06-reference/research

nand leads dram rollover ordering n3

2026-11-03·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
investingmemory-cycletripwiresnanddram

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)

What the web says

Convergences and contradictions

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

Related

Sources

Vault

Web (fetched and quoted)

Web (cited from search results, body not fetched)

Web (TrendForce press releases, URLs carried in the vault anchor CSV with title and summary quote)

Not retrieved (flagged, not retried)