Newsletter · · Ashutosh Agarwal

SK Hynix Storms Nasdaq While Samsung's Record Gets Shrugged Off - HBM & The Memory Supercycle - Week of July 11, 2026

HBM & The Memory Supercycle, Issue #2 for the week of July 11, 2026. Samsung printed the most profitable quarter in its history and SK Hynix pulled off the largest first-time US listing by a foreign company ever, yet both stocks sold off in a sharp sell-the-news reversal, while SK Chairman Chey Tae-won said customers want five to six times more capacity and Q3 pricing guidance still points up.

HBM & The Memory Supercycle

Week of July 11, 2026: SK Hynix Storms Nasdaq While Samsung's Record Gets Shrugged Off

Issue #2 - Week ending Saturday, July 11, 2026

TL;DR

  • The stock market and the memory bosses are telling two completely different stories. Samsung just posted the most profitable quarter in its history, profit up roughly nineteen times from a year ago, and the stock fell. Micron and SK Hynix also sold off hard even as prices keep ripping. This was the week the "sell the news" reflex met the "we are sold out for years" reality, head-on. (The Rundown, Jul 7; Bloomberg Intelligence, Jul 7)
  • SK Hynix put a price on the whole trade. It listed shares in the US for the first time, priced at $149, and opened around $177, a ~19% first-day pop. The book was over seven times oversubscribed, north of $160 billion of demand. It is the largest first-time US share sale by a foreign company, ever. (Bloomberg Tech, Jul 9)
  • The clearest, most credible voice all week was an insider. SK Group Chairman Chey Tae-won sat down with Bloomberg on listing day and said the quiet part out loud: customers are asking him to build not two times but "five times, six times" more capacity, and he thinks supply "is never going to catch up" until society reaches "some settlement with AGI." Take that with the salt it deserves, he was ringing the Nasdaq bell that morning, but it is the operator, not a pundit, saying it. (Bloomberg Tech, Jul 10)
  • Pricing is still going up, just a little less violently. Fresh third-quarter guidance points to DRAM contract prices rising another 13–18% and NAND another 10–15%, on top of last quarter's ~60% moves. Samsung has reportedly told some customers it wants another 20% on DRAM. (Mac OS Ken, Jul 6)

What's New

This was a two-headline week, Samsung's earnings on Monday–Tuesday and SK Hynix's US listing on Friday, and almost everything else was reaction to those two events. I've ranked the items by how much they should actually move your thinking, not by how loud they were.

1. An IDM boss finally talks, and he is not hedging. The single most valuable thing on the podcasts this week was an operator interview, not analyst chatter. On listing day, SK Group Chairman Chey Tae-won joined Bloomberg's Ed Ludlow live from the Nasdaq (Bloomberg Tech, Jul 10).

The headline number: SK plans to invest close to $1 trillion over the next ten years in AI data centers, targeting "about 15 gigawatt in Korea" and roughly "5 gigawatt outside of Korea." (A gigawatt is the power draw of a very large data-center campus, so this is enormous.)

On capacity, he was blunt about how far behind demand still is:

"So we just announced that a few months ago, so we're going to double up our capacity within five years... Well, people see that that's unrealistic and probably the oversupply and people worry about it. But my customers said that that's not enough. We need it more. So double up is not... It's not enough, actually. Well, they want us that we're closing up five times, six times."

And on how long the shortage lasts:

"I have some confidence that the demand will grow. And our supply capacity is never going to catch up... I think it's going to be until that our society, human society has said there was some settlement with AGI. Then... the world does kind of normalize all the cycles. But anyhow... Until then, well, we need that a lot of memories."

Why it matters: This is the man who runs the world's #1 high-bandwidth-memory maker saying, on the record, that even a doubling of capacity is not enough for his customers. He also framed the long-term contracts that have de-risked this cycle as customer-driven, not vendor-driven: "there is a supply shortage and every customer wants to... secure their own supply... they're asking us that we need some kind of the long-term contracts. It's not suggested by us." He floated a new idea, "memory as a service," arguing memory is "not a cyclical business anymore" once you lock in multi-year volumes and price floors. That last claim is exactly what the bears think is management talking its own book. But it's the freshest operator signal we've had in two weeks, and it directly fills a gap I flagged in Issue #1. (One honest caveat: he was interviewed the morning his company debuted on the Nasdaq. Nobody rings the bell and then tells you the party's ending. Weigh it accordingly.)

2. Samsung printed a record, and the market yawned. Samsung reported one of the most lopsided quarters you will ever see. Per The Rundown, Jul 7, host Zaid Admani laid out the numbers: operating profit of about 89.4 trillion won (~$58 billion), up 56% from the prior quarter and roughly 19x year-over-year, on record revenue of about $112 billion that "more than doubled from a year ago." He cited Bloomberg data that DRAM prices "rose more than 40% last quarter, while NAND prices jumped more than 50%."

The reaction? The stock fell about 7% in Korea, dragging Micron down ~8% and Intel down ~10% in sympathy. Admani's read: "we've hit the point now where all the upside in these chip stocks are already priced in," and Samsung only beat estimates by about 6% while planning "around $70 billion in CapEx this year alone," which "could mean that the memory supply shortage today could turn into a supply glut."

Bloomberg's Ed Ludlow put the cleanest frame on the paradox (Bloomberg Intelligence, Jul 7):

"Numbers good. You know, revenue more than doubled, profit 19-fold. And what's so interesting about it is fundamentally, very deliberate choice of word, nothing in the memory markets changed from when you and I went to bed last night... there is still tight supply pricing of DRAM and NAND... and Samsung has no line of sight to when that improves. It's really a stock story where Samsung was a stock that was up 150% going into it."

Why it matters: When a company prints its best-ever quarter and the stock drops, the market is telling you positioning, not fundamentals, is now the swing factor. That's a different regime than three months ago, when any good print sent these names straight up.

3. SK Hynix's listing became the market's referendum on memory. SK Hynix listed ADRs (American Depositary Receipts, a way for US investors to buy a foreign stock without touching the Korean market) on the Nasdaq. Per Bloomberg Tech, Jul 9, it priced at $149 per ADR, a ~3% premium to the Korean close, raising $26–28 billion. Bloomberg's reporter noted the book was "over seven times oversubscribed... nominally that's north of $160 billion in demand," making it "the biggest first-time US share sale by a foreign entity," outpacing Alibaba and behind only Saudi Aramco's domestic listing. On listing day the ADRs opened around $177 versus the $149 price, roughly a 19% pop (Chairman Chey confirmed the open in the Bloomberg interview above).

Bloomberg's Ian King gave the investor frame: revenue "tripling to more than $200 billion this year at an 80% gross margin." That is a staggering margin for a business that, as King reminded listeners, "came very close to being bought by its competitor, Micron" a few years ago.

Why it matters: For the first time, US institutions and index funds can own the world's HBM leader directly. Watch whether the ADR premium to the Korean shares widens from that 3% toward double digits, that's a live read on Western appetite for the trade.

4. The next leg of pricing is guided, and it's still up. The most concrete pricing datapoint came, oddly, from an Apple podcast. Ken Ray aggregated the latest TrendForce guidance (Mac OS Ken, Jul 6): DRAM contract prices are forecast to rise 13–18% in Q3, and NAND flash 10–15%, on top of last quarter's increases "of as much as 63%" for DRAM and "as much as 60%" for NAND. Separately, the Korea Herald reported Samsung "plans to raise DRAM prices 20% this quarter." TrendForce's warning: "price tolerance among consumer customers has reached its limit."

Why it matters: Prices are still rising, but the rate is decelerating (mid-teens vs. ~60%). That is the first number the glut-worriers will point to, and the first data the bulls will dismiss as "still up and to the right."

5. Micron raises the capex bar to $250 billion. Buried in the SK Hynix coverage: Micron is boosting planned US spending to $250 billion from $200 billion (Bloomberg Tech, Jul 9), and separately signed a "massive upstate New York facility" deal (The Compound and Friends, Jul 10). Ian King's caution is the one to keep taped to your monitor: "this is an industry that's never got it right over the long term... if we see a massive flood of that $250 billion getting deployed in the U.S. very quickly, then that is something that investors will pay close attention to."

The Debate: Structural Shortage, or a Cycle About to Roll?

This week sharpened both sides. Here's the honest steel-man of each.

The structural / "this time is different" case:

  • Demand is a black hole. The Limitless hosts framed it well (Limitless: An AI Podcast, Jul 9): every AI model rereads its full weights on every prompt, and "every model... 10 to 20x is the requirement of memory," while chat memory and KV cache pile onto the NAND side. Bloomberg Intelligence's Mandeep Singh made the same point from the sell-side chair: "longer context... for all these models has been a big driver of KV cache and HBM demand and that seems to be something that will stay for a while... I don't see how that changes until you've got a new player that can add more supply, which we know is not going to happen anytime soon."
  • Supply physically can't respond fast. HBM cannibalizes normal memory: "one gigabyte of HBM consumes the factory capacity of roughly four gigabytes of regular DRAM," per Limitless, so every AI wafer is a wafer that never becomes a laptop. New fabs don't arrive "until late 27" at the earliest (Wedbush's Matt Bryson, below), and the Limitless hosts argued meaningful new supply won't land until closer to 2030.
  • The contracts changed the game. Multi-year customer agreements with "relatively stark" cancellation penalties now give vendors "visibility into longer-term cash flow" for the first time, Wedbush's Matt Bryson on Closing Bell, Jul 9, noting Micron has said these will be "50% of their sales," so "we could be looking at five years of strong earnings, even if only half of their volume is set at current pricing."
  • The moat is real. Mandeep Singh: the Koreans "manage their own fabs. That's been a huge moat, which is why we don't hear about new entrants getting into the memory market," backed by talk of ~$880 billion of Korean sovereign support "very similar to how TSMC built its... moat over time."

The cyclical / "we've seen this movie" case:

  • Margins always revert. Ed Ludlow's warning is the whole bear case in three sentences: "Historically, memory is a very cyclical market, boom and bust... in those periods where the memory names have had very high margins, they've never been maintained. They ebb and flow, they crash to a floor. And I think everyone might be a little bit worried about that."
  • The chart looks toppy. Guy Adami on RiskReversal Pod, Jul 8: Micron ran from ~$300 in March to over $1,000 by early June "pretty much a straight line," and now trades in daily 8–10% up-then-down swings, "not symbolic of something that's healthy... this is typically the churn you see towards the... top."
  • The buyers of the buildout are rolling over first, the 1999 tell. This was the sharpest bear argument of the week, from Michael Cembalest of JPMorgan on The Compound and Friends. He drew the Cisco/telecom analogy: in 1999 the infrastructure stocks kept zooming while the companies paying for the infrastructure quietly rolled over. "The last three months, the hyperscalers are starting to roll over even though the semiconductor index is zooming... if the big four hyperscalers plus Oracle can't convince the market that they're earning a good return on a trillion-five of capital spending outlays, those infrastructure stocks are at risk at some point. So I'm starting to feel a little nervous about where we are right now." (The Compound and Friends, Jul 10)
  • The financialization worry. Dan Nathan (RiskReversal) flagged how the buildout is increasingly debt-funded, Amazon's $25 billion bond deal, Blackstone/Apollo's ~$35 billion private-credit vehicle to buy chips, echoing a Jim Chanos-circulated note calling it "a credit-driven real estate cycle hidden in tech clothing." (RiskReversal Pod, Jul 8)
  • Supply is coming, on management's own timeline. Samsung's ~$70 billion and Micron's $250 billion capex plans are exactly what turns today's shortage into tomorrow's glut. And even Chairman Chey conceded, in the same breath as his bullish comments, that "nobody expected... I don't know what's going to happen in next years."

My read: The near-term price action is positioning, not fundamentals, record prints selling off is a classic late-stage "everyone already owns it" signal, not evidence the shortage is ending. The fundamental clock is set by two things nobody can fake: when new fabs actually come online (late 2027 at the earliest) and whether hyperscaler capex holds. Cembalest's hyperscaler-rollover point is the one genuinely new bear argument this week and the one I'd watch hardest into the coming earnings season. If Microsoft, Meta, Amazon and Alphabet reaffirm capex, the structural case holds; if any of them blink the way Meta did last week, the whole hardware complex gets a demand scare regardless of how "sold out" 2026 is.

Stocks in Play

Every name that got discussed by ticker this week, with the bull, the bear, and the next thing to watch.

Micron (MU). Bull: The "cleanest" way to play memory, per Futurum's Daniel Newman (Futurum Equities Podcast, Jul 8), who ranks it above SK Hynix and Samsung because "shock risk is the lowest, there's no tariff risk, it's U.S.-based." Pricing power is extraordinary: DRAM average selling prices up ~60% year-over-year and NAND up ~80%, plus 16 long-term agreements worth ~$100 billion signed last quarter with $22 billion in cash deposits (Stephanie Link, Money Rehab, Jul 6). Gross margin "more than doubled to almost 85% from 39%" a year ago (Mac OS Ken). Newman notes Micron trades around 6x earnings, "you really have to believe the cycle is peaking this year" to be bearish here. Bear: Now a >$1 trillion company that went straight up and is chopping violently; "$200 below its highs even with its bounce" (Matt Bryson, Closing Bell). Cembalest flagged that Micron is "now bigger than Meta, or right there," a valuation the market historically never lets a cyclical hold. Watch: Whether it holds the 50-day moving average around the high-$800s (Futurum's Shai Belor was eyeing ~$880 as a re-entry). Bigger picture: pace of that $250 billion capex deployment.

SK Hynix (000660 KS / new US ADRs). Bull: The HBM leader at 57% market share (Mandeep Singh, Bloomberg Intelligence, Jul 10), NVIDIA's most strategic supplier, ~$200 billion revenue at 80% gross margin. The listing was 7x oversubscribed and popped ~19%. Historically cheap at a 6–8x forward P/E versus Micron's ~14x (Brew Markets, Jul 6). Bear: Newman "doesn't love the ADR nature of it," and both Futurum hosts flagged something "wonky" and "dangerous" in the Korean market, reports of retail investors putting retirement savings into 2–3x leveraged memory plays. And chasing a listing that's "3x, 4x oversubscribed" can leave you a "bag holder" if you buy the pop. Watch: Does the ADR premium to the Seoul shares widen past ~3%? And how large does that "$35 billion and going to get much, much, much bigger" US investment (Ed Ludlow) actually get?

Samsung Electronics (005930 KS). Bull: Record profit (~$58 billion operating, +19x y/y); on track to "make more money in a year than it made in the last 19 years" (Futurum). #2 in HBM and ramping HBM4, plus optionality from custom AI chips, reportedly "in discussions with Anthropic about building specialized AI processors" (AI Chat, Jul 7). Newman says Samsung has "overcome a lot of execution risk." Bear: Ranked last of the three by Futurum. Its ~$70 billion capex plan is the single biggest glut risk. And its guidance was called "opaque at best," which RiskReversal's Guy Adami tied directly to the sell-off in equipment name Applied Materials. Watch: HBM4 qualification progress at NVIDIA (notably absent from this week's podcasts) and whether the "opaque" guidance sharpens on the next call.

SanDisk (SNDK). Bull: The week's most jaw-dropping numbers. Per Stock Club, Jul 9: a $42 billion backlog, gross margins at 78% (up from 23% a year ago), revenue of $6 billion in its latest quarter versus $1.7 billion a year earlier (+250% y/y), and operating income that went "from $2 million last year to $4.2 billion this year." Wedbush raised its price target to $2,000 from $1,200, targeting 84–85% gross margins by 2027 (Closing Bell). Up roughly eightfold year-to-date. Bear: "100% peak, peak, peak at the cycle," in Stock Club's own words. The hyperscaler customers "don't want SanDisk to have 78% gross margins" and will hunt for alternatives. This is the most cyclical, most beta-heavy way to own the theme. Watch: NAND contract pricing direction (guided +10–15% for Q3) and any sign the backlog stops growing.

The hyperscaler buyers, NVIDIA (NVDA), AMD, Microsoft (MSFT), Meta (META), Amazon (AMZN), Alphabet (GOOGL). Bull: All four hyperscalers are "trading at least 10% below their 52-week highs," down 35–40% on a multiple basis, "de-risked, and ultimately they control the compute," iCapital's Shonali Bassik and Cantor's Eric Johnston on Closing Bell see them as the relative-value trade into earnings. On NVIDIA specifically, it trades ~14x forward estimates for a company growing ~50% (Stephanie Link, Money Rehab), "crazy cheap for what you're getting." Bear: Cembalest's rollover warning, and Meta's capex-capping hint from last week still hangs over the group. Zuckerberg said Meta will "aggressively price" its AI model, a commoditization signal. Watch: Upcoming hyperscaler earnings. As Johnston put it, "the most important part of all of earnings season is going to be their commentary on CapEx." Any hint of discipline is a direct demand signal for HBM.

Read-Throughs

  • Memory equipment (Advantest, BESI, Camtek, KLA, Lam, Applied Materials). Thin week for hard data, and mostly negative price action. Applied Materials (AMAT) fell from ~$740 a few weeks ago to ~$550, which Guy Adami tied directly to Samsung's "opaque" guidance, "you would think they'd have much more clarity than they do" (RiskReversal). KLA (KLAC) and Lam Research (LRCX) moved with it. The bull counter came from Stephanie Link, who owns Lam, AMAT and KLA and calls them long-term winners because "it doesn't matter who wins, all of these players need the cap-equipment companies," advising buyers to "wait for a pullback." Net: the equipment names are trading as a leveraged proxy for memory-maker guidance confidence, when the IDMs sound vague, equipment gets hit hardest. No specific commentary this week on Advantest (test), BESI (hybrid bonding) or Camtek (inspection).
  • Packaging / substrates (CoWoS, hybrid bonding). Essentially absent. The one relevant nugget: Chairman Chey noted SK Hynix works "with TSMC on the base die, which the others do not do," a reminder that the HBM-to-logic packaging relationship is a competitive edge, but there was no dedicated CoWoS or hybrid-bonding discussion this week.
  • GPU makers (NVIDIA, AMD). Both are the pull-through demand for HBM. Matt Bryson's line is the read-through that matters: NVIDIA's edge isn't just the chip, it's that "they have the best supply chain in the tech world... you can't get enough memory, you can't get enough substrate, you can't get enough logic, and NVIDIA's in a better position than anyone else." Memory scarcity is, paradoxically, a moat for whoever has locked up allocation.
  • PC / handset OEMs facing rising memory costs. This is where the supercycle stops being abstract. Apple raised prices "across its Macs, iPads, and more" by hundreds of dollars, citing memory and storage shortages (Business of Tech, Jul 8). Concrete numbers from Limitless: MacBook Air $1,100 to $1,300; MacBook Pro $1,700 to $2,000; Mac Studio $4,000 to $5,300; a 32GB RAM kit is now 2–3x last year's price, and memory is "over a third of the cost" of a self-built PC. Dave Sobel cited a common 8GB DRAM module going "from around $35 to roughly $300, nearly 9 times over," and flagged that the data-center buildout is set to pull "15–20% of all consumer memory manufacturing capability... into data centers in 2027." TrendForce expects smartphone brands to raise retail prices in Q3, which "could weigh on handset sales," the first genuine demand-destruction feedback loop worth tracking. IDC's Nabila Popal called it "an absolute existential crisis" for sub-$100 Android makers, who "won't be able to get the memory because memory suppliers are only answering calls of big players."

What Changed vs. Last Week

A lot, actually, this was a much busier week than Issue #1.

  • Last week's story was Micron's blowout and Meta's capex scare (the Thursday July 2 sell-off that hit KLA, Micron, SanDisk and AMAT). This week the catalyst shifted to Samsung's record print and SK Hynix's listing, and crucially, the reaction pattern flipped from "buy any good news" to "sell the news," with record numbers greeted by 7–15% single-day drops before chips bounced back mid-week.
  • A gap I flagged last week is now filled: Issue #1 noted "no direct IDM exec interview." This week we got Chairman Chey Tae-won on the record, the highest-quality operator signal we've had.
  • New pricing data: TrendForce's Q3 guide (DRAM +13–18%, NAND +10–15%) and Samsung's reported +20% DRAM plan give us fresh, if decelerating, contract-price direction.
  • New capex bar: Micron $200 billion to $250 billion.
  • New price target: Wedbush took SanDisk to $2,000 from $1,200.
  • New bear framework: Cembalest's hyperscaler-rollover / 1999-Cisco analogy is the most substantive fresh argument on the skeptical side.
  • Still quiet on the technical core: last week I noted no HBM3E/HBM4 qualification or yield update. Same this week, HBM4 came up only as "sells for more," never as a qual-or-yield story.