's NEC, Toshiba, and Hitachi had spent a decade perfecting and that American firms like Intel would soon abandon.
The announcement was later called the Tokyo Declaration. It was made in exactly the city where it would be most easily dismissed. Japan controlled 66 percent of the global 64K DRAM market. Samsung had no chip engineers, no fabs, no customers. Lee committed four hundred million dollars against the unanimous opposition of his executive team. Nobody in Tokyo noticed.
Ten months later, Samsung produced its first chip — a 64-kilobit DRAM, primitive by Japanese standards. But the catch-up was faster than anyone modeled. Samsung released a 1-megabit DRAM in 1986, a 4-megabit in 1988, a 16-megabit in 1990, each one closing the gap. By 1992 Samsung's 64-megabit DRAM arrived six months ahead of the Japanese designs. In 1993 Samsung passed Toshiba to become the world's largest memory maker, producing more DRAM revenue than any individual Japanese competitor.
The Korean government had helped — a joint development program, favorable lending, industrial policy that treated chips the way other nations treated defense. Hyundai and LG built their own fabs. Within a decade Korea went from zero to dominant, and Japan's chip industry began a collapse that ended with the bankruptcy of Elpida Memory in 2012 — the last Japanese DRAM maker — and its sale to Micron.
But before Japan's exit, something else happened. Everyone built at the same time. More than fifty fabs were announced during 1995 and 1996. DRAM prices peaked in late 1995 and collapsed — falling 51 percent in 1996, another 65 percent in 1997. A megabit that cost three dollars in 1995 sold for sixteen cents by 1998. Micron's stock, which had run roughly five-fold in barely a year, crashed 82 percent between September 1995 and July 1996. The S&P 500 gained roughly 13 percent over that span. If you owned memory, you were ruined.
◉ THE RHYME — WHAT'S IDENTICAL
A state-backed challenger enters the memory market with cheap capital, catches up faster than anyone expects, and the incumbents' stock prices crack before the pricing does. The rhyme is clean.
◉ THE DIVERGENCE — WHAT'S DIFFERENT THIS TIME
CXMT is behind, not ahead. When Samsung took the crown in 1993, its 64-megabit DRAM was six months ahead of the Japanese competition. CXMT is roughly three to four years behind SK Hynix and Samsung in high-bandwidth memory — the chips that matter most for AI. It's manufacturing second-generation HBM, equivalent to what Korea was shipping in 2016. The gap is narrowing, but it's real.
Three incumbents, not twenty. In the mid-1990s, more than twenty companies made DRAM. Today there are three — Samsung, SK Hynix, and Micron — and they learned from the last cycle. The oligopoly is supposed to prevent the kind of capacity race that produced the 1996 crash. CXMT's 8 percent share is meaningful but not enough to break pricing alone. The question is whether its arrival makes the Big Three lose discipline.
Export controls create a ceiling that didn't exist. Samsung had access to every tool the industry offered. CXMT is building fabs with stockpiled deep-ultraviolet machines because the U.S. has blocked its access to extreme-ultraviolet lithography. It can flood commodity DRAM but may struggle in the highest-margin segments where the real money is being made.
The demand driver is different. The 1990s memory boom was powered by PCs. Today's boom is powered by AI infrastructure spending — data centers are buying HBM faster than anyone can make it. If AI demand stays strong enough, it might absorb the extra supply before pricing cracks. That's the bull case. The bear case is that it never has before.
◉ THE RECKONING — WHAT HAPPENS NEXT
The crash didn't come from Samsung alone. It came from everyone building at once. Samsung, Hyundai, LG Semicon — all expanding. Micron was spending 2.5 billion dollars on a new Utah plant even as its stock was falling. Capex as a share of semiconductor production exceeded 30 percent. The result was mechanical.
Micron survived. Its revenue actually grew in fiscal 1996 even as prices cratered — volume offset some of the pricing damage. But the stock didn't care about volume. Between September 1995 and July 1996, Micron lost 82 percent of its value while the S&P returned roughly 13 percent. Owning the right sector mattered more than owning the right market.
The Japanese firms didn't survive. NEC and Hitachi combined their DRAM operations into what became Elpida. LG Semicon and Hyundai Electronics were forcibly combined during the Asian Financial Crisis, later renamed Hynix. Elpida filed for bankruptcy in 2012 — the largest manufacturing bankruptcy in Japanese history. Micron bought its assets for roughly two billion dollars, a fraction of what one advanced fab costs today.
The pattern: the challenger's arrival doesn't crash the market by itself. What crashes the market is the incumbents' reaction — the panic-building, the refusal to cut production, the faith that demand will absorb every new wafer. Micron has guided fiscal 2026 capex above $25 billion. SK Hynix is spending roughly $27 billion. Samsung is expanding aggressively. If those numbers go higher while CXMT's fabs are ramping, the 1996 script is back in play.
The edge: In 1995, Micron's stock peaked not when DRAM prices peaked, but weeks before. The stock saw the oversupply coming while the earnings were still good. Micron just reported record revenue of $41.5 billion in its fiscal Q3 — a 346 percent increase. The earnings look perfect. They looked perfect in September 1995 too.
◉ TOMORROW’S WATCH
SK Hynix reports Q2 earnings today, and any softness in DRAM pricing guidance will echo the analyst downgrade that sent Micron into freefall in November 1995. Wednesday brings the Fed decision — and the question of whether a central bank will choose to cut rates into a semiconductor downturn.
