Newsletter · · Ashutosh Agarwal

Micron's CEO Breaks Cover as Memory Stocks Roar Back - HBM & The Memory Supercycle - Week of August 22, 2026

Memory podcast roundup for the week of August 22, 2026. Micron CEO Sanjay Mehrotra went on record calling memory the strategic infrastructure for AI, SK Hynix launched a $29 billion buyback, and J.P. Morgan tripled its Micron price target as memory stocks roared back.

HBM & The Memory Supercycle

Week of August 15–22, 2026: Micron's CEO Breaks Cover as Memory Stocks Roar Back

For six weeks running, the memory story has had the same strange shape: the companies keep printing extraordinary numbers, and the stocks keep going down anyway. This week the shape finally broke. The stocks bounced hard, Micron pushed back above $1,000, and, for the first time in a while, an actual memory-maker CEO sat down and told us how he sees it. Sanjay Mehrotra of Micron went on camera in Boise. SK Hynix backed up the truck with a $29 billion buyback. And J.P. Morgan slapped a price target on Micron that would have sounded insane a year ago. The bears are still out there, and they made their case too. But the balance of the week tilted back toward the bulls.

Quick note on what a podcast can and can't give you: this is a recap of what investors and executives said out loud on shows in the last seven days, not a data feed. Where someone runs the company, I'll call them an operator. Where they're a fund manager or a pundit talking their book, I'll say so. That distinction matters more than usual this week.

TL;DR

  • A memory-maker CEO is finally on record. Micron's Sanjay Mehrotra, in a CNBC exclusive from Boise, made the whole bull thesis in one line: "Memory is no longer a component in a system. Memory is the strategic infrastructure for AI. And it's no longer a commodity." He's putting $10 billion into a new long-horizon research lab to prove it.

  • SK Hynix is showering cash on shareholders. A $29 billion buyback (40 trillion won), done by mid-November, plus a pledge to hand back more than half of free cash flow, funded, in the company's words, by "increases in the prices of memory chips."

  • The stocks rebounded and the Street chased. J.P. Morgan more than tripled its Micron target to $1,550. Memory names bounced off their lows as 13F filings showed big investors doubled down.

  • The bear didn't blink. One veteran advisor called memory "by far the riskiest part of the market right now" and told listeners to "run far, far away," arguing that everyone adding capacity at once is "a tell."

  • Still missing: no hard HBM4 yield or qualification update, no dedicated deep-dives from the semiconductor specialist shows, and no clean contract-price print, just "prices are up 10–15x" in round numbers.

What's new

1. Micron's Sanjay Mehrotra goes on the record, the strongest primary source of the week

This is the item that matters most, because it's an operator, not a pundit. Micron CEO Sanjay Mehrotra sat down with Jim Cramer in Boise for a CNBC exclusive on the network's Squawk on the Street (Aug 20). The hook was a new $10 billion "Micron Research Labs", long-horizon basic research, groundbreaking next year, with satellite campuses wherever Micron already has R&D.

But the reason to care is why he's spending it. Mehrotra's framing was blunt and repeated:

"Memory is no longer a component in a system. Memory is the strategic infrastructure for AI. And it's no longer a commodity. It is a high value... Without memory, you cannot make AI smarter. You cannot make AI faster. You cannot scale up AI."

And the reason memory demand keeps compounding, in his words: "A lot of context is getting generated. All of that context has to be stored, has to be processed through memory. So AI is driving a whole hierarchy of memory requirements, from high bandwidth memory to DRAM to SSDs."

Why it moves the thesis: for a month the entire bull case has been argued for memory makers by outside investors. Now the person who runs the biggest U.S. player is saying the same thing on camera and putting $10 billion behind the claim that memory is a durable, high-value business rather than a commodity that mean-reverts. Note the limits, though: Mehrotra gave the vision, not the granular data. No HBM4 yield figures, no customer allocation numbers, no contract-pricing direction. This was strategy, not a datasheet.

2. SK Hynix's $29 billion buyback, cash generation "phenomenal"

On Bloomberg Tech (Aug 19), executive editor Peter Elstrom broke down SK Hynix's move to win investors back after a rough slide: a $29 billion (40 trillion won) share buyback, to be completed by mid-November, plus a pledge to return more than 50% of free cash flow to shareholders going forward. Elstrom's team did the math and figures that could total "something like $170 billion over the next two years." The engine, in his description: "They're making a lot of cash because of these increases in the prices of memory chips."

There's a clever financial trick inside it, too. SK Hynix sold shares in the U.S. (its ADRs trade at a higher multiple than the Korean stock) and is buying back the cheaper Korean shares, an arbitrage regular investors can't do because of conversion caps. Shares bought back will be cancelled. SK Group Chair Chey Tae-won told Elstrom the company is expanding at home in Korea and has U.S. ambitions as well.

Why it moves numbers: a buyback of this size and speed is a confidence signal, management is effectively saying the cash is real and here to stay. It gave a tailwind to the whole group; both Micron and SanDisk got a lift from what one show called SK Hynix's "vote of confidence in their own shares."

3. J.P. Morgan triples its Micron target; the Street is chasing

On Real Vision's Macro Mondays (Aug 17), macro strategist Andreas Steno Larsen (a fund manager, so treat as informed pundit) flagged that J.P. Morgan raised its Micron price target from $500 to $1,550, more than triple. His line: "The memory trade is working wonders. Again, it's probably the best performing trade over the past week or so." He argued the assumptions baked into memory stocks are "incredibly conservative still, which is in sharp contrast to the run-up to the year 2000... maybe slightly too conservative if anything."

He also made a point worth holding onto: memory prices are "10, 15X" higher, yet they "are nowhere to be seen in the CPI index" because consumer electronics is a tiny slice of the basket. In plain terms, this is real inflation in chips that the average shopper barely feels, which is exactly why the Fed isn't forced to react to it.

4. 13F season: the smart money doubled down (and one famous investor blew up)

On Limitless (Aug 18), hosts Josh Kale and Ejaz (analysts) walked through the quarterly filings that show what big funds actually own. Two takeaways. First, the cautionary tale: Leopold Ashenbrenner, the AI investor who was "very memory-pilled", got liquidated. His book was 28.5% SanDisk and 28% Micron, over half in two memory names, and leverage, not the thesis, is what killed him. As Ejaz put it: "The thesis is coherent and it still stands... He just did it all wrong because he was using improper position sizing and, most importantly, leverage."

Second, the bullish read: other big investors bought more at the recent lows. And the punchline that keeps recurring: "All the supply that is available for memory for next year is completely sold out." The hosts flagged SanDisk's new high-bandwidth flash, a fast storage chip aimed at AI inference, as "already backlogged into the end of 2027."

5. SanDisk's investor day: 80% margins guided all the way to 2030

On The AI Investor Podcast (Aug 20), the hosts unpacked SanDisk's investor event, where management guided to 80% adjusted gross margins and 75% operating margins for 2026 through 2030, with long-term agreements now covering two-thirds of output. The bull framing here is subtle but important: everyone's been waiting for the first dip in margins as the classic signal that the memory cycle has peaked. SanDisk is arguing that margins can come off their absolute peak and still stay historically elevated for years, smoothed out by those long-term contracts rather than whipping around on the spot market.

The debate

Bull case (structural, this is a new kind of memory market): The clearest articulation came from Gil Loria on The Rundown (Aug 16). His argument, verbatim: "High bandwidth memory is something that is designed into a data center and into a server years in advance. It's not about month to month right now. It's being bought five years in advance. It does not fit any definition of a commodity." He can't reconcile the valuations: "In what world are AMD and Intel worth 35 to 60 times earnings and Micron six times earnings?" His view is that memory is "the biggest pocket of profit... in the semi pocket for the next at least two to three years," and that people pricing Micron like it's 2023 are stuck in an old mental model, "three years ago memory served a completely different role. Memory was used for storage... Now memory is used to run AI models." Mehrotra's on-record comments this week are the operator version of exactly this argument.

Bear case (it's still a cycle, and cycles always end): The sharpest bear voice was Justin Klein on InvestTalk (Aug 22). He's unmoved by the new-business-model talk: "The average retail investor... has no clue about the cyclicality of especially the memory market... all of them are trying to add capacity right now. That's a tell. That's not a plan. That's a tell that they know that this is going to be very short lived." His verdict: memory is "by far the riskiest part of the market right now, and I would run far, far away," and "these are the type of names that look the cheapest at the top and look the most expensive at the bottom." A softer version came from Ankur on Digital Disruption (Aug 17), who notes memory prices "have more than quadrupled" but warns the whole build-out is capped by physics and power, with GPU utilization he pegs "below 30%", implying the demand math may not hold once investors start asking about returns.

The synthesis: notice what both sides now concede. Neither camp thinks the physical shortage unwinds this year. The fight has narrowed to two questions: (1) how long the ~80% margins last, and (2) whether the eventual capacity wave, most likely from China, turns memory back into a commodity. The bull rebuttal to "margins can't stay at 80%" is genuinely interesting: as one AI Investor host put it, you'd "almost rather see these companies at a 70% or even a 60% gross margin because you can bring all this additional demand online", a healthier, longer cycle beats a shorter, hotter one.

Stocks in play

Micron (MU)

  • Bull: CEO now publicly framing memory as strategic AI infrastructure and investing $10B in research; J.P. Morgan target $1,550; trades ~6x earnings vs AMD/Intel at 35–60x (Gil Loria); White House pressure on Apple to avoid Chinese memory is a direct tailwind. On Market Mondays (Aug 18), the hosts hold a $1,600 year-end target and flagged the reclaim of $1,000 as the technical "all clear."
  • Bear: the InvestTalk cyclicality warning applies most directly to Micron; adding capacity into a boom is the classic top signal. Market Mondays' own aside: "Micron at 500 is not something you want to see", i.e., if it breaks, it breaks hard.
  • Next catalyst: NVIDIA earnings next week (the demand read-through the whole group keys off), plus any hard HBM4 qualification or pricing detail from Micron's own next print.

SK Hynix (000660 KS)

  • Bull: $29B buyback by mid-November, >50% FCF return, possibly ~$170B of cash returned over two years; still the HBM share leader.
  • Bear: the buyback is a shareholder-return story, not new evidence on demand; the stock had fallen ~20% into this, and the ADR-vs-Korea multiple gap shows how contested the valuation is.
  • Next catalyst: completion pace of the buyback into mid-November; any update on U.S. expansion plans that Chair Chey hinted at.

SanDisk (SNDK)

  • Bull: investor-day guide of 80% adjusted GM / 75% operating margin through 2030; two-thirds of output under long-term contracts; high-bandwidth flash backlogged into end-2027; Cramer noted on Squawk on the Street (Aug 17) it went from a "$25 billion company" to "$244 billion... in two years."
  • Bear: it "went bankrupt a number of years ago because of how bad the industry is" (InvestTalk), the poster child for NAND cyclicality; a guide is a forecast, not a result.
  • Next catalyst: whether the LTA-covered two-thirds of output actually holds pricing if spot softens.

Samsung Electronics (005930 KS)

  • Mentioned mainly as one of the two Korean memory giants and as a China-exposure risk (its Chinese fabs need U.S. license renewals, see read-throughs). No standout operator commentary this week.

Western Digital / Seagate (adjacent NAND/HDD)

  • Cramer on Squawk (Aug 17) can't square WD's move: "I don't know how Western Digital doesn't go higher... it's up 195%. And it may not be done." Filed here as sentiment, not fundamentals.

Read-throughs

Memory equipment (Advantest, BESI, Camtek, KLA, Lam, AMAT). Thin on named equipment operators this week, but two useful pundit reads:

  • Applied Materials (AMAT), on Chip Stock Investor (Aug 17), Nicholas Rossolillo walked through a record quarter: ~$9.1B revenue (+25% YoY), 50% gross margin, 30%+ operating margin, adjusted EPS +41%, guiding to over $10B next quarter (~+50% YoY at the midpoint) and a $40B+ annual run rate. Management is doubling system-manufacturing capacity by 2028 and said China is "back in growth mode" (China was ~26% of systems + services, ~$2.29B). A reverse-DCF, he notes, now prices in ~24% EPS growth a year, not cheap.
  • Lam Research (LRCX), on Full Signal (Aug 19), strategist Jacob Kuykowski pitched Lam as the highest-memory-exposure name among the big equipment makers, with a "razor-and-blades" recurring revenue stream (software subscriptions, spare parts, service). His wry twist: if China is the thing that eventually pops the memory bubble, "to pop that bubble, they have to buy so much equipment, and they're going to buy it from Lam." Heads-you-win framing.
  • Genuinely absent this week: any named commentary on Advantest, BESI, Camtek, or KLA.

Packaging / substrates (CoWoS, hybrid bonding). Nothing substantive this week, no CoWoS or hybrid-bonding technical discussion surfaced tied to memory. A gap worth noting given how central packaging has been in prior issues.

GPU makers (NVIDIA, AMD). NVIDIA reports next week and is the single biggest swing factor for the group's sentiment. Ambient color this week was cautious on the tape (per InvestTalk's weekly tally, NVIDIA fell 5.3%, Broadcom 10.5%, Intel ~14% on the week) even as memory rebounded, a reminder the memory bounce was somewhat its own story.

PC / handset / consumer OEMs. The pass-through theme is now everywhere in pundit commentary, though no OEM executive was on record. Full Signal: "Apple has had to raise prices on their computers. Nintendo had to raise prices... The consumer is going to feel it," with a prediction that fall video-game price hikes stoke "even more anti-AI" consumer sentiment. Digital Disruption echoed that Apple and others are passing memory costs into iPhones and tablets, and questioned how much more consumers will absorb.

China (CXMT / YMTC), the swing risk everyone circles back to. Still no deep-dive, but three threads:

  1. Policy tailwind for the incumbents: the White House is pressuring Apple not to buy Chinese memory, Cramer said the report that the administration "doesn't want Apple using foreign memory chips" was directly helping Micron (Squawk, Aug 17; Market Mondays, Aug 18).
  2. The long-term bear: Full Signal's Kuykowski warned China could "drastically increase DRAM and NAND production and cause prices to go down," pointing to CXMT's record-setting IPO and China's playbook in EVs, steel, and solar, "China has ruined the party for a lot of investors."
  3. A geopolitical wildcard: on Real Vision, Steno floated a theory that Trump's move to limit joint military drills with South Korea could be tied to HBM allegedly reaching China via Malaysia, and warned that Samsung's and SK Hynix's China fabs need renewed U.S. licenses, "if we see a rift developing between the U.S. and South Korea, it comes at a very, very bad time." Speculative, but a risk to keep on the radar.

What changed vs last week

A real shift. Last week (Aug 8) was "beat-and-sink", record prints, sinking stocks, and the new "memory tax" framing, with the frustrating feature that no memory-maker executive was on record (a step back from the Jensen Huang and Chey Tae-won appearances in earlier weeks).

This week reversed both:

  • The stocks bounced. Micron reclaimed $1,000 and the group rebounded off its lows; the "beat-and-sink" pattern finally broke.
  • We got a memory-maker operator back on the record, Sanjay Mehrotra of Micron, on camera, making the structural case himself. That directly fills the gap that reopened last week.
  • Concrete new capital-return news: SK Hynix's $29B buyback and >50% FCF-return pledge.
  • The Street re-rated up: J.P. Morgan's $500 → $1,550 Micron move.
  • A policy tailwind hardened: the White House pushing Apple off Chinese memory is now a recurring, market-moving talking point rather than a rumor.

The debate itself narrowed: it's no longer "is demand real" (both sides concede the shortage lasts through this year) but "how long do 80% margins last, and does China eventually commoditize this again."

Gaps this week

Being honest about what the podcasts did not deliver:

  • No HBM4 yield, qualification, known-good-stack, or layer-count update. Mehrotra named the memory hierarchy (HBM → DRAM → SSD) but gave no HBM4 qualification status. This has now been absent for several weeks running.
  • No clean contract-pricing print. Only round numbers, "10–15x" (Real Vision) or "more than quadrupled" (Digital Disruption), not a specific quarter-over-quarter DRAM or NAND contract figure.
  • No named commentary on Advantest, BESI, Camtek, or KLA; equipment read-through this week was limited to AMAT (record quarter) and a Lam thesis, both from pundits, not equipment executives.
  • No CoWoS or hybrid-bonding discussion tied to memory packaging.
  • No China memory-capacity deep dive, CXMT and YMTC surfaced only in passing (IPO scale, Apple-sourcing politics), not with capacity or share numbers.
  • No specialist deep-dive shows in the window, nothing from SemiAnalysis, Odd Lots, Stratechery/Sharp Tech, Asianometry, Six Five, or Tech Disruptors on memory this week. The coverage this week leaned CNBC/Bloomberg and generalist-investor podcasts.