Newsletter · · Ashutosh Agarwal

HBM & The Memory Supercycle - Week of June 20, 2026: Micron's Blockbuster Looms as Memory's Hottest Cycle Ever

HBM and memory newsletter for the week of June 20, 2026. Micron heads into its June 24-25 print on the loudest setup of the cycle, with the tape near-unanimous that this is the strongest memory cycle in history and the only real debate being whether the pricing power is cyclical or secular.

HBM & The Memory Supercycle

Week of June 20, 2026: Micron's Blockbuster Looms as Memory's Hottest Cycle Ever

HBM & The Memory Supercycle, week ending June 20, 2026

TL;DR

  • Micron reports June 24-25 into the loudest setup of the cycle: TD Cowen just more than doubled its target to $1,500, the CEO has hinted at a quarter that could throw off over $27B in gross profit, and DRAM contract is up ~15% month-to-date. The number is a foregone conclusion; the reaction is the trade.
  • The tape is unanimous that this is the strongest memory cycle in history, DRAM/NAND prices up 50-100%, HBM "sold out" possibly into end-2027, and a three-supplier oligopoly running mid-to-high-single-digit capacity adds against 2-3x demand. Operators (HPE's CEO) and analysts agree supply doesn't normalize until 2028 at the earliest.
  • The one debate that matters: cyclical or secular? Bulls say long-term agreements with floor margins have permanently re-rated the group. Bears remember that memory "always destroys numbers when the stock peaks." Nobody on this week's tape claims to know for sure, and that honesty is the tell.

What's New

1. The $1,500 Micron call, and why it's not just a number. On CNBC's Power Lunch (June 15), TD Cowen senior analyst Krish Sankar (sell-side) walked through more than doubling his target to $1,500 from $660. The mechanism isn't just price, it's a re-rating: "into the print last time, Micron was trading at almost four times earnings. Now it's probably trading at eight times our numbers, probably 10 times consensus." His durability thesis: "You never see stability in pricing leading to durability of earnings. And I think that's something you might start seeing today." Note the demand mix shift, "first you saw the GPUs driving memory. Now you're seeing the CPUs driving memory" (agentic AI = more DRAM per CPU). Moves numbers because it reframes memory from a price-to-book cyclical into a PE compounder.

2. A buyer says the quiet part out loud. On Squawk on the Street (June 17), HPE CEO Antonio Neri, an actual operator and memory buyer, was asked whether the price surge is denting demand: "Not at all... it's actually an acceleration of demand." On timing: "we're going to see slight improvement in '27, but that's going to be through yields improvements, packaging strategies... the true wafer capacity is not coming online until sometime '28... So this is going to be here with us for a long period of time." When the customer tells you he can't get enough at any price, believe him over the strategists.

3. The supercycle's physics, explained. TechSurge: Deep Tech Podcast (June 16) ran the best structural deep-dive of the week (analyst/pundit). The framing: "this is probably the strongest memory cycle in history right now... don't forget 18 months ago, we were in the midst of the worst memory cycle, probably since the tech bubble." The HBM kicker is the trade ratio: "you need something like four times as much silicon area to make a gigabyte of high bandwidth memory" versus standard DRAM, so adding wafer capacity doesn't add many bits. February industry revenue was "up something like 80% year over year... almost all memory pricing." This is the bit-supply argument the whole bull case rests on.

4. Apple offers to help foot the bill. On Hedgeye's "Protect the Pile" (June 19, buy-side research), the actionable nugget: "DRAM pricing is up like 15% month to date." Plus a fresh wrinkle, Apple signaling it will "use their balance sheet to help partially fund some of the supply challenges... particularly DRAM and NAND." Translation: "the capex burden... at the margin may not be as steep this cycle" if marquee customers co-fund supply. Better FCF math for suppliers, but Hedgeye's warning lands harder: memory names "always [destroy] numbers when the stock peaks."

5. The oligopoly, quantified. On Bloomberg Daybreak (June 19), Bloomberg Intelligence's Mandeep Singh sized it cleanly: memory is "an oligopoly with Micron, SK Hynix and Samsung... mid to high single digit type of capacity expansion" against demand growing "2x, 3x," with Micron set to "post over 200% growth... in 2026." His release valve for the mismatch: bigger capex, "or... a new memory player out of China."

The Debate

Steel-man the bull (structural shortage): Three suppliers, capex discipline (mid-single-digit adds), and an HBM trade ratio that eats four wafers' worth of area per usable bit. Demand is AI bit-demand, not consumer, and it's signed under long-term agreements with floor gross margins, the SanDisk LTA structure Sankar credits for the re-rating. HPE's CEO says the market has "structurally changed... In the past, it was driven by consumer demand. Now it's driven by [AI] demand." On Futurum Equities (June 18), the bolder version: if pricing power runs three years, "this might be a $2,000 stock," with consensus' 2027 peak-earnings call "offsides."

Steel-man the bear (a cycle dressed up as a religion): The single sharpest piece of thinking this week came from Imran Khan (ex-sell-side, now investor) on RiskReversal Pod (June 19): "this is not a cyclical, this is really a secular growth... but I also think we don't have a clear answer, whether it's a cyclical or secular." Memory has always been a boom-and-bust valued on price-to-book (1-3x). If 2029-30 growth is positive double-digits, "the stock is super cheap." If earnings go negative as they always eventually have, "they're very expensive." His scar tissue: in 2020 "Zoom was 20% of Google's market cap. Today it's less than 1%."

The honest read: the bull and bear are arguing about the same fact, durability of pricing, and neither can prove it yet. June 24 won't settle it. The 2027 capacity ramp will.

Stocks in Play

Micron (MU) : Bull: "81% gross margins... record free-cash-flow machine... only US-based advanced memory maker," still ~8x earnings and a lower PEG than Nvidia per Futurum. Bear: per Daily Stock Picks (June 15), "you're trying to buy the best company... at the most euphoric time. With the CEO selling shares... and a 157% gap between the stock price and its 200-day average. History is clear that even when Micron delivers monster beats, the stock often sells off." Next catalyst: earnings June 24-25, watch the capex guide (Hedgeye: "that's like the biggest number") more than the EPS beat. RiskReversal's host cited ~$88 EPS this year vs. a $4.50 loss in 2023.

NVIDIA (NVDA) : Bull: memory cost pass-through isn't hurting it, Khan notes "everybody was concerned that their margin is going to get hit hard because of memory prices. We haven't really seen it," thanks to long-term supply deals. Bear: it raised $25B in debt this week (first IG issuance since 2021) into a frothy tape. Watch: whether HBM allocation, not GPUs, becomes the gating item.

AMD : Bull: HPE will support the AMD Helios rack stack (with Juniper QFX fabric), a real enterprise design win. Bear: Futurum flags "the market share that AMD is going to lose due to Nvidia is not being priced in" on the CPU/PC side. Watch: Helios deployment traction.

Microsoft (MSFT) : Called out repeatedly as "the dog of the Mag 7," down ~15% YTD, trading under 20x forward. Bull: Azure + Copilot re-rating optionality. Bear: it's a payer of rising memory costs and its custom silicon is "years behind." Watch: Copilot revenue inflection.

SK Hynix (000660 KS) / Samsung (005930 KS) : No insider commentary on the tape this week; both appear only as the other two legs of the HBM oligopoly (Samsung + Hynix named as the scarce HBM vendors by Lambda's founder). Watch: any HBM3E/HBM4 qual or KGS yield update, conspicuously absent this week.

SanDisk (SNDK) : Not discussed by name this week, but its LTA-with-floor-margin structure is the template analysts now credit for memory's durability re-rating.

Read-Throughs

  • Memory equipment (Advantest, BESI, Camtek, KLA, Lam, AMAT): Thin week, no episode dug into test/inspection/bonding by name. TechSurge's only color: the WFE names "haven't gone up 10x, like the memories" (memory upside is price-driven, not unit-driven), and 2026 is "a constrained year" for WFE because fabs/clean rooms must be built before tools ship, the equipment leg lags into 2027.
  • Packaging / CoWoS / hybrid bonding (BESI, TSMC): Essentially absent from this week's tape. Noted as a gap, not a non-event.
  • GPU makers (NVDA, AMD): Memory is "a big part" of NVIDIA's bill of materials (Lambda's Stephen Balaban, The MAD Podcast, June 18), rising HBM cost is a GPU gross-margin watch item, so far absorbed via LTAs.
  • PC / handset OEMs: Squarely in the crosshairs, Apple is raising device prices and co-funding component supply. Rising DRAM/NAND is now a consumer-hardware margin and pricing story.

What Changed vs. Last Week

This is the inaugural issue, so there's no prior baseline to diff against. Setting the marker for next week: the open questions are (1) Micron's June 24-25 capex guide and the post-print reaction into quarter-end flows, (2) any first hard HBM4/HBM3E qualification or yield data point, and (3) whether equipment names start to catch the memory bid.

Gaps (Honest Absences This Week)

  • No HBM3E/HBM4 qualification, KGS, or yield specifics : the bull case's core technical milestones went unmentioned.
  • No equipment or packaging deep-dive : Advantest/BESI/Camtek/KLA/CoWoS/hybrid bonding barely surfaced.
  • No SK Hynix or Samsung insider commentary : Micron is the sole memory proxy on the tape.
  • No hard NAND spot-pricing print : only directional "DRAM/NAND up 50-100%."
  • China (CXMT/YMTC) appeared only as a qualitative supply-threat aside, not a fresh data point.