Newsletter · 2026-10-03 · Ashutosh Agarwal

Micron Guides to a Tighter 2027 and 2028 as Investors Still Shrug - HBM & The Memory Supercycle - Week of October 3, 2026

HBM & The Memory Supercycle for the week of October 3, 2026. Podcast synthesis on Micron's record fiscal Q4 beat and above-consensus guide, CEO Sanjay Mehrotra's call that memory supply stays tighter in 2027 and 2028 than in 2026, why the stock slipped anyway, the December 9 buyback lockup, China's CXMT and YMTC capacity push, slowing but still-rising DRAM prices, and chip designers trying to route around HBM.

HBM & The Memory Supercycle

Week of October 3, 2026: Micron Guides to a Tighter 2027 and 2028 as Investors Still Shrug

Micron just posted one of the best quarters any chip company has ever reported. Revenue nearly 5x in a year. A gross margin of 87%, higher than NVIDIA's. Its CEO went on television the next morning and said next year and the year after will be even tighter than this one.

The stock went down.

That gap is the whole story this week. On the numbers the shortage keeps getting worse. Investors are still pricing Micron as if the party ends on a known date. Nobody in the podcasts this week argued that demand is cracking. The fight is about how long this lasts, and whether the market will ever pay more than about 6x earnings for a memory company.

TL;DR

• Micron beat and raised, and the stock barely moved. Fiscal Q4 revenue was $54.23B vs. $51.0B expected, EPS $33.42 vs. $31.61, and gross margin 87% vs. 86.3%. Next-quarter revenue guidance is $60–63B, against roughly $56.8B expected. The stock still slipped 2–3% the next morning (Closing Bell, 9/30; Schwab Network, 9/30; The Rundown, 10/1).
• The CEO extended the outlook to 2028. Sanjay Mehrotra said supply is "tight in 27 and 28, even tighter than 26." Micron now has 26 long-term customer agreements, up from 16 in August, with some running into 2031. About 75% of 2027 supply is already spoken for (Squawk on the Street, 10/1; The AI Investor Podcast, 10/1).
• The next catalyst is a buyback, not a chip. Micron's CHIPS Act lockup on buybacks ends December 9. Sell-side analysts on the podcasts talked about buying back 8–10% of the company a year, and one said "15% a year, easy." China came back into the bear case too: CXMT is adding 100,000 wafers a month of capacity by end-2027 (Closing Bell, 10/1; Squawk on the Street, 10/1; The Circuit, 9/28).

What's new

Ranked by how much each one matters for a position. Labels: [INSIDER] means a company executive speaking on the record. [OPERATOR] means someone who builds chips or systems but is not a memory maker. [SELL-SIDE] means a bank or research-firm analyst. [PUNDIT] means a host, investor or commentator.

1. Micron's CEO, on the record: "We cannot fulfill the demand of our customers" [INSIDER]

Podcast: Squawk on the Street, "9AM HOUR: Playbook for Q4 and October, Micron CEO Exclusive, Bond Sell-off Accelerates 10/1/26" Speaker: Sanjay Mehrotra, Chairman, President and CEO of Micron, interviewed by Jim Cramer

This is the most important source of the week, and the first time a memory-maker CEO has given hard numbers on these podcasts in weeks. (Micron's Jeremy Werner in mid-September and Mehrotra in August talked vision and factories, not numbers.)

Long-term contracts jumped in six weeks. Cramer pointed out that Micron had 16 "strategic customer agreements" when he visited Boise on August 20 and now has 26. These are multi-year supply contracts. Many are "take-or-pay," which means the customer pays whether or not it takes delivery. Mehrotra's answer:

"Even the customers where we have already strategic customer agreements in place, they are bringing upsides to us for demand. In fact, we cannot fulfill the demand of our customers. Customers are extending these agreements. Even now, into 2031 timeframe."

The outlook now runs to 2028. Asked about buybacks, he said:

"We see going forward in an environment where supply is tight in 27 and 28, even tighter than 26, we see an environment of continued, strong, durable, predictable financial performance."

Capex: about $25B in the first half of fiscal 2027 alone. Capex means capital spending, mostly on new factories ("fabs").

"In the first half of our fiscal 27 ... CapEx, mostly going into construction, and of course, bringing up supply in a disciplined manner will be around $25 billion. That's almost as much as what we invested in full fiscal year 26."

Cash and buybacks are on hold until December 9. He said Micron expects "well over $100 billion in cash, by the end of our current fiscal quarter." Asked about buybacks, he said only "stay tuned" and that details come "after December 9," the second anniversary of its CHIPS Act award.

Why margins dipped. Next quarter's gross margin guide is 86.25%, just below this quarter's 87%. Gross margin is the share of revenue left after the direct cost of making the chips. Mehrotra put the dip on employee pay: "Very pleased and proud to be able to incentivize our team members in line with our record performance."

Why it matters: The bear case on Micron has always been "2028 is when new factories come online and pricing breaks." The CEO went straight at that year and said it will be tighter, not looser. That does not make him right. But a CEO who has signed 10 new multi-year contracts in six weeks is a different source from one making a forecast.

Cramer's own read, which is opinion: "the long-term agreements that they have are really not, are a floor for what they can earn. And a lot of people, I think, think they're a ceiling, and that's a misread."

2. The numbers, and why the market shrugged [SELL-SIDE / PUNDIT]

Podcasts: Closing Bell, "Markets Close Out September With AI in Focus 9/30/26"; Schwab Network, "EARNINGS ALERT: MU"; The Rundown, "Micron Posts Monster Quarter, Apple Readies Smart Home Devices"; The AI Investor Podcast, "Google Strikes Back With Argon, Where Memory Stocks Go Next, And Viewer Q & A"

The scorecard for Micron's fiscal Q4 (the quarter ending around late August 2026), from CNBC's Seema Modi on Closing Bell and Schwab Network:

Put it in context. Zaid Admani on The Rundown noted Micron did $11.3B in the same quarter a year ago, so "in one year, the company nearly 5X their revenue." On The AI Investor Podcast, Eric Bleeker (24/7 Wall St.) cited revenue up 379% and read from the company's own statement:

"We expect memory and storage supply demand conditions to be much tighter in calendars of 2027 and 2028 than they were in 2026."

He also pulled out two contract numbers that matter more than the quarter itself. 75% of 2027 output is covered by these agreements, and 35% of revenue through 2030 is covered too. Management said this quarter is the low point for margins, and margins should rise again through fiscal 2027 as more high-bandwidth memory (HBM) ships. HBM is the stacked, very fast memory that sits next to AI chips. A consensus that assumed peak margins was not expecting that.

So why the shrug? Three explanations, from three kinds of speaker:

• The beats are getting smaller. Dan Nathan on CNBC's Fast Money (9/30) [PUNDIT]: Micron "lost money 3 years ago in 2023. They had negative gross margins," and this quarter beat consensus by only "about 5, 6%," with guidance "maybe ... up 8%." His line: "the year of the days or the quarters of having this like 50%, 100% kind of beats, they're gone."
• Not much left to surprise anyone with. Jake Silverman, semiconductor analyst at Bloomberg Intelligence (10/1) [SELL-SIDE]: "the upside is somewhat limited now because you can only have so many price increases. You can only have so much margin expansion from here."
• Investors don't believe it lasts. Admani on The Rundown: "Wall Street just doesn't believe that this memory boom and massive gross margins will last ... that's probably why Micron currently trades at less than 7 times next year's earnings," against about 18.5x for the S&P 500.

One detail is easy to miss. Kevin Green on Schwab Network said Micron's automotive business beat by about $2 billion, on roughly $6.8 billion of sales. Take that out and "it wasn't actually that much of a beat on the revenue front." Worth checking in the filing before you lean on the headline beat.

3. China is back: CXMT and YMTC are getting bigger, and better, fast [ANALYST, close to industry]

Podcast: The Circuit, "Memory Pricing Pressures, Meta's Muse AI, and the Agentic CPU Crunch" (9/28) Speakers: Jay Goldberg and Ben Bajarin, semiconductor analysts and hosts

China was missing from these podcasts for two weeks in mid-September. It is back, and with numbers. Bajarin opened the segment by calling the Chinese makers "the fourth and fifth horsemen of the memory apocalypse." Goldberg gave the detail:

"CXMT is on track to add another 100,000 wafers per month by the end of next year. So that's a 33% increase for them year on year in capacity ... it sounds like their yields are improving and they're starting to branch out in different flavors of DRAM."

(A "wafer" is the silicon disc that chips are cut from, so "wafers per month" measures factory output. "Yield" is the share of chips on each wafer that actually work.) He added that YMTC, China's NAND maker, "has pulled forward some of its capacity expansion." He was careful about the tone: "not like the sky is falling, China is coming kind of news. But just a periodic reminder that these are serious providers of memory."

The Apple angle is what should worry the incumbents. Goldberg said reports that Apple "tested CXMT memory and was pushing very hard to get a waiver ... from the US government to be able to buy it" got the other memory makers' attention, because "if Korea and Idaho can't provide the memories, China can." Bajarin's reply: "Once you're in the door with another alternate player ... Apple's not going to all of a sudden go, oh we're going to stop using them."

A second voice on China supply. On Investing Experts, "What will Micron's gross margin guide be?" (9/27) [PUNDIT], the guest said CXMT is "expected to end the year within 20,000 wafer starts range as Micron's own monthly output."

The YMTC lawsuit. Telltales, "Weekend Update - W2639" (9/27) [PUNDIT; an automated, AI-voiced show] reported that "a Munich court handed China's YMTC two injunctions over 3D NAND patents that could restrict Micron's supply into Germany." That is new, small in dollars, and worth tracking.

Why it matters: The bear case is no longer "demand will fade." It is "China builds a Samsung's worth of capacity and pushes prices down, the way it did in solar panels." A 33% capacity increase at CXMT next year, plus better yields, is the most concrete version of that risk we've heard in these podcasts.

4. Consumer memory prices may be leveling off, while data-center prices keep climbing [ANALYST]

Podcast: The Circuit (9/28), Goldberg and Bajarin

This is the most interesting new pricing detail of the week, and it is flagged as secondhand. Goldberg:

"I'm starting to hear different rumblings ... this is mostly coming from interested parties, like memory company CEOs. But it sounds to me like there is some plateauing of consumer-grade memory pricing ... I don't think memory is getting cheaper in consumer, but maybe prices stop going up again. As opposed to data center pricing. Memory is just going to keep going up."

His view is that the smart memory makers will treat these as two separate markets: price-sensitive consumers, and data centers that "will take all the memory they can get." Bajarin said two forecasts out that week cut consumer-device volumes "really across the board ... it wasn't just PCs. It was phones and everything because of the memory price bug." Chinese phone makers are holding memory specs flat or even cutting them on some models, because "you don't want to get stuck" at a higher, more expensive memory tier.

Supporting pricing data (all secondhand, and all percentages rather than dated $/GB prices):

• TrendForce says DRAM average selling prices rose 13–18% in Q3, down from roughly 50–60% in Q2 (Investing Experts, 9/27).
• Matt Bryson, Wedbush [SELL-SIDE], on Closing Bell (9/30): "All of my price checks have suggested that pricing is still moving up. It's at a little bit slower pace."
• Jake Silverman at Bloomberg Intelligence (10/1) said "Samsung mentioned $100 increase in at least some of their SKUs this morning ... we're just going to have to pay more for our phones."

Why it matters: Prices are still going up, just more slowly. For the stocks, the risk is less about prices falling than about the rate of increase slowing. A market that trades on rate of change will notice.

5. A hostile crowd at Hot Chips: does the industry even want taller HBM stacks? [ANALYST]

Podcast: The Circuit (9/28), Goldberg

This one hasn't shown up in these podcasts before. Goldberg said memory presenters at the Hot Chips engineering conference "faced a pretty hostile crowd":

"There were a lot of people pushing back saying why are you moving to hybrid bonded memory, why are you going higher. We don't need that because that is just going to be more and more wafers that are going to be needed ... let's try a different approach to memory. Doesn't have to be so high ... so many stacks of HBM."

(Hybrid bonding fuses chip layers directly together, without tiny solder bumps, so you can stack more layers of memory. "Higher" means more layers per HBM stack.)

Why it matters: It is a small but real crack in the "HBM content only goes up" assumption. More layers per stack eats more wafers, and that is part of why HBM tightens supply for all memory. If chip designers push back, it points to the other architectures coming up in item 6.

6. The chip designers say memory is now the bottleneck, and some are trying to route around HBM [OPERATOR]

Three chip-company leaders on three different podcasts said the same thing: the limit on AI is no longer raw compute. It is getting data in and out of memory fast enough, at a sane cost.

Raja Koduri (former head of graphics at AMD and Intel, now running his own chip startup), on TechSurge: Deep Tech Podcast, "Gaming Chip Pioneer Raja Koduri on China's AI Cost Advantage and Moving Beyond the GPU" (9/29):

"You can take TSMC 7 nanometer die ... and attach it to DRAM ... 3D stack, hybrid bond it. You get 10x the bandwidth than the current HBM. And 10x token generation rate. That package will beat Vera Rubin."

His point is that older chips with memory packed tightly against them can beat the newest chips, which matters for China. He said "the bottleneck has shifted quite significantly to how you deal with memory hierarchy," and that the real contest is "between China Inc. and rest of the world."

Walter Goodwin, founder and CEO of Fractile, on No Priors, "The Future of Frontier Model Architectures with Walter Goodwin, Fractile Founder and CEO" (10/2): "When you look at what it looks like to run inference at data center scale for thousands of users, the economics actually collapsed down to essentially a kind of cost per gigabyte of the memory that you're employing." Fractile is working "more closely with memory vendors" on a regular-DRAM-based platform ramping in the second half of next year. He claims "25 times more bandwidth per chip than an HBM based chip."

Sumti Jairath, Chief Architect at SambaNova, on DataFramed, "#379 The Secrets of Deploying AI in Production" (9/28): "What is the cost expensive part on this whole thing is ... the whole memory hierarchy." SambaNova combines on-chip SRAM, HBM and ordinary DDR memory ("tens of terabytes of memory per machine") to run many AI agents on fewer, lower-power racks.

Why it matters: For memory makers this cuts both ways. Bullish: everyone agrees memory is where the value sits. Mildly bearish for the HBM premium specifically: well-funded engineers are building designs that use cheaper, plain DRAM. That helps total DRAM demand, but it could cap how far HBM pricing runs over time.

7. Google says its newest AI model is saving it memory [JOURNALIST, reporting company claim]

Podcast: Closing Bell (9/30), CNBC's Mackenzie Sigalos

A small item that bears should watch. Google told CNBC it has been using its new Gemini 4 Argonne model internally "including to optimize its data centers, potentially freeing up hundreds of terabytes of memory without having to spend more CapEx dollars on additional hardware." It's a company claim and it's small against total demand, but it is the first time a hyperscaler has pitched AI-driven memory savings publicly. If that becomes a trend, it's the efficiency risk bears have been waiting for.

The debate: structural shortage, or a classic cycle with a later ending?

Both camps now agree on more than they used to. Nobody in the podcasts this week said the shortage ends in 2026. The disagreement is about 2028, and about what multiple a memory company deserves.

The structural case (steel-manned)

• Demand grows far faster than supply can. Vivek Arya, BofA Securities senior semiconductor analyst [SELL-SIDE], on Closing Bell (10/1): "the demand is growing over 100% a year. But the industry capacity is just unable to grow more than 40 or 50% a year." And: "half of the bill of materials now for these data center products is memory. Without memory, there is no AI." BofA raised its AI total addressable market forecast to $2.2 trillion by 2030, from $1.8 trillion.
• This isn't dot-com fiber. Arya again: "There is no chip that is just sitting there doing nothing, as opposed to the prior cycle where we had a lot of dark fiber. There is no dark compute right now."
• New supply arrives late and still falls short. Matt Bryson (Wedbush) [SELL-SIDE]: "Supply is always what ends these cycles. It's late '27, '28 that we get that supply going on or coming on." He added that with Broadcom "talking about demand doubling and then doubling again ... even if you get more supply coming on, it's really unclear to me that supply catches demand." Jake Silverman (Bloomberg Intelligence) went further: new fabs give "meaningful capacity" only "at the end of 2028," and "it's more like a '29 story."
• The contracts change the shape of the cycle. Ben Reitz, head of tech research at Melius [SELL-SIDE, $2,200 price target], on Squawk on the Street (10/1): "there's a view that 2028 was going to have all these capacity additions that made the story kind of roll over, and then they basically are saying to you, yes, we know about all the capacity, and we're not rolling over." On margins: "HBM margins are actually going up a lot. So margins might be able to stay in the mid to high 80s much longer than people think."
• Customers are handing chip capacity back for lack of memory. On the Futurum Equities Podcast, Ep 59 (10/2) [PUNDIT, citing unnamed insiders]: "literally there are accelerator companies giving wafers back to TSMC right now because they can't supply. They can't secure the memory." They read that as bullish for NVIDIA, which "has secured memory more effectively than any other company."

The cyclical case (steel-manned)

• About 6.5x earnings has always been the right multiple. Steve Weiss on Halftime Report, "The Setup into Q4 9/30/26" [PUNDIT; he took profits in Micron after a 300%+ gain]: "Everybody talks about how cheap it is, 6.5 times earnings. Guess what? That's where it trades. That's where it's traded forever."
• Contracts don't hold when the glut hits. Weiss again: "This is not the first time they've come out with these contracts guaranteeing price. They do it every cycle. Guess what? When things get tough, when there's a glut of memory chips, nobody cares about the contracts." He also argued that only "about 40 to 50% of their forward business is guaranteed in terms of price," and the rest is "enough to kill the earnings." His concession: "I do think the cycle will be longer."
• Shortages can turn into gluts quickly. The Investing Experts guest (9/27): with Chinese supply coming and "PCs weak, smartphones weak ... we could go very quickly from a shortage to a glut." They also noted AI labs are cutting prices hard (Anthropic's Opus 5.5 at "40% less to run," OpenAI cutting API prices "by 50%"). That squeezes the economics that pay for all this hardware.
• China is building fast. CXMT +33% capacity next year, with yields improving (The Circuit). See item 3.
• The easy money has been made. Dan Nathan (Fast Money): with smaller beats now, "investors are going to start pricing in best-case scenarios."

Where it nets out: The operator evidence (Mehrotra's contract count, the wafers being handed back to TSMC, chip designers calling memory the bottleneck) still supports the structural camp for 2027. The cyclical camp's best evidence is no longer demand. It's Chinese supply plus slowing price increases. Watch both every week.

Stocks in play

Micron (MU)

• Bull: 87% gross margin; next quarter guided to $60–63B; 26 long-term contracts out to 2031; 75% of 2027 sold; management says margins rise through FY27; "well over $100 billion in cash." On the Futurum podcast, one host called for $200 of EPS in calendar 2027 at a 10x multiple, which is $2,000 a share. Melius has a $2,200 target.
• Bear: Beats shrinking to about 5–6%; gross margin dips next quarter; a ~$25B first-half FY27 capex budget that the bears will call the start of the next glut; a possible YMTC patent injunction in Germany; the stock is still well below its June high.
• Next catalyst: December 9 buyback lockup ends. Arya said Micron can buy back "8% to 10% of their stock every year." Reitz said "might be 15% a year, easy." Also watch the full-year FY27 capex number and any HBM4 share update.

SK Hynix (000660 KS)

• Bull: It moves with Micron. Cramer on Squawk on the Street (10/1): "If Korea's up, they'll buy Micron." The Kospi was up 2% overnight after the print. Micron's 2028 comments apply across the industry.
• Bear: CXMT's capacity push hits commodity DRAM first. Hynix is the biggest name in the Roundhill DRAM ETF basket, so leveraged ETF selling hits it hard (see below).
• Next catalyst: Q3 results later in October. No SK Hynix executive appeared on any podcast this week.

Samsung Electronics (005930 KS)

• Bull: Raising prices. Bloomberg Intelligence relayed "$100 increase in at least some of their SKUs."
• Bear: Consumer-device demand is weak, and Samsung is the most exposed to phones and PCs.
• Next catalyst: Q3 preliminary results, usually in the first week or so of October. They were not discussed in any podcast in this window.

SanDisk (SNDK)

• Bull: Its earlier guidance for about 80% adjusted margins through 2030 "is what really led to a rebound in a lot of these stocks" (The AI Investor Podcast). Reitz grouped it with Micron for "a buyback of epic proportions."
• Bear: YMTC is pulling forward NAND capacity (The Circuit).
• Next catalyst: Next quarterly results; watch NAND pricing commentary.

Lam Research (LRCX)

• Bull: On The MoneyFlows Show, Ep. 44 (10/1) [PUNDIT; the hosts own it], the hosts said Lam told the Goldman Sachs Communacopia conference that "customers are asking for equipment faster than it can deliver it." Its strongest growth is in etch and deposition for "leading-edge foundry, and DRAM investments." EPS estimates are $9.47 (2027), about $12 (2028) and $14.02 (2029). P/E is down from 53x in June to 31x. The stock is 27% off its high.
• Bear: Chinese DRAM capacity growth is a double-edged sword. It's a customer today and a pricing threat to Lam's other customers tomorrow.
• Next catalyst: October quarterly results; Micron's ~$25B first-half capex is direct demand.

Advantest (6857 JP)

• Bull: The ADR jumped 8.4% to a breakout on 10/2. The company is "over a hundred billion dollar market cap" (Stock Market Today With IBD, 10/2).
• Bear: The discussion was purely about the chart. Nobody tied it to HBM test demand.
• Next catalyst: Late-October quarterly results.

NVIDIA (NVDA)

• Bull: "Getting close to 6 trillion" in market value. The Futurum hosts' insider claim is that NVIDIA secured memory better than rivals.
• Bear: Rising memory costs keep pressure on its margins (the margin squeeze covered in issue #9).
• Next catalyst: Its next results in November.

Read-throughs

Memory equipment (Advantest, BESI, Camtek, KLA, Lam, AMAT). Positive, though the evidence is mostly secondhand. Micron is spending about $25B in six months, "mostly going into construction," and that money eventually turns into tool orders. Lam says it can't ship fast enough (MoneyFlows). A Halftime Report panelist said to watch "what happens to semi-cap equipment names" after Micron, arguing the industry "stopped investing" in 2022–23 and the catch-up "is a 2027 story." BESI, Camtek, KLA and AMAT were not discussed in memory terms this week.

Packaging and substrates (CoWoS, hybrid bonding). Mixed. Koduri's pitch, a hybrid-bonded DRAM stack on an older logic chip giving "10x the bandwidth than the current HBM," is a strong long-run case for hybrid bonding, and so for BESI. But The Circuit's Hot Chips report has engineers pushing back on hybrid-bonded, taller HBM stacks. There was no CoWoS-specific capacity data this week.

GPU makers (NVIDIA, AMD). NVIDIA comes out ahead if the Futurum report is right that other accelerator companies are "giving wafers back to TSMC" because they can't get memory. The custom-chip crowd (Fractile, SambaNova, Koduri's startup) is designing around HBM supply, which signals how tight HBM is.

PC and handset makers. Clearly negative. Two forecasts cut consumer-device volumes "really across the board" (The Circuit). Samsung is raising some prices by about $100 (Bloomberg Intelligence). Chinese phone makers are freezing or cutting memory specs. The one bit of good news is Goldberg's secondhand "plateauing of consumer-grade memory pricing."

Retail leverage, a warning. On Barron's Streetwise, "Buybacks, Bond Doubts, DRAM, and T-Rex" (10/2), Jack Hough walked through the Roundhill Memory ETF (ticker DRAM). Its top holdings are Micron 26%, Samsung 25%, SK Hynix 22%, with some exposure held through swaps. Its 2x leveraged cousin (ticker RAM) is "down 45%" over three months. Hough's view: "a hot fund at the moment that's launched to capitalize on speculative fervor." After KMEM and YRAM in earlier weeks, that makes three retail memory products. This one has lost nearly half its value since launch. That shows how volatile these stocks are, separate from how the businesses are doing.

What changed vs. last week (issue #12, Sept 26)

• The big event happened. Last week was all previews. This week Micron reported, guided above consensus, and extended its outlook to 2028, where before it had talked about 2027. That was the incremental news Melius' Reitz highlighted.
• A memory-maker CEO is back on the record, with numbers. It's the first time in weeks a memory executive gave capex, margin guidance and contract counts on these podcasts. This fills the "no memory-maker exec" gap that ran through issues #11 and #12.
• Last week's checklist, scored:
  • Guidance vs. consensus: beat ($60–63B vs. $56.77B).
  • FY27 capex vs. the roughly $45B estimate: partly answered. About $25B in the first half alone, and Bloomberg Intelligence said capex rises "all the way through ... the second half of fiscal 2027." No full-year number on the podcasts.
  • HBM4 share: not answered on any podcast.
  • Contract commentary: answered and bigger. 26 agreements, out to 2031, 75% of 2027, 35% of revenue through 2030.
  • Whether other phone makers matched Samsung's $2.00/Gb price: not answered. The only new Samsung pricing point is the ~$100 SKU increase.
• China went from a footnote back to a main storyline. CXMT +100k wafers a month, YMTC pulling capacity forward, Apple's CXMT waiver push, and the YMTC injunction against Micron in Munich.
• Pricing tone softened slightly. "A little bit slower pace" (Wedbush), TrendForce Q3 +13–18%, and a possible consumer-pricing plateau. Prices are still rising, just more slowly.
• New this week: the December 9 buyback has become the main bull catalyst, replacing "the next beat."

Gaps (what we didn't hear this week)

• No HBM3E or HBM4 qualification, yield or "known good stack" data. Known good stack means a fully tested HBM stack that works before it's mounted. This is still the longest-running gap in the series. There was no HBM4 share number from Micron on any podcast, though Silverman said NVIDIA tries to be first to adopt HBM4.
• No dated contract price in $/GB. We only have percentages (TrendForce Q3 +13–18%) and "slower pace."
• No SK Hynix or Samsung executive. Samsung's Q3 preliminary results weren't discussed in the window.
• Equipment names were chart-only. Advantest came up purely on its price action. BESI, Camtek, KLA and AMAT weren't discussed in a memory context.
• No CoWoS capacity update.
• Missing shows: none of Odd Lots, SemiAnalysis, Stratechery/Sharp Tech, Asianometry, Acquired or Korean business-press podcasts had a relevant episode in the window. Coverage was heavy on CNBC, Bloomberg and investor shows.