In short: Held — her memory position, sized up using Nintendo as the offset. Memory is one of the trades she front-ran: "buying all of the memory companies before everyone figured out that we were sold out for the next 4 years." Owning Nintendo "mentally… helps me justify a larger position in SK Hynix."
SK Hynix is a Korean maker of memory chips, including the high-end memory AI chips need. Wilde bought memory companies before the market realised supply was "sold out for the next 4 years" — one of her front-run-the-change trades.
She pairs it with Nintendo: Nintendo buys memory at whatever the current price is, so rising memory prices hurt it. Holding a bit of each softens the blow if she is wrong about memory, which lets her hold a bigger SK Hynix position.
In short: Subscriber view, passing: named with Micron as the memory suppliers that benefit "regardless of whether the winner is Nvidia, a hyperscaler, an AI lab, or some architecture we haven't seen yet."
13:22In this knife fight, you want to be the one selling the knives. Upstream semiconductor suppliers such as ASML, KLA, LAM Research, and Applied Materials, together with memory suppliers such as Micron and SK Hynix. They benefit regardless of whether the winner is Nvidia, a hyperscaler, an AI lab, or some architecture we haven't seen yet.
In short: Together with Samsung it is such a large weight in the KOSPI that the whisper-number miss dragged the whole Korean market into a ~40% drawdown — since largely recovered.
SK Hynix is Samsung's Korean rival in memory chips and the other giant weight in Korea's KOSPI index. She names the pair to explain the size of the damage: because those two companies dominate the index, one whisper-number miss pulled the entire Korean stock market into roughly a 40% drawdown (a fall from its peak), since largely recovered.
The lesson is about concentration risk, not Korean semiconductors. When a handful of names carry an index, a stumble at one of them is not a stock story — it's a market story. She points at Korea as the small-scale rehearsal for what over-concentration could do elsewhere.
29:23they fell in between those two and it caused a rout in the stock because of how big a weight both Samsung and SK Hynix are in the KOSPI the Korean stock market that had a drawdown to the tune of about 40%. We've since seen a recovery there, but that's what I think probably the next step will be. And I'm not suggesting it comes as soon as third quarter reporting season, but the next step would be dislocations that you start to see that are a little bit more at the individual stock level and you start to see more dispersion. We're
In short: Named as the downstream beneficiary of look-through lending: money that never asks what happens inside the data center "just leads to much more data center construction and far more GPUs and a lot more SK Hynix high bandwidth memory, NANDs and everything else." A demand pull he treats as financing-driven, not end-use-driven.
SK Hynix makes high-bandwidth memory (HBM), the fast memory that sits next to AI accelerators, plus NAND storage. Every new data center pulls more of it through.
Kedrosky mentions it as a consequence, not a pick. His point is where the demand comes from: lenders finance data centers by looking through the building to the hyperscaler on the lease, without much interest in what the machines inside are doing. That financing structure keeps construction going, which keeps orders for GPUs and HBM going.
So the demand is real but its origin is a credit decision rather than end-user consumption — which, on his framing, makes it vulnerable to the same funding-market shift that drives the rest of his argument.
15:45credit on the other side. It's a 12-year lease renewable and they're good for it and this is secure cash flow much better than holding a 10-year or anything else. And so I'll hold this and then at the end I look through whatever is going on inside the data center. And that of course just leads to much more data center construction and far more GPUs and a lot more SK Hynix high bandwidth memory, NANDs and everything else.
In short: The other half of the Korean AI-memory pair driving that index — peaked in June with it and down 30–35%; "the valuations of these stocks can't support a second derivative slowing."
SK Hynix is the other half of that Korean pair — the leading supplier of the high-bandwidth memory that sits alongside AI processors. Its shares and Samsung's together dominate the Korean index Dowd is watching.
Same read: the pair peaked in June and has fallen 30–35%. Because these companies are so directly geared to AI hardware orders, their decline is his cleanest evidence that the second derivative — the rate at which AI spending is growing — is turning down before it shows up in Western headlines.
11:49and it's down 30, 35% trying to rally. If that goes to new lows, I think we have the answer that the second derivative is shifting and the valuations of these stocks can't support a second derivative slowing. It's classic growth stock investing 101. I was a growth stock investor at BlackRock and in growth stock land, especially in very hyper growth stocks, second derivative does not bode well for valuations.
In short: The memory-cartel trade in its purest form: named alongside Samsung and Micron as holding the leverage over Nvidia on DRAM and NAND, then re-rated on its own capital return — "SK Hynix upgrades because of the sharp price increases in NAND and DRAM and the fact that SK Hynix announced a surprise 30 billion share buyback, caught the market off guard. JPMorgan has now kept an overweight on SK Hynix because of that massive buyback plan."
SK hynix is the Korean memory maker that, with Samsung and Micron, controls almost all global production of the chips that AI processors depend on. It is on the same side of the trade as Micron: Nvidia's decision to raise server prices rather than absorb memory costs is an admission that the memory makers, not the chip designer, currently hold the pricing power.
What re-rated the shares this week was capital return rather than the chip cycle. The company announced a surprise share buyback of 30 billion — buying back its own stock, which shrinks the share count and lifts earnings per remaining share. It caught the market off guard, and JPMorgan kept its positive rating specifically because of it. A buyback of that scale is also a management signal: it is what a board does when it believes its own shares are cheap relative to what is coming.
Full passage: premium transcript (PDF).
In short: Bought in the July capitulation "just for a trade, not for long-term" — memory names were changing hands at four times earnings after the Korean leveraged-ETF unwind. Also one of the gross-margin comps he watches into Nvidia's print: "the memory guys are 85% gross margins which is insane."
SK hynix is one of the three companies in the world that make the memory chips AI servers depend on. Korean retail investors had piled into leveraged funds tracking it; when the government clamped down, forced selling knocked the whole complex down 45% and left memory names trading at roughly four times earnings.
He bought that dislocation — but labelled it plainly as a trade, not a long-term holding, with an exit already scheduled ahead of November. He also watches its 85% gross margin as the comparison point for whether Nvidia's own margins can hold.
9:58So, we bought SK Hynix, we bought the DRAM ETF. We bought CoreWeave, we bought Nanya, and just for a trade, not for long-term. And so this end of July and August has been a really good backdrop for us, but I think we'll get cautious again going into November into the midterm elections because we feel after November that our president is not really going to hold back and we might see a re-escalation in Iran, we might see new tariffs announced because at that point
In short: Dean. Named alongside SanDisk as the memory complex that has fallen 40–50% in a month on ROI doubts — the setup for the whole conversation. He owns no semiconductors, and separately flags Chinese DRAM capacity as the reason.
1:28But gentlemen, how about we start off with a broad view of the hyperscaler spending on AI, the consequences and your thoughts, because I know you really think this is a very big deal. — Yeah, it's very timely. I think SanDisk, Skhinx, that entire memory complex and AI complex has been down I think 40, 50% over the last month and the market's really questioning if ROI is there anymore.
In short: Pitched twice in the same compilation. Artisan Partners (Emerging Markets Fund): "the company's key differentiator is its strength in high-bandwidth memory (HBM)," reinforced by "ongoing collaboration with NVIDIA on next-generation AI memory and continued investment in packaging and HBM testing capacity in Korea" — with an explicit caveat that "the pace and magnitude of recent memory price increases may prove difficult to sustain, particularly if elevated costs begin to weigh on end demand." Buffalo Funds (International Fund) makes the broader version: "memory is what makes agentic AI possible," and agent workloads "drive demand not just for HBM… but across the entire memory spectrum, including conventional DRAM and NAND for AI servers," with SK hynix's HBM "made to order and customized." Both funds name it a top contributor.
SK hynix is the Korean memory maker that leads in high-bandwidth memory, the stacked chips that sit next to every AI accelerator. Two separate funds in this compilation pitch it, which makes it the most-agreed name in the sample.
Artisan's version is about technical leadership: HBM is what lets AI chips process data quickly without wasting energy, and SK hynix is co-developing the next generation with Nvidia while spending on the packaging and testing capacity that stacking requires. They attach an honest caveat — memory prices have risen so fast that customers may eventually balk.
Buffalo's version is broader and simpler: memory is what makes agent-style AI possible at all, and agent workloads need not just the exotic stacked memory but ordinary memory and flash storage as well. SK hynix's particular advantage is that its HBM is built to order for each customer rather than sold as a commodity.
Full passage: premium transcript (PDF).
In short: The other half of the "dynamic duo." With Samsung it is the KOSPI's semiconductor sector — and it is down 40% over the last five weeks "despite ballistic earnings, leading to modest P/Es." Read as the transmission channel from the Korean blow-off into the US semis: "this is also when the U.S. Semiconductor Index, the SOXX, began its descent."
In short: Owned — flagged years ago by teammate Devon, who noticed after a visit to Nvidia's head office that "there's only one supplier" of high-bandwidth memory across the street. "I'm not in the memory bubble camp": an oligopoly of two or three suppliers, free-cash-flow yield in the high teens to low 20s, fabs that can't be thrown up quickly. The AI bubble, he says, is in "the fourth or fifth LLM model," not here.
SK Hynix makes memory chips, including the "high-bandwidth memory" that sits next to AI processors and feeds them data. The position came from process, not a hunch: his teammate Devon visited Nvidia's head office years ago, learned that when Nvidia needed this kind of memory there was literally one company it could phone, noticed that company had offices across the street, and said they had to buy it. A sole supplier to the fastest-growing customer in technology is, as Morrison puts it, "an awesome business."
He's blunt that he is "not in the memory bubble camp." Memory used to be brutally cyclical, but there are now only two or three credible suppliers — an oligopoly — and the shares throw off free cash flow equal to the high-teens to low-20s percent of their market value, which is extraordinarily cheap for a business in a boom. His discipline is to watch supply rather than guess demand, and supply can't respond quickly: chip fabs take years and the equipment queue at ASML and Lam Research is long. The bubble, in his view, is elsewhere in AI — "the fourth or fifth" large language model, which is already being commoditised.
50:50He said, "We got to buy this SKX because when I talked to Nvidia, there's only one supplier. That's an awesome business." So, I'm not in the memory bubble camp to be frank. — And this time is different. — I would not say this time is everything is cyclical. There will eventually be a moment in time where before I always say I try and gauge.
In short: Cited as an exhaustion signal, not a pick: its $27B equity raise last week (like SpaceX) is well-timed selling into the top — part of the shift from buybacks to "massive equity issues" that marks a market top.
SK Hynix is a giant Korean memory-chip maker, but Fred mentions it as a warning flag, not a pick. It just raised $27 billion in new stock last week. When companies rush to sell their own shares to the public (like SpaceX's IPO the same week), it usually means insiders think the timing is good — i.e. prices are near a top. The market has swung from companies buying back their shares to companies flooding the market with new shares, which is classic late-cycle behavior.
29:59We no longer have those buybacks. We have massive equity issues now. So we had SpaceX, right, with, it was the largest IPO ever. Two point two trillion dollar, was an $85 billion IPO. That's huge. We had SK Hynix just come last week, $27 billion of equity. We've had almost $200 billion of debt from the hyperscalers.
In short: Core of the three-player memory oligopoly — "providing the pickaxes in a gold rush." A protected, near-impossible-to-enter business (patents + "black magic" fab know-how; the three can't keep up with demand). But don't chase: "if you already own it, don't sell it; if you don't own it, don't buy it — the party has only just started." Regrets selling his own stake.
SK Hynix is one of only three companies in the world that make computer memory chips (the others are Samsung and Micron). Pabrai loves the business: memory used to be a brutal 20-player price war where nobody made money, but it consolidated down to three, and now it's nearly impossible for a fourth to break in — you'd need to violate patents, poach scarce senior engineers, and spend 10-20 years, and even then some of the manufacturing is "black magic" that can't easily be copied. With AI demand booming, the three can't make enough, so they're "selling the pickaxes in the gold rush."
But he's careful not to tell people to pile in now. His stance is: if you already own it, don't sell; if you don't own it, don't buy — "the party has only just started" but chasing it here isn't his style. He admits he broke his own never-sell rule by selling his memory stakes and regrets it.
16:46So when you look at a country like Korea, the indices are dominated by these two companies. They make up a large portion of the Cosby index — and it has been a wonderful ride for investors. So I would just say: if you already own it, don't sell it. But if you don't own it, don't buy it. — So now is not the time to buy.
Nothing matches this filter.
Verbatim excerpts from the public transcripts (auto-pulled at each mention's timestamp, lightly cleaned). Timestamps link into the video; "source page" opens that commentator's full analysis at this row. Click a mention's header line to expand it (one open at a time). For personal study — not investment advice.