Micron's Name Is in Idaho. Its Capability Lives in Taichung.
Notes from Supply Chained, 2026-09-08: why an 'American memory company' keeps its capacity and leading-edge nodes in Asia, why buying capacity and building capacity are opposite businesses, and how long a stopgap like HBM can hold. Educational, not investment advice; no tickers, no price targets.
Whatever you are told must be examined; passed along often enough, white becomes black and black becomes white. — Lüshi Chunqiu, “Examining What Is Transmitted” (Warring States period; translation mine)
What this episode is about
In the 2026-09-08 episode of Supply Chained, John of Asianometry talks with Tim Culpan of Culpium about Micron. Tim had just published a piece called “Micron and the myth of American memory,” arguing that Micron doesn’t really qualify as an American memory company. They warm up with a bad joke — John says he has a pain, Tim asks whether it’s a big pain or a micron-sized one — and end by agreeing to check back on all of it in 2035.
What stayed with me wasn’t the question of which flag the company flies. It was the gap underneath it: how far apart a company’s registered home and its actual capability can drift. That gap doesn’t show up on a financial statement.
The points worth keeping
One: the name is in Idaho, the assets are in Asia. Micron was born in Boise, funded by a potato magnate — Tim’s line is that he went from chips to chips. But since the late 1990s, its long-lived assets, meaning fabs and equipment, have mostly sat elsewhere. Taiwan is now the number one manufacturing hub, with Japan and Singapore high on the list.
Two: being behind was a sequencing decision. Micron adopted EUV later than SK Hynix and Samsung — a deliberate choice, on the view that they didn’t need it until a later node. When they did adopt it, the first went into Taichung, the second into Hiroshima. The first EUV node in the United States arrived only this year. Japan and Taichung run two nodes ahead of the US fabs, which works out to five or six years.
Three: Tim frames this as strategy, not as an insult. Samsung and SK Hynix expanded by building. Micron expanded by buying. When Texas Instruments couldn’t stomach the memory cycle and exited in the late 1990s, Micron picked up its stakes in a Singaporean and a Japanese company. Around 2010, as the Japanese and Taiwanese players fell apart, Micron swept in — the Japanese assets, the German stake from Qimonda, the Taiwanese capacity. That is how Taiwan became its production hub. Buying capacity beats building it, especially when the chips are down.
Four: it survived because it refused to die. John supplies the other half. Micron came out of design and backed into manufacturing; the founders were identical twin brothers, an engineer and a lawyer, and the lawyer served as CEO the longest. Memory looks like a commodity, but design carries enormous weight — one reason Taiwan’s own memory industry never worked is that the designs weren’t theirs. Micron’s process instinct was never to chase the next node but to squeeze the old one. Most other players were dabbling in DRAM on the side; TI couldn’t stomach the roller coaster and got out, which John calls a sound business decision. The difference is that Micron had nothing else.
Five: it nearly bought SK Hynix. Hynix was then an amalgam of two companies, carrying an LED line and a power chip line alongside the DRAM. Micron offered to take only the DRAM part, and the understanding at the time was that the point was to take that capacity off the market. The Korean administration wanted the deal done — the books looked bad. The company itself rebelled and chose to stick it out. Its creditors owned it for a decade.
Six: memory has one scoreboard, cost per bit. On Elpida, the verdict is blunt: its failure was a very Japanese failure. They took pride in yields up around 99%, but the yield percentage mattered less than the cost per bit. What Samsung mastered was driving that number down. The Japanese government put billions in, aimed at the wrong scoreboard.
Seven: there’s a chapter that never happened. Yukio Sakamoto, the elder statesman of Japan’s memory industry who died a few years ago, pushed for a Japanese-Taiwanese tie-up — a cross-border conglomerate called Taiwan Memory Co, partly funded by Taiwanese taxpayers. Politicians in Taipei asked why they should fund a foreign chipmaker. The rebuttal was that it wasn’t just a foreign chipmaker; it was Taiwan’s chip industry surviving alongside a strong partner against the Koreans. Tim says the business people on all sides were willing to do the deal, and the bureaucracies on both sides killed it. Underneath sat something harder to admit: a defeatism that Samsung simply couldn’t be beaten. John points out that the tone matches how people talk about Chinese manufacturing today.
Going further
”So which country is this company from?”
If you hold semiconductor exposure, this isn’t abstract. Tariffs, subsidies, export controls, geopolitical risk — each of them lands on a specific piece of ground.
The test I took from this episode: look at where the long-lived assets and the leading-edge nodes are, not where the headquarters is. Tim draws the distinction carefully. Apple, HP and Dell also manufacture overseas, but they keep core technology and design at home. Memory is different: only a few steps separate design from manufacture, and the coupling between designing a chip and ramping it is tight enough that engineering capability grows wherever the leading node runs. His most honest line is that Micron didn’t engineer this — over time, R&D and process engineering migrate toward the fabs on their own.
It comes with a condition that would break it: if the US nodes ever catch up to the Asian ones, the judgment needs rewriting. Until then, counting Micron’s capacity as American capacity is calling black white.
”They’re building new US fabs — is that good news?”
This is the part of the episode I’d pay for, and no headline will hand it to you.
Micron is expanding in Boise, expanding in Virginia (that fab was bought from Toshiba around 2002, and the first EUV node just went in there), and building in New York with a lot of money from the state. All three are moving. But Tim names the thing underneath: the next ten years are a different business from the last twenty-five. The old expansion bought capacity that already existed, at the moment when the owner hurt most — market share purchased at distressed prices. There is no one left to buy. Expanding now means building, which adds net new supply to the market, in a more expensive place, far from the ecosystem. His phrasing is restrained: the economics may not work out.
In my own words: the same company, the same verb “expand,” with opposite sources of return on either side of the line. The first version earned the discount on cyclical-trough assets. The second pays the premium on cyclical-peak construction. When an expansion headline crosses your screen, ask which one it is — bargain hunting, or added supply. The multiple you’d justify differs.
“AI is booming — isn’t memory a free ride?”
I replayed the closing stretch. John’s question is whether HBM is really the future. His description is that HBM is something of a kludge, the thing that happened to be sitting there when ChatGPT blew up. He isn’t alone: Pat Gelsinger has said publicly that it’s a silly idea and inefficient, and when SK Hynix was challenged on it, the answer was — you’re right, but it’s the best we have for now.
Between “the best we have for now” and “the long-term answer” sits a window nobody can measure. If something better arrives in the next ten years, it could nix the HBM piece, and that’s a problem for everybody.
Here’s how I handle this kind of setup, and I could be wrong: when a moat rests on “nothing better exists yet,” I treat the returns as rent rather than as an asset. Leases expire; you just don’t know the date. As for the constitution of this industry, the two of them repeat it three times — cost per bit. Yield, node generation, scale: all of it converts back into that one number before it counts.
Worth a look
- Supply Chained (John / Asianometry with Tim Culpan / Culpium), episode of 2026-09-08
- Tim Culpan’s piece, “Micron and the myth of American memory,” at Culpium
- For background: the mid-1980s memory price war and US-Japan trade friction, Elpida’s 2012 bankruptcy, and Yukio Sakamoto’s Taiwan Memory Co proposal — those three episodes are the bedrock under everything discussed here
One thing to take with you
Where a thing’s name is registered and where its capability lives are two different places — and capability grows wherever you spend your hours.
Micron’s name is in Idaho while two generations of leading-edge process run in Taichung and Hiroshima. Nobody deceived anybody; R&D and process engineering simply drift toward the fab. People work the same way. Your title sits on a business card. Your capability grew in whatever you actually spent last week doing.
Something I tried, and it takes under ten minutes: open last week’s calendar and write beside each block which capability that block was feeding. Leave it blank where you can’t answer. Then set that page next to the sentence you use when you introduce yourself to a stranger.
The first time I did it, three blocks matched my introduction, and close to half the page came back blank. That’s not there to feel bad about. It’s there to build next week’s calendar with.
This article is an educational discussion of investment method. It is not advice to buy or sell any individual security, offers no target prices, and does not analyze any current holding. Investing carries risk; make your own decisions or consult a qualified professional.