The "NAND leads DRAM by ~3 quarters" ordering fails the n=3 test: observed leads are 0, ~1 and ~2-3 quarters, and the 2022 cycle inverts it
The question
Verbatim: "Does the 'NAND rolls over ~3 quarters before DRAM' ordering hold in the 2008 and 2022 memory cycles? At n=3 it becomes a usable tripwire rule; if it fails, the 2018 observation is coincidence and should be demoted."
Context: this is a falsification test of one tripwire in memory-cycle v1.1 (Phase 2), not a bull or bear case. The claim under test came out of [[2026-09-09-2018-capacity-announce-asp-rollover-lag]], which said the NAND-leads-DRAM sequence was "the one piece of 2018 structure that repeats cleanly."
What we already know (from the vault)
- The claim as written. The parent brief's conclusion: "in 2018, NAND rolled over roughly three quarters before DRAM... Watch the NAND-leads-DRAM spread, not the calendar" ([[2026-09-09-2018-capacity-announce-asp-rollover-lag]]). Its 2018 evidence: NAND contract prices dropping in 1Q18 (DRAMeXchange, 2017-12-11), DRAM growth stopping after 2Q18, plateau 3Q18, first QoQ decline 4Q18 (TrendForce 2018-08-16 and 2018-09-26). On its own numbers that is a 2-3 quarter lead, not a clean 3.
- The tripwire it feeds. Tripwire #4 in the Phase 2 to Phase 3 set is "NAND contract prices going negative QoQ while DRAM stays positive," named the highest-value indicator ([[2026-07-07-dram-hbm-phase2-phase3-early-signals]]).
- Verified data limitation, now confirmed directly. The anchor CSV
trendforce-historical-2017-2026.csvholds 252 rows, true range 2021-02-25 to 2026-04-24. Zero pre-2021 rows, exactly as the parent brief warned. Coverage is also uneven: 24 rows in 2021 and only 8 rows in all of 2022. It is a press-release scrape with adirection_numericflag, not a price series. - The 2022 cycle is nevertheless readable from it, because the 2021 rows carry TrendForce's own forward QoQ contract-price guidance for both DRAM and NAND quarter by quarter.
- The vault's own phase labels make "the 2008 cycle" ambiguous.
phase-history.csvlabels the 2008 cyclecapacity-announcement 2006-12,capacity-online 2008-07,down-cycle 2008-09(GFC) ([[phase-history-notes]]). The ASP peak for that era was 2006, so the 2008-09 label is a second-leg demand collapse, not a rollover from a positive QoQ regime.
What the web says
- 2006-07 rollover, DRAM side, dated to the week. DigiTimes, 2007-01-09, reporting DRAMeXchange's first contract update of 2007: DDR contract prices rebounded but DDR2 "saw a reduction for the first time since July 2006." So DRAM contract prices were rising July-December 2006 and first turned negative in January 2007 (1Q07). DRAMeXchange put the 1Q07 DDR2 average at -20% QoQ, and by early March was reporting a further ~10% contract drop in the first half of the month.
- 2006-07 rollover, NAND side, dated to the quarter at best. InformationWeek, 2007-02-02: "Starting late last year, poor seasonal demand prompted Samsung, Hynix and others to dump 8-Gbit NAND" onto the spot market; NAND ASPs "fell sharply" in January 2007, and 2006 full-year ASPs were down about 60%. Gartner then forecast -25-30% for 1Q07 and Hynix -30-35%. That puts the NAND turn in 4Q06, roughly one quarter ahead of DRAM.
- The same article supplies the structural objection. Gartner: NAND prices have historically dropped about 40% per year as a baseline. NAND's per-bit series carries a large negative secular drift that DRAM's does not.
- Within the 2007 downturn the sign relationship inverted. DRAMeXchange reported fixed-density NAND MLC contract prices (4/8/16/32Gb) up 4-7% in late March 2007, up more than 10% in 2Q07 on easing oversupply, and up 15-25% in the first half of August 2007 after a Samsung fab power disruption. Over the same stretch iSuppli had DRAM per-megabit ASP -39% QoQ in 2Q07 (ElectronicDesign, 2007-07-30). For two quarters of 2007 the condition was DRAM negative while NAND positive, the exact opposite of the tripwire.
- 2022 cycle, both legs from TrendForce's own contract guidance (via the vault CSV's sourced rows). 3Q21: DRAM +3-8%, NAND +5-10% - both positive. 4Q21: DRAM -3 to -8%, NAND -0 to -5% - both negative, same quarter. TrendForce's first published call that DRAM would decline came 2021-08-10 (PC DRAM -0-5% in 4Q21); its first call that NAND would decline came 2021-09-23, six weeks later.
- 2022's inversion is on the record too. TrendForce, 2022-03-28: overall DRAM to drop 0-5% in 2Q22. TrendForce, 2022-03-30: NAND to gain 5-10% in 2Q22 following raw-material contamination at Kioxia and Western Digital. Again DRAM negative, NAND positive.
- Current cycle reads inverted as well. TrendForce, 2026-09-30: 4Q26 conventional DRAM +10-15% QoQ, NAND +15-20% - NAND is the stronger of the two right now ([[2026-10-03-dram-asp-series-basket-consistency-tripwire-1]]).
Convergences and contradictions
- Convergence on direction, divergence on magnitude. Across the three testable rollovers NAND turned no later than DRAM in all three, but the lead is 0 quarters (2021-22), ~1 quarter (2006-07), ~2-3 quarters (2018). The spread covers the entire decision-relevant window. The vault's "~3 quarters" is matched by none of the three observations, including the one it was derived from.
- Direct contradiction of the parent brief's generalization. [[2026-09-09-2018-capacity-announce-asp-rollover-lag]] called the ordering the piece of 2018 structure that "repeats cleanly." On the 2021-22 cycle it does not repeat at all: the two rolled in the same quarter, DRAM's decline was the deeper one (-3-8% vs -0-5%), and DRAM was flagged first.
- The sign condition is the weakest part, not the strongest. In two of three cycles (2007 and 2022) the tripwire's literal state - NAND negative while DRAM positive - occurred with the signs reversed during the downturn, driven by NAND-specific supply accidents (Samsung Hwaseong power loss Aug 2007; Kioxia/WDC contamination Feb 2022). A rule keyed on the sign pair would have produced a false all-clear in both.
- Measurement bases do not match across cycles, so the three numbers are not the same measurement. 2006-07 mixes blended per-megabyte ASP (iSuppli/Gartner) with fixed-density contract prices (DRAMeXchange); 2018 rests on directional press-release language; 2021-22 rests on TrendForce forward contract guidance. No cycle uses a common $/Gb series, because the vault has none before 2021 and the public ones are gated.
Synthesis for RDCO
Verdict: the ordering FAILS as a calibrated lead-time rule. Demote the 2018 observation. The concrete change to memory-cycle v1.1: tripwire #4, "NAND contract prices negative QoQ while DRAM stays positive," should be reclassified from leading indicator to corroborating co-indicator, and stripped of any implied lead time. It stays on the scoreboard, because NAND turning first is still weak evidence of a cycle turn, but it must no longer be described as the highest-value indicator, must no longer carry "~3 quarters," and must no longer be allowed to move the Phase-3 prior on its own. The parent brief's instruction to "watch the NAND-leads-DRAM spread, not the calendar" does not survive this test; it was an n=1 extrapolation that the 2021-22 cycle contradicts outright.
Why it fails is more useful than that it fails. Two independent problems. First, a drift problem: NAND per-bit prices fall roughly 40% a year as a structural baseline (Gartner), while DRAM's secular drift is far shallower. Comparing the sign of two series with different drifts will tend to show NAND going negative first regardless of cycle phase, which means the 2018 "hit" carries much less information than it looks like it does. Second, a shock problem: NAND is a six-supplier market with a documented history of single-fab accidents that reprice the whole market within weeks. Aug 2007 Samsung power loss, Dec 2021 Xi'an lockdown, Feb 2022 Kioxia/WDC contamination. In 2007 and 2022 those shocks put NAND up while DRAM was down, producing exactly the false all-clear a sign-pair rule is most vulnerable to.
On the sample size, plainly. This is n=3 at best, and the three readings are not the same measurement. One cycle (2018) was read off directional press-release language; one (2021-22) off forward guidance ranges with only 8 vault rows in the whole year; one (2006-07) off two different ASP conventions that disagree with each other. A fourth candidate, the vault's labelled down-cycle 2008-09, is untestable for this question: by September 2008 both NAND and DRAM had been falling for roughly seven quarters, so there was no positive QoQ regime to roll over from. The honest statement is that we have three heterogeneous observations of a quantity measured in quarters, with values 0, ~1 and ~2-3. No average should be computed from that, to any number of decimal places, and the range is wide enough that it cannot time anything. Even if all three had agreed, three same-direction observations of an industry-wide cycle are not three independent trials - the same three suppliers, the same demand base, overlapping decision-makers.
What survives, narrowly. One claim does hold 3-for-3: NAND has never rolled over later than DRAM in the cycles we can read. That is a real asymmetry and it is consistent with NAND's heavier consumer and retail exposure, which turns earlier than server and PC OEM contract demand. But "no later than" with a lead that is sometimes zero is a sanity check, not a tripwire. It is worth keeping as a falsifier in the other direction: if DRAM ever goes negative QoQ while NAND is still clearly positive on a demand-driven basis - no NAND supply accident in play - that would be a genuine surprise worth investigating, because it has not happened in three cycles outside of shock conditions. Note that the current print is the benign version of exactly that shape: 4Q26 NAND +15-20% against conventional DRAM +10-15%, both strongly positive, so nothing is firing and the ordering premise is currently inverted.
Why this is in the vault
This closes the last open follow-up carried in [[2026-09-09-2018-capacity-announce-asp-rollover-lag]] and changes a specific row in the Phase 2 to Phase 3 tripwire table in [[2026-07-07-dram-hbm-phase2-phase3-early-signals]]: tripwire #4 drops from "highest-value indicator" to corroborating co-indicator with no lead time attached, and the memory-cycle v1.1 scoreboard should no longer let it contribute to a Phase-3 call on its own. It also adds dated 2006-07 and 2021-22 NAND-vs-DRAM rollover rows that /investing:label-historical-phases can use, and it records that the vault's down-cycle 2008-09 label cannot serve as a rollover-ordering test case.
Open follow-ups
- Does the ordering hold on a drift-adjusted basis - NAND QoQ relative to its own trailing 8-quarter average decline rate, rather than against zero? That removes the 40%/yr structural drift that may be manufacturing the 2018 hit, and is the only version of this tripwire that could still be informative.
- How often do NAND-specific supply shocks (fab accident, contamination, power loss, regional lockdown) reprice NAND contract prices by more than 5% QoQ against the prevailing cycle direction? If the base rate is high, any NAND-keyed tripwire needs a shock exclusion and the question is what observable triggers it.
- Which demand segment actually leads: is the 2018 NAND lead better explained as a smartphone and retail-storage air-pocket arriving before server and PC OEM contract demand? If so, the leading indicator is segment mix (consumer-led vs datacenter-led turns), not the NAND/DRAM product split, and the 2021-22 tie is explained by its being a PC-inventory-led turn.
- Did enterprise SSD contract prices lead commodity NAND and DRAM in any of these cycles? The 2021 vault rows track enterprise SSD separately (+15% QoQ in 3Q21), and it sits closer to the datacenter demand that drives the current cycle than consumer NAND does.
- Can a consistent $/Gb contract series for 2006-2008 be built from Micron, Hynix and SanDisk filings and earnings decks, so the 2006-07 rollover stops resting on two incompatible ASP conventions? This is the same recovery problem the parent brief flagged, re-aimed one cycle earlier.
Related
- [[2026-09-09-2018-capacity-announce-asp-rollover-lag]]
- [[2026-07-07-dram-hbm-phase2-phase3-early-signals]]
- [[2026-10-03-dram-asp-series-basket-consistency-tripwire-1]]
- [[2026-06-28-chip-memory-cycle-phase2-phase3-indicators]]
- [[2026-09-25-hbm-dram-phase-2-3-transition-q3-2026]]
- [[2026-09-29-phase3-memory-cycle-rotation-basket]]
- [[2026-05-18-memory-cycle-v1.1]]
- [[phase-history-notes]]
Sources
Vault
~/rdco-vault/06-reference/research/2026-09-09-2018-capacity-announce-asp-rollover-lag.md- the claim under test and its 2018 evidence~/rdco-vault/06-reference/research/2026-07-07-dram-hbm-phase2-phase3-early-signals.md- tripwire #4 as currently written~/rdco-vault/06-reference/research/2026-10-03-dram-asp-series-basket-consistency-tripwire-1.md- 4Q26 conventional DRAM +10-15% vs NAND +15-20%, basket-label discipline~/rdco-vault/01-projects/investing/anchors/dram-spot/trendforce-historical-2017-2026.csv- verified 252 rows, 2021-02-25 to 2026-04-24, 8 rows in 2022; source of the 2021-22 contract-guidance readings~/rdco-vault/01-projects/investing/anchors/memory-cycle-v1.1/phase-history.csv- 2008-cycle labels (2006-12, 2008-07, 2008-09) and 2022-cycle labels (2021-12, 2022-03)~/rdco-vault/01-projects/investing/anchors/memory-cycle-v1.1/phase-history-notes.md- labelling methodology and confidence tiers~/rdco-vault/01-projects/investing/theses/2026-05-18-memory-cycle-v1.1.md- thesis the tripwire belongs to
Web (fetched and quoted)
- InformationWeek, 2007-02-02 - "Flash Memory Prices In Precipitous Plummet": https://www.informationweek.com/it-leadership/flash-memory-prices-in-precipitous-plummet - NAND dumping "starting late last year," January 2007 ASP collapse, 2006 ASP -60%, Gartner 1Q07 -25-30%, 40%/yr secular decline
- ElectronicDesign, 2007-07-30 - "DRAM Market Hits Bottom": https://www.electronicdesign.com/news/dram-market-hits-bottom - iSuppli DRAM per-megabit ASP -39% QoQ in 2Q07
Web (cited from search results, body not fetched)
- DigiTimes, 2007-01-09 - DDR2 contract price falls "for the first time since July 2006" per DRAMeXchange: https://www.digitimes.com/news/a20070109VL204.html
- DRAMeXchange weekly research archive - 1Q07 DDR2 average -20% QoQ; ~10% contract drop in 1H March 2007; NAND MLC contract +4-7% late March 2007; NAND contract +15-25% in 1H August 2007 after Samsung fab power disruption: https://dramexchange.com/WeeklyResearch/Post/2/567.html and https://www.dramexchange.com/WeeklyResearch/MarketView/131
- siliconsemiconductor.net / iSuppli - NAND per-megabyte ASP -40% in 1Q07: https://siliconsemiconductor.net/article/69994/iSuppli_sees_reasons_for_optimism_in_2007
Web (TrendForce press releases, URLs carried in the vault anchor CSV with title and summary quote)
- 2021-07-13 - DRAM 3Q21 +3-8% QoQ: https://www.trendforce.com/news/2021/07/13/dram-prices-for-3q21-projected-to-undergo-minor-qoq-increase-of-3-8-owing-to-rising-demand-from-peak-season-says-trendforce/
- 2021-07-14 - NAND 3Q21 +5-10% QoQ: https://www.trendforce.com/news/2021/07/14/nand-flash-contract-prices-likely-to-increase-by-5-10-qoq-in-3q21-as-quotes-continue-to-rise-says-trendforce/
- 2021-08-10 - first DRAM decline call: PC DRAM -0-5% in 4Q21: https://www.trendforce.com/news/2021/08/10/contract-prices-of-pc-dram-expected-to-decline-by-0-5-in-4q21-as-spot-prices-of-dram-modules-continue-to-weaken-says-trendforce/
- 2021-09-22 - DRAM -3-8% QoQ in 4Q21: https://www.trendforce.com/news/2021/09/22/dram-prices-projected-to-decline-by-3-8-qoq-in-4q21-due-to-rising-level-of-client-inventory-says-trendforce/
- 2021-09-23 - first NAND decline call: NAND -0-5% in 4Q21: https://www.trendforce.com/news/2021/09/23/nand-flash-market-will-see-falling-quotes-and-0-5-qoq-declines-in-contract-prices-for-4q21-says-trendforce/
- 2021-10-12 - DRAM exits a three-quarter bullish run, 4Q21 -3-8%: https://www.trendforce.com/news/2021/10/12/dram-prices-projected-to-enter-period-of-downswing-in-2022-as-demand-lags-behind-supply-says-trendforce/
- 2021-10-20 - NAND upturn "lasted only two consecutive quarters," 4Q21 -0-5%: https://www.trendforce.com/news/2021/10/20/nand-flash-prices-projected-to-enter-cyclical-downturn-in-2022-due-to-modest-demand-growth-and-competition-for-higher-layer-nand-says-trendforce/
- 2022-02-22 - 4Q21 actual: NAND ASP fell nearly 5%: https://www.trendforce.com/news/2022/02/22/total-nand-flash-revenue-drops-2-1-qoq-in-4q21-due-to-slowing-demand-and-falling-prices-says-trendforce/
- 2022-03-28 - DRAM 2Q22 -0-5%: https://www.trendforce.com/news/2022/03/28/russia-ukraine-pincered-by-russian-ukrainian-war-and-inflation-dram-price-drop-forecast-to-continue-in-2q22-by-0-5-says-trendforce/
- 2022-03-30 - NAND 2Q22 +5-10% on Kioxia/WDC contamination (the inversion): https://www.trendforce.com/news/2022/03/30/nand-flash-pricing-to-gain-510-in-2q22-on-the-heels-of-material-contamination-at-kioxia-and-wdc-says-trendforce/
Not retrieved (flagged, not retried)
- A common-basis $/Gb contract ASP series covering 2006-2008 for both DRAM and NAND. TrendForce's gated
priceChartmember API and paid DRAMeXchange/iSuppli/Gartner feeds remain the only known routes; the 2006-07 leg of this brief therefore rests on two incompatible ASP conventions, which is why its ~1 quarter lead is stated as approximate.