06-reference/research

micron tongluo fab first dram output date

2026-11-09·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
memory-cycledramcapacity-onlinemicronphase-markers

Micron has never guided a first-DRAM-output date for Tongluo — it guides shipments, and the newest guide (mid-CY2027) is a pull-in

Not a recommendation. No buy, sell, short, or allocation is expressed or implied anywhere below. This fills one row in a pre-registered indicator register.

The question

Verbatim: "What is the guided first-DRAM-output date and capacity for Micron's Tongluo (Taiwan) fab, its largest FY2026 capex item, which is absent from its online-date guidance?"

Context: open follow-up #2 of [[2026-09-15-dram-greenfield-fab-online-dates]], which left the Tongluo row as "Not found / Not found" in the per-fab online-date register behind tripwire #3 of the memory-cycle Layer-1 phase tracker. The parent brief anticipated Micron's FQ4 FY2026 call as the answering event; that call happened 2026-09-30, and the FY2026 10-K landed 2026-10-09.

What we already know (from the vault)

What the web says

The Tongluo row, filled

Field Value Evidence tier
Guided first DRAM wafer output date Not guided. Micron has never published one for this site. Absence, verified across LOI, completion release, FQ4 remarks, FY2026 10-K
Guided meaningful shipments date (newest) mid-calendar 2027 Company-guided, FQ4 FY2026 prepared remarks, 2026-09-30
Prior guides for the same milestone H2 CY2027 (2026-01-17); fiscal 2028 (2026-03-15) Company-guided
Revision direction Pulled in ~1-3 quarters vs the March statement Inference from the three company statements
Implied first wafer output ~Q4 CY2026 to Q1 CY2027 Ray inference — back-solved from mid-CY2027 shipments using Micron's own "a few quarters after initial output" ramp rule plus ~2-3 month DRAM cycle time. Not company-guided. Low-to-moderate confidence; the ramp rule is stated about new fabs and this is a retrofit
Capacity, phase 1 Not guided. Analyst estimate: >10% of Micron's Q4 2026 global capacity Analyst estimate (TrendForce via trade press)
Capacity, full build (P5 fab) Not guided. Reported design max ~50,000 wpm; ~8,000 wpm installed at acquisition Analyst / trade-press estimate
Capacity, full site Second fab ~270,000 sq ft cleanroom, construction from end-FY2026; PSMC's original campus plan was ~120,000 wpm across three fabs Company-guided (cleanroom area, second-fab timing) / trade press (wpm)
Only company-published physical size ~300,000 sq ft existing 300mm cleanroom; +~270,000 sq ft planned Company-guided
Product positioning Leading-edge DRAM "including HBM" — mixed, mix undisclosed Company-guided (Mar 2026 release); the Jan release said only "DRAM"
Classification for tripwire #3 Acquired/retrofit shell, NOT greenfield Vault ruling, [[2026-09-15-dram-greenfield-fab-online-dates]]

Convergences and contradictions

Synthesis for RDCO

The register row fills, and it fills on the earlier side of the staircase without touching the tripwire. The parent brief's three-step model put Step A (brownfield/HBM conversion, live, little commodity relief) in 2026 and Step B (the real greenfield clock) at H2 2027-2028, with #3 flipping on first-wafer-out at two of three makers, earliest Q3-Q4 2027. Tongluo does not belong to either step as written. It is a fourth category the model did not have: acquired, de-tooled, already-permitted 300mm cleanroom, retrofitted rather than built. That category has a materially shorter lead time than greenfield — retrofit started March 2026, shipments mid-CY2027, so roughly fifteen months shell-to-shipment versus the five-plus years ID1 took. It also sits outside tripwire #3 by construction, since #3 is explicitly greenfield-only. The honest read is that Tongluo makes commodity-bit relief arrive slightly earlier and slightly larger than the register implies, while leaving the date on which #3 can legally fire exactly where it was: no earlier than Q3-Q4 2027. The register should carry Tongluo as a non-triggering supply addition — a row that adds bits to the Step-B step without being able to flip the state.

Its product mix is the part that actually matters, and it is the part Micron will not say. The March release's "leading-edge DRAM products, including HBM" is the only mix language on record, and it is deliberately both-ways. The timing is suggestive: Micron's Singapore HBM advanced-packaging facility hits initial output early CY2027, and Tongluo shipments arrive mid-CY2027 — a sequencing consistent with Tongluo die feeding Singapore stacks. That is inference, not guidance. It matters because of the wafer-area asymmetry the parent brief established: HBM consumes far more wafer area per bit than DDR5, so a 50,000-wpm fab pointed at HBM4 delivers a fraction of the commodity-bit relief that the same fab pointed at DDR5 would. Until the mix is known, Tongluo's capacity number cannot be converted into commodity bit supply — the same unresolved conversion that blocks Y1 and P5. This is now the third fab blocked on the same missing ratio, which promotes that open follow-up from a nice-to-have to the single highest-value unanswered question in the supply leg.

The revision direction is the cleanest signal in this brief, and it reinforces a pattern already logged. The parent brief's third finding was that almost every date in the set has been pulled in since announcement, which it read as the late-Phase-2 defensive land-grab shape. Tongluo is a new, independent instance: fiscal 2028 in March became mid-calendar 2027 in September, and Micron spent $1.8B to buy a shell rather than pour one — the most explicit "we cannot wait for concrete" purchase in the whole register. Buying capacity instead of building it is a stronger land-grab tell than accelerating a build. The symmetric watch item is also now live and dated: Micron said second-fab construction starts by the end of fiscal 2026, i.e. by roughly September 2026. Whether that pour actually happened is checkable in the FQ1 FY2027 call, and a slip there would be the first discipline signal available from Micron rather than from Samsung.

For the capital-cycle frame this is duration information, not direction information, and it is small. One retrofit fab moving a shipments date by a quarter does not move a position horizon. It modestly widens the band of outcomes in late 2027 and modestly raises the eventual overshoot, consistent with how [[2026-10-01-micron-fq4-fy2026-dram-hbm-tripwire-rescan]] read the FY2027 capex doubling. The scoreboard verdict on #3 is unchanged: not fired, and nothing here can fire it.

Why this is in the vault

It closes open follow-up #2 of [[2026-09-15-dram-greenfield-fab-online-dates]] by filling the Tongluo row of the per-fab online-date register that tripwire #3 of memory-cycle-v1.1 is scored against, and it proposes one structural change to that register: a fourth fab category (acquired/retrofit) that adds bits without being able to trigger a greenfield-defined tripwire. It also retires the vault's stale "investors.micron.com returns 403 on all paths" access note.

Open follow-ups

Related

Sources

Primary — company-guided (Tier A, retrieved 2026-11-09)

Secondary — analyst and trade press (Tier C, explicitly labelled as such)

Vault

Blocked / not obtained (flagged, not retried)