06-reference/research

2018 capacity announce asp rollover lag

2026-09-09·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
investingmemory-cyclephase-historycapexdram

The 2018 capacity-announcement → ASP-rollover lag is 12-16 months for DRAM and ~6 for NAND, and it is a floor, not an estimate

The question

Verbatim: "Re-key the 2018 memory cycle's 'Samsung capacity-announcement → ASP rollover' lag interval to calibrate the lead-time between a capex-composition shift and ASP rollover for the current window."

Context: this is the calibration the vault has twice flagged as missing. Two prior briefs state the capex→ASP lag is untested at n=0 cycles, and the 45% shell_share threshold in the capex-composition anchor is explicitly a guess until this interval is measured.

What we already know (from the vault)

What the web says

Convergences and contradictions

Synthesis for RDCO

The measured interval, keyed to Samsung's 2017-07-04 announcement. NAND contract ASPs rolled over in 1Q18, a lag of about 6 months. DRAM contract-price growth stopped after 2Q18 (~12 months), went flat-to-weak in 3Q18 (~12-14 months), and posted its first QoQ decline in 4Q18 (~15 months). Micron's gross margin peaked at FQ-ended 2018-08-30 (~14 months). The honest statement is a bracket: DRAM 12-16 months, NAND 6-7 months, with NAND leading DRAM by roughly three quarters. Keying instead to the vault's labelled capacity-announcement 2017-06 widens the DRAM bracket to 13-17 months. This is n=1. It is not a distribution, and it should never be quoted as "the" lag.

The single most important finding is that this number is a floor for the current window, not a transferable estimate. Samsung's July 2017 money went into a shell that was already standing: Pyeongtaek P1 broke ground in 2015 and began mass production on the same day as the announcement. The 2017 announcement was largely tool-fill and second-phase equipment on built real estate, so announcement→capacity-online was compressed to roughly twelve months. The current window is the opposite shape. Micron's CIP near-doubled to $10,935M while equipment-not-placed-into-service rose only 3.7%, which is buildings-first spending with an 18-30 month tool lag, and the vendors' own guidance dates meaningful new DRAM wafers to mid-2027 through 2028. Applying the 2018 lag unadjusted will call the rollover 2-4 quarters early.

What this does to the current window's dates. Taking the three plausible current keying events and adding the raw 12-16 month DRAM bracket: TrendForce's 2025-11-13 capex print gives Nov 2026 - Mar 2027; Micron's 10-Q period ended 2026-05-28 gives Jun 2027 - Oct 2027; the 2026-07-28 SK Hynix capex raise and the July 2026 Korea greenfield announcement give Jul 2027 - Nov 2027. The union is Q4 2026 through Q4 2027. Adding the shell-first adjustment of 2-4 quarters pushes the center to roughly Q4 2027 - Q1 2028, and an honest uncertainty band on an n=1 calibration is at least ±3 quarters, so Q1 2027 - Q3 2028. That band is wide enough that it cannot by itself trigger a position change. Note that the center lands near the useluminix probability-weighted peak of ~Q3 2027 already in the DIO brief, but those are two weak estimators agreeing, not independent confirmation; do not treat the overlap as corroboration.

Practical implication for the anchors. This calibration is too coarse to set the shell_share 45% threshold, and it does not rescue that threshold from being a guess. What it does justify is a sequencing rule that needs no lag estimate at all: in 2018, NAND rolled over roughly three quarters before DRAM. The Phase-3 tripwire already names "NAND contract prices going negative QoQ while DRAM stays positive" as the highest-value indicator, and TrendForce's 2026-07-30 forecast has the NAND sufficiency ratio turning positive in 2027 while DRAM's gap widens. The 2018 re-keying says that ordering is the one piece of 2018 structure that repeats cleanly and is observable monthly, whereas the capex-announcement lag is neither. Watch the NAND-leads-DRAM spread, not the calendar.

Why this is in the vault

This closes the open follow-up carried in [[2026-06-28-chip-memory-cycle-phase2-phase3-indicators]] ("re-key the 2018 cycle's Samsung capacity-announcement → ASP rollover lag") and supplies the empirical input that [[2026-09-06-vendor-capex-composition-memory-anchor]] said its 45% shell_share threshold was missing. It also adds a dated 2018 evidence row that /investing:label-historical-phases can use to tighten the 2017-06 and 2018-09 transitions in phase-history.csv from fab-date inference to contemporaneous price-commentary evidence, and it establishes that the resulting lag is a floor for the current cycle rather than a usable point forecast.

Open follow-ups

Related

Sources

Vault

Web (fetched and quoted)

Web (cited from search results, body not fetched)

Paywalled / not retrieved (flagged, not retried)