The bigger the boom, the harder the (K-)pop
The AI gold rush discovered gravity
If you’ve been reading me for a while, you might remember this article:
In which I called the AI spending binge a “Gazillion-Dollar Oops”.
And if you’ve been even longer with me, you might remember:
And because the horse wasn’t dead enough, I kept beating on it in:
In which I reported on Anthropic’s 80x growth numbers, and a CEO admitting - on the record to Fortune - that growing only 5x a year would send them bankrupt.
Seems I keep being right about those things. Just way early… Some1 told me to develop a different hobby.
Anyway.
There’s been a development.
South Korea’s KOSPI hit an all-time high of 9,385 on June 19.
The world’s best-performing major index in 2026.
Up roughly 90% year-to-date. Nearly tripled in a year.
One small detail… Only two stocks drove roughly 70% of those gains:
Samsung Electronics & SK Hynix.
At the end of 2025, those two companies were about 34% of the KOSPI. Already way overweighted in my opinion, but the Koreans said: hold my soju…
By June 18, they were over 50%.
A whole frickin’ national stock market, concentrating itself into two chip factories over the course of six months. And on June 22 - the day before the crash - SK Hynix surpassed Samsung Electronics as Korea’s most valuable company for the first time in 26 years.
Make it make sense… I’m sure it does to someone.
Less than a week after the record, a Korean newspaper published twelve words:
“SK하이닉스, HBM4 생산 속도조절… ‘공급 부족’ 범용 D램 늘려 추가 수익 모색”
The KOSPI closed June 23 down 9.99% - 910 points - triggering double circuit breakers - COMEX is calling them to see how those things actually are supposed to work.
Oh, you don’t understand Korean (No1 doesn’t either!) so the translated headline is: “SK Hynix Adjusts HBM4 Production Speed… Seeking Additional Revenue by Increasing General-Purpose DRAM Amid Supply Shortages”.
BAM. 10% down.
The fifth-largest single-session fall in the index’s history, surpassing even the GFC volatility. Samsung went down 12.3%. SK Hynix 12.5%.
Foreign investors - the ones that were still left after the relentless dumping - dumped another $3 to 4 billion of Korean shares in that single session.
For what it’s worth, I also kept beating that dead horse (it’s practically steak tartare at this point) here:
Korea rang the bell.
Chosun Biz reported that SK Hynix found that commodity DRAM margins had risen sharply enough to sometimes beat HBM margins (HBM are specialized high-speed memory chips used in AI accelerators).
It’s only a rational business decision to follow the yield.
But the yield isn’t from “AI”. So it’s not hype, not fancy enough for the crowds… Hell, do they know it’s only borrowed money pooling in one corner of the supply chain long enough to create shortages in the next one over. Or that three years of circular capex has inflated everything adjacent to HBM.
So that company went to deprioritise AI chips and - I know … revolutionary concept! - actually try to make the most money possible.
Underneath all that there IS of course a demand question, if we forget the debt there for a second.
Nvidia’s next-generation Rubin platform - the one requiring HBM4 - is seeing lower-than-expected production forecasts. One word: China.
GLM-5 was trained earlier this year on 100,000 Huawei Ascend 910B chips. Zero Nvidia. Zero AMD. Zero AMERICAN silicon.
Woops
It scores within single digits of GPT-5.2 on major benchmarks. Frontier-class AI on domestic Chinese hardware, produced after China was placed on the US Entity List specifically to prevent this.
If you’re a Chinese hyperscaler, you won’t be buying Rubin chips. And if you’re not buying Rubin, you’re also not ordering the HBM4 those chips require. And if No1 is ordering HBM4 chips, SK Hynix has no incentive to produce them… I KNOW. Absolutely shocking. Demand affects supply. Write it down!
But those pesky export controls didn’t stop China’s progress on AI. They did what they always do: reroute past it. And with it a meaningful slice of global HBM demand went the way of the dodo.
But the fuel under the raging bonfire was the leverage.
By end of May, Korean retail investors held 38.5 trillion won - roughly $25 billion - in margin debt. A record. South Korea had just launched its first single-stock leveraged ETFs tied to Samsung and SK Hynix. The Financial Supervisory Service had expressed concern. Their concerns were bypassed after the age-old argument: “this time it’s different”.
The largest ETF: the Hong Kong-listed CSOP SK Hynix 2x Leveraged product, with $16.7 billion in assets. The world’s largest single-stock leveraged ETF. Anywhere. In the world.
(checking notes)
The FSS Governor publicly stated he “regrets not blocking the launch”.
(re-checking notes)
Right.
Water, meet bridge.
Moving on.
Now what might happen when a 2x product on a stock that is 26% of your national index starts forced selling? Asking for a friend…
The Korean-listed leveraged ETFs fell an average of 25.6% on a day the underlying stock was down 12.5%.
Goldman’s desk projects CTAs will dump in every scenario over the coming week.
A leverage unwind doing a convincing impression of a correction.
And while the circuit breakers were going off, Micron reported their earnings. The actual memory business, unadorned: just chips and customers and quarterly results.
Revenue: $41.5 billion. Expected: $35.8 billion. EPS: $25.11. Expected: $20.71. Gross margins: 84.9%. Q4 guidance: $50 billion, against analyst estimates of $43.6 billion.
A monster beat with a monster guide. Stock up 3% (and 15% after hours).
So.
Chip-buying is real. But three years of circular debt has to land somewhere, and it landed in Micron’s quarterly results - and according to them, next quarter it’s apparently going to land even harder.
The companies selling the shovels are gaining more gold than the actual gold diggers… A story as old as time I guess…
I’m not going to beat on that pulp anymore unless I find myself a director for a decent horror movie.
You know where I stand.
All that capex, that’s actually opex, sloshing around in a complete circle jerk. Sam’s “earning” money by riding the bubble while oil and silver are being punished for the one thing that AI can’t deliver: REALITY.
It’s a make-believe world.
“Slightly less spending” on AI chips topped the KOSPI. No1 saw that coming.
Yup... everything’s fine.
Nothing to see here.
Please disperse.
My other publications:
[Daily Digest] → The news in 5 minutes, without the forty open tabs.
[Portfolio] → What I do with my own money
One is emboldened to ask whether "the Street" is at long last catching up to what you've been reporting, along with Gary Marcus, et al...there seems to have been an industry-wide self-blindness that other countries, specifically China, can indeed competitively play in the AI sandbox...I mean, Deep Seek surely was an early-warning message, now all but forgotten. Competing on cost AND open-source potentially is a killer for super-capex exposed US AI frontier outfits...small wonder that OpenAI has reportedly shelved an IPO for later this year.
But AI can do this now ;-}
So a 'Jellyfish' Drone Swarm Might Have Shot Down Our F-15 Over Iran
“Multiple drones interconnected and moving as one with smaller drones below the bigger drones like legs,” shot down a U.S. Air Force F-15E strike fighter over Iran in April, according to a CNN source familiar with the apparently bizarre shootdown. “Real alien s***," they called it...
..A second source confirmed the almost unbelievable claim, telling CNN that the downed pilot described witnessing a “minefield of drones” in the air...
..While the pilot's name still hasn't been released, he's believed to be one of the same Eagle pilots accidentally shot down by Kuwait in a friendly fire incident earlier in Operation Epic Fury. According to the Washington Examiner, the unnamed pilot is "believed to be the first fixed-wing Air Force pilot since the Vietnam War era to survive being shot down twice in the same conflict."...
.."Had he witnessed a mature capability that US intelligence wasn’t aware of?" CNN's Zachary Cohen and Katie Bo Lillis asked. "A beta test? A mirage in the desert?" ...
..If Beijing also possesses some sort of drone-swarm antiaircraft capability, Epic Fury would have been a sweet opportunity to put it to a real-world test... Another option is that Tehran's homegrown drone technology is already advanced enough to destroy American fighters. If that's the case, the Pentagon needs to recalibrate how best to handle any future conflict. https://pjmedia.com/vodkapundit/2026/06/23/so-a-jellyfish-swarm-might-have-shot-down-f15-over-iran-n4954272