Newsletter · · Ashutosh Agarwal

Micron Reports 84.9 Percent Margins and a 100 Billion Contract Book on the Memory Supercycle - HBM & The Memory Supercycle - Week of June 27, 2026

HBM and memory supercycle newsletter for the week ending June 27, 2026. Micron's FY-Q3 print landed at $41.5B revenue and 84.9% gross margin with $100B in contracted revenue, CEO Sanjay Mehrotra said he has no line of sight to supply catching demand before 2028, Apple raised hardware prices, and SK Hynix set a July 10 $29B US ADR listing.

HBM & The Memory Supercycle

Week of June 27, 2026: Micron Reports 84.9 Percent Margins and a 100 Billion Contract Book on the Memory Supercycle

HBM and The Memory Supercycle, week ending June 27, 2026

TL;DR

  • Micron's FY-Q3 print broke the model. Revenue ~$41.5B vs. ~$35B expected, gross margin 84.9% (above NVIDIA's), guidance to ~$50B next quarter at ~86% margins, and a stack of long-term, floor-priced customer contracts that the bulls say re-rates memory out of "commodity."
  • The CEO removed the timer. Sanjay Mehrotra told the Street he has "no line of sight" to when supply catches demand, and doesn't see industry supply improving even gradually until 2028. Apple capitulated the same week and raised hardware prices.
  • The debate is no longer "shortage or not." It's duration. Even committed bears (Jim Chanos) won't short DRAM here; the cleanest bear signal, SK Hynix nudging capacity toward commodity DRAM, got read two opposite ways.

What's new

1. Micron's quarter was the whole week (operator data). On CNBC's Squawk on the Street, David Faber walked through a print where "net income doubles quarter on quarter, gross margin guidance is better than even NVIDIA's at their top… operating cash flow, $25 billion for the quarter, free cash flow, $18 billion a record" at "84.9 percent gross margins." The thesis-mover is structural, not the beat: CEO Sanjay Mehrotra's line that "we currently do not have line of sight as to when memory supply will be able to catch up with increasing demand," with no gradual relief until 2028 (The Rundown).

2. The order book is the re-rating (operator data). Micron locked in "14 strategic customer agreements, securing $100 billion in contracted revenue, $22 billion in cash just to secure capacity… sold out until at least 2027," per Stephanie Link on CNBC's Halftime Report. She pegged ASP gains at "DRAM up 60% and NAND up 80%." On The AI Investor Podcast, the read was that these deals run "through 2030" and could push ">50% of revenue" under contract, and, tellingly, that the only line Micron missed was capex: "they're not building capacity as fast as they want." A missed capex number as a bull signal, that's the whole supercycle in one data point.

3. The pricing print, from outside the hype (industry expert). The sharpest supply-side framing came from an auto show. Kearney partner Kushal Fernandes, on Automotive News Daily Drive, put DRAM "up roughly 450% in just four months" and called it "a structural reallocation of supply driven by AI companies in an arms race," not a normal cycle. His killer technical point: "for the same one gigabyte of HBM, you need three to four times the wafers as you would for one gigabyte of conventional DRAM," demand is up and each unit eats more supply. He also flagged the discipline: the three suppliers "have been remarkably disciplined… that has not been to the same extent as we did during the COVID shortage," with new capacity "two to three years" out.

4. Apple blinked (read-through). Memory is now an inflation vector. On Squawk on the Street, CNBC noted "memory prices have quadrupled in the past three quarters" and that Apple is modeling the "iPhone 18 Pro… to cost $280 more at that Pro level." Apple's own words: the industry faces "an unprecedented challenge with AI data centers driving an extraordinary surge in demand for memory and storage." When the most powerful buyer in consumer electronics eats the price, that's the cycle telling you who has the whip.

5. SK Hynix is about to tap US markets (operator/structural). Per Bloomberg Tech, Hynix, having "jumped ahead of Samsung" in HBM, plans to "start trading these ADRs on July 10th," raising ~$29B to fund fab expansion. Potentially a top-five offering ever, and a fresh way to play the trade.

The debate

Steel-man the bulls (structural, multi-year). The supply side is physics, not sentiment. Long/short investor Val Zlatev at the MacroMinds Symposium (Monetary Matters) argued the equipment makers "cannot really grow their revenues or their shipments by much more than 30%… 30% to 35% a year. That's kind of a max," so bit supply is capped regardless of demand. Memory is "up four to five X," bill-of-materials for a PC/phone has gone "20%… now it's like 50%," and units are "down mid teens." His punchline: the market is pricing "a big sharp price decline… six to nine months out," and "that is very unlikely to happen," the peak is "shallow… two, three years or whatever it is, four years" before any rollover.

Steel-man the bears (cycle, about to roll). It's still DRAM. Will Kerwin on The Morning Filter called memory names supply-constrained commodity chips, with "six primary memory makers… building substantial new supply in late 2027–2028" and Micron trading "more than twice" Morningstar's $455 fair value. The cleanest tell: a local report that SK Hynix is shifting from higher-margin HBM toward cheaper commodity DRAM, which one Bloomberg guest read as "an early warning sign that the current cycle of elevated memory pricing will eventually reverse" (Bloomberg Businessweek).

The tell that this is a duration argument, not a direction one: even Jim Chanos won't take the short. "In my 40 years, I don't think I've ever made a single dollar being short the DRAM companies… we've never been able to time correctly" (Monetary Matters). Or, as Adam Parker put it on CNBC: "long for three months, short three years. And there's just a big truck you can drive in between."

Stocks in play

  • Micron (MU). Bull: floor-priced contracts through 2029–2030, 84.9% margins, Anthropic co-design tie-up (Buy Hold Rant). Bear: trough-cycle multiple optics, ~2x Morningstar FV, capacity glut risk in 2027–28. Watch: Q4 print (guided ~$50B / ~86% GM) and whether 2027 gets called sold out, not just 2026.
  • SK Hynix (000660 KS). Bull: HBM share leader, ahead of Samsung. Bear: the HBM-to-DRAM capacity nudge is the bears' favorite chart. Watch: the July 10 US ADR listing (~$29B) and any color on HBM4 allocation.
  • Samsung (005930 KS). Bull: sector capex backstop and HBM catch-up optionality. Bear: still the HBM laggard. Watch: HBM qualification at NVIDIA.
  • SanDisk (SNDK). Bull: NAND ASPs "up 80%." Bear: pure-play cyclicality, Micron's NAND alone (~$9.9B/qtr) is now "bigger than the entire company of SanDisk" (Buy Hold Rant). Watch: NAND contract direction.
  • NVIDIA (NVDA). Coatue's Philippe Laffont called it "a very cheap stock… 13 or 14 [times] 2027" on Squawk Pod. Watch: HBM cost as a gross-margin input.

Read-throughs

  • Memory equipment (Advantest, BESI, Camtek, KLA, Lam, AMAT): Quiet on names this week. The relevant signal is Zlatev's "30–35% a year" shipment ceiling, bullish for backlog visibility, but nobody named orders. Laffont's framing is the cleanest equipment thesis on the tape: own "a supplier to the fabs" so "I don't need to make an exact bet on which of the chips is going to win."
  • Packaging / substrates (CoWoS, hybrid bonding): Thin. AT&S's CFO discussed substrates and namechecked CoWoS on In the Know but gave no HBM-specific capacity or yield. Hybrid bonding surfaced only as a China workaround for EUV-less memory.
  • GPU makers (NVIDIA, AMD): HBM is now a visible cost line; memory is reportedly ~48% of next year's hyperscaler spend (The AI Investor Podcast). A "tax" on the whole AI trade.
  • PC / handset OEMs: This is the pain trade. Apple raising prices; Zlatev sees PC/phone units down mid-teens as memory eats half the BOM. Watch OEMs without Apple's pricing power.

What changed vs. last week

A lot. Last week was Micron preview and a Kospi-led chip sell-off (SK Hynix and Samsung each -12% mid-week). This week the actual print landed, and it was a blowout that flipped the tape from "is the cycle rolling?" to "how long is the plateau?" New this week: the $100B contract book, the explicit "no line of sight" CEO language, Apple's price hike, and a firm July 10 date on the SK Hynix ADR.

Gaps

  • No hard HBM3E/HBM4 qual or yield (KGS) update. HBM3E was called "digital gold… the backbone for NVIDIA's Blackwell" (Motley Fool), but no stack counts, bin splits, or yield figures surfaced.
  • No clean contract-pricing level. Best directional reads: DRAM ASP +60% / NAND +80% (Halftime), the 450%/4-month DRAM figure (Kearney), and a "DRAM contract pricing… set to jump 50–55% this quarter" claim from Telltales, note that show is AI-voice-generated and aggregates, so treat as secondary.
  • China (CXMT/YMTC) stayed qualitative. CXMT is reportedly "adding three times as much capacity as Micron this year" (The AI Investor Podcast) and building a facility to "flood the global market with cheap legacy DRAM" (Motley Fool), no node, yield, or bit figures. One analyst (Mehdi Hosseini, Closing Bell) flagged a US export-control uplift on YMTC/CXMT as a risk to watch.
  • No SK Hynix / Samsung earnings detail. All commentary was stock-price or ADR-listing related, not segment financials.