Sizing the subsidy-insulated tail in the 2027-28 memory capacity wave
The question
"Quantify the subsidy share of the 2027-28 wave: what fraction of Micron Idaho/NY, Samsung Taylor, and the Korean clusters is CHIPS/KDB-funded vs maker-funded, since that fraction is the size of the 'capacity insulated from discipline' tail?"
This is open follow-up #2 from [[2026-07-21-memory-maker-fab-financing-supply-side]], which rejected the "external capital marks the end of oligopoly discipline" hypothesis but found a third term the FCF-vs-market binary was blind to: state subsidy. The job here is to size that tail, not re-argue it.
What we already know (from the vault)
- The parent brief's tail-risk register entry, verbatim: "CHIPS (Micron, Samsung-US), KDB sub-market loans (both Korean makers), and CXMT's ~77% state-backed intensity mean a slug of the 2027-28 wafer wave is insulated from both cash-flow discipline and market discipline." It named three channels and quantified none of them at project level. [[2026-07-21-memory-maker-fab-financing-supply-side]]
- The only subsidy figure anywhere in the vault's memory-cycle set is the Korean policy-loan program: ₩10T at ~2.85% (≈1pt below commercial), part of >₩60T policy funding; Samsung applied for ₩2T, SK Hynix for ₩1T. Those are applications, not awards or drawdowns. Micron's capex is quoted "net of anticipated government incentives" — a disclosure convention, not a number. [[2026-07-21-memory-maker-fab-financing-supply-side]]
- Section 48D, the 25%/35% CHIPS investment tax credit, the K-Chips Act, Korea Eximbank and cluster infrastructure spending have zero mentions across the entire memory-cycle brief set. The largest subsidy mechanism in the stack was absent from the vault before this brief.
- The fab-by-fab online-date register is the binding constraint on the window: no greenfield DRAM wafer before H2 2027; material greenfield bits reach market ~Q1-Q2 2028; Micron NY is the only date that slipped (to ~2030) while almost every other date pulled in. [[2026-09-15-dram-greenfield-fab-online-dates]]
- That register already excludes Samsung Taylor on the grounds that it is a foundry fab, not DRAM — so the question as posed contains a category error the vault had already caught. [[2026-09-15-dram-greenfield-fab-online-dates]]
What the web says
The question's three named US/Korean anchors mostly fall out of the window on inspection.
- Samsung Taylor is a leading-edge logic foundry (2nm SF2/SF2P+, Tesla AI5 as anchor customer), not memory. Trial production ~late Q3 2026, customer production 2027. It adds zero DRAM or NAND bits and must be excluded from any memory numerator or denominator. Its CHIPS award was also cut from a ~$6.4B April 2024 preliminary memorandum of terms to ~$4.745B finalized ~2024-12-20 — a useful datapoint on ceiling-vs-award, but irrelevant to memory supply. [SECONDARY: Electrek, TechTimes; commerce.gov returned 403 to both WebFetch and curl-with-browser-UA, so the award figures are not primary-verified.]
- Micron Clay NY is out of the 2027-28 window entirely. Up to $100B over 20+ years, four fabs, first phase $20B by end of decade, groundbreaking 2026-01-16 — but first production has slipped to late 2030, and Micron's own 10-K sequences it "after the completion of the second Idaho fab." [SECONDARY: TrendForce 2025-11-10 and 2026-07-10; the 20+-year/four-fab program shape is PRIMARY from the FY2025 10-K.]
- Micron ID1 (Boise) is in the window but barely. First DRAM wafer output 2H calendar 2027, customer qualification after, and Micron itself guides no meaningful output until 2028. [PRIMARY: Micron FY2025 10-K; SECONDARY for the "no meaningful output" framing.] Micron does not disclose per-fab wafers-per-month for ID1, ID2 or Clay — only cleanroom square footage and dollars. Any kwspm figure for a US fab is a third-party estimate.
The US subsidy stack is far larger than the grant everyone counts, because the tax credit is bigger and nobody counts it.
- Micron signed definitive direct funding agreements with Commerce on 2024-12-09 for up to $6.1B covering one Boise fab and two Clay NY fabs; amended 2025-06-11 to add a second Boise fab with the $6.1B ceiling unchanged (a reallocation, not a cut), plus a separate $275M Manassas VA agreement — $6.4B total CHIPS grants. Funding is milestone-based on construction, tool installation and wafer production, with clawbacks, a 5-year special-dividend restriction and upside-sharing capped at 75% of award disbursements. Critically: "We retain discretion with respect to capacity and production volume ramp of each project." [All PRIMARY: Micron FY2025 Form 10-K, FY ended 2025-08-28.]
- Section 48D rose from 25% to 35%, enacted in the One Big Beautiful Bill Act on 2025-07-04, for qualified property placed in service after 2025-12-31. The credit does not apply to property whose construction begins after 2026-12-31 (48D(e)), and 48D(d) permits an elective payment — i.e. it behaves like cash, not a tax-attribute carryforward. [PRIMARY: Micron 10-K verbatim for the rate change; 26 U.S.C. 48D for the construction deadline and elective payment.]
- Micron treats the credit as cash-equivalent: credits are recognized in receivables or other noncurrent assets as qualified investments are made, with capex-related incentive cash shown as an investing activity "including by constructive receipt." [PRIMARY: 10-K]
- Because ID1 first output is 2H2027 and everything after it is later, essentially all of Micron's US fab capex lands in the 35% window, not the 25% window. That step-up carries no Commerce discretion, no milestone negotiation and no clawback schedule.
The money has started moving, but no project-level disbursement figure exists.
- Micron FY2025 "proceeds from government incentives" was $2,005M, against $315M in FY2024 and $710M in FY2023 — a ~6x jump. Cumulative PP&E reduction from capex-related incentives $5.04B ($3.11B of it on FY2025 expenditures). Incentive receivables $1,572M, other noncurrent assets $914M, unearned incentives $1,018M. Remaining US commitment subject to performance conditions $5,206M — and the ITC is explicitly excluded from that line. [All PRIMARY: FY2025 10-K.]
- All of these are global and commingled across CHIPS, India, Japan and Singapore. A CHIPS-specific disbursed-to-date number is not publicly disclosed at project level. Micron's FY2026 10-K is not yet on EDGAR; the FY26 Q3 10-Q and the 2026-09-30 8-K were not scanned, so any 2026 award renegotiation remains unchecked.
The Korean headline package decomposes into almost nothing that functions as a subsidy to the memory makers.
- The ₩26T package (MOEF, May 2024; expanded to ₩33T, April 2025) breaks into: financial support ₩18.1T structured as up to ₩2T of government capital injected into a state bank by 2027, capitalizing a ₩17T low-interest loan facility; a ₩1.1T Semiconductor Ecosystem Fund whose equity is explicitly earmarked for "materials, parts, equipment, and fabless companies" (structurally unavailable to Samsung/SK Hynix); ₩2.5T infrastructure; ~₩5T fiscal/R&D. Preferential rate is −0.8 to −1.0pp for large companies, putting them at ~3.0% vs ~4% commercial. [PRIMARY: MOEF/MOFE release; note moef.go.kr now 301-redirects to mofe.go.kr.]
- Samsung and SK Hynix together had an outstanding balance of only ₩6.5T (~$4.6B) on the program as of January 2026 (Samsung ₩3.6T, SK Hynix ₩2.9T), up from ₩2.67T a year earlier. That is ~20% of the ₩33T headline and ~4% of a single year of the two makers' combined capex (Samsung >₩110T + SK Hynix high-₩40sT ≈ ₩155-160T for 2026). [SECONDARY: Businesskorea 2026-02-04.]
- K-Chips Act: the national-strategic-technology facility credit for large and mid-size companies rose 15% → 20% (National Assembly, Feb 2025), SMEs 25% → 30%, sunset extended to 2029-12-31. [PRIMARY: korea.kr.] The "25% large-company rate" in circulation appears to be a crossover with the old SME rate. Both makers are deeply profitable in 2026 (SK Hynix Q2 operating margin 76%), so unlike the 2023-24 loss years the credit is now fully usable.
- Infrastructure is the one genuinely large and genuinely unquantified leg. The Special Act enforcement decree in force 2026-08-11 permits state and local government to bear 50-100% of total cluster power/water/road project cost, KEPCO absorbs shared long-haul transmission, and 3GW of phase-one LNG is being built by state generators (EWP, KOSPO, KOWEPO). But the decree directs priority to non-capital-region sites, and Yongin is capital-region (Gyeonggi). No won-denominated commitment to Yongin specifically was found. The only hard number remains the national ₩2.5T line. [SECONDARY: multiple Korean outlets reporting the State Council resolution; 2024-11-27 infrastructure MOU with KEPCO, K-water and LH.]
- The makers are funding the clusters themselves, and the direction of travel is away from policy credit. SK Hynix's total debt fell to ₩18.59T from ₩19.32T during 2026 while capex rose ~70% in H1; it sits in ~$44.8B net cash and raised $26.5B via ADRs in July 2026. Samsung's 2025 actual capex was ₩52.7T (₩47.5T in Device Solutions) against a 2026 plan of >₩110T. The government's own stated denominator is ₩957T of planned private semiconductor investment against a ₩33T mostly-loan package. [Samsung 2025 capex PRIMARY via public disclosure; the rest SECONDARY.]
Convergences and contradictions
- Convergence, and it reverses the parent brief's emphasis. The vault recorded KDB sub-market loans as a live insulation channel on the strength of ₩10T program size and ₩2T/₩1T applications. The web adds the number that matters: ₩6.5T actually drawn, at a ~100bp discount. The annual value of that interest subsidy is
₩65B ($46M) — roughly 0.04% of one year of the two makers' capex. KDB is not an insulation channel at the scale the thesis needs; it is a rounding error. - Contradiction the vault had half-caught. The question names Samsung Taylor and Micron NY as 2027-28 anchors. Taylor is logic, and NY's first wafer has slipped to ~2030. [[2026-09-15-dram-greenfield-fab-online-dates]] already excluded Taylor; it dated NY at ~2030 but kept it in the register. Both belong out of a 2027-28 capacity numerator, while staying in a 2029-31 one.
- Contradiction on the denominator, unresolved. Free sources put global DRAM capacity at ~1.92M wpm (TrendForce fragment) and ~2.25M wpm (TechInsights fragment) — a ~17% disagreement I could not reconcile; the scope difference is my inference, not a sourced explanation. SEMI's own releases return 403 to WebFetch and curl. Per-fab kwspm for US fabs is not disclosed. The reconciled series is paid product (SEMI World Fab Forecast, TechInsights, TrendForce). Every ratio below is therefore a band, not a point.
- One widely-repeated figure failed verification and is excluded. A search summarizer attributed to TrendForce (2025-10-17) the claim that HBM consumes 22%/30% of DRAM wafer input in 2026/2027 while producing 9%/13% of bits. The cited article contains none of those figures. The wafer-to-bit mechanism is nonetheless well-sourced at Micron's own statement that HBM3E consumes ~3x the wafer supply of DDR5 per bit, widening with HBM4 — but the year-by-year percentage split is unsourced and is not used here.
Synthesis for RDCO
The honest answer is that the question's own framing overstates the tail, and the tail that is real sits somewhere the question didn't name. Three corrections have to land before any ratio means anything. Samsung Taylor is a logic foundry and contributes zero memory bits. Micron Clay NY first produces in ~2030. And Micron ID1, the one US memory fab actually inside the window, has first wafer in 2H CY2027 with Micron itself guiding no meaningful output until 2028. So the CHIPS-funded memory capacity landing as bits inside 2027-28 is a sliver — plausibly 15-50k wpm of nameplate running at partial ramp, against global DRAM additions I bottom-up at roughly 435-600k wpm across Korea, the US, Japan, Taiwan and China. That is ~2-5% of 2027-28 incremental wafer additions, with the KDB component at approximately zero. The mechanism the parent brief named as the tail is, inside the window the parent brief cared about, immaterial.
Two separate ratios were being conflated in the question, and they answer differently. The subsidy share of the capex behind in-window capacity is large: for Micron's US fabs I put the federal share at roughly 30-45% of qualified project cost, and the striking part is that this is dominated by the 48D investment tax credit at 35%, not by the $6.4B of CHIPS grants that gets all the attention. (Derivation, stated so it can be attacked: 35% statutory rate applied to a qualified base of 75-95% of total project cost gives 26-33 points, plus a few points of Idaho-weighted grant allocation. The low end assumes a narrow qualified base and thin grant allocation; the high end assumes buildings and tools almost entirely qualify and Idaho carries a disproportionate share of the $6.1B.) For Korea the same figure is roughly 15-28% — a 20% K-Chips facility credit on a 70-90% qualified base, plus an infrastructure-absorption leg of 2-10 points that is the single largest unquantified number in this brief, plus an interest subsidy of effectively zero. But the share of in-window capacity that is subsidy-dependent is a different question, and on that test Korea does not qualify: a 20% credit does not turn a negative-NPV fab positive, SK Hynix cut debt while raising capex 70% and funded Yongin from cash flow and a $26.5B equity raise, and Samsung demonstrably did pause P5 in the downturn and only restarted it in 2026. That is direct evidence that Korean state support did not prevent discipline from operating.
The genuinely insulated leg at scale is Chinese, not CHIPS and not KDB. CXMT at ~320k wpm heading to ~420k by 2027 and YMTC at ~200k heading to ~300k NAND by early 2027 are each adding ~100k wpm inside the window on something the vault already measures at ~77% state capex intensity. Bottom line, stated as a range: 20-35% of 2027-28 incremental memory wafer capacity is insulated from both cash-flow and market discipline, and 15-30 of those points are Chinese state capital. The low end assumes Chinese ramps at the low end of reported targets and global additions at the high end; the high end assumes CXMT reaches 420k and YMTC 300k while global additions come in at the low end of my bottom-up. In bit terms the Chinese share is probably higher still, because CXMT and YMTC are making commodity DDR5/LPDDR and NAND while the Korean additions are HBM-weighted at ~3x the wafer per bit — so the state-funded leg converts to commodity bits far more efficiently than the privately-funded leg. That is directional, not computed; the year-by-year HBM wafer-to-bit split is not available free, and the one circulating figure set failed verification.
For the capital-cycle thesis the practical implication is a timing one, and it cuts in a usefully unexpected direction. Western subsidy is not cushioning the 2027-28 window — it is cushioning 2029-2031, when Clay NY, Yongin Y2 (₩35.2T, cleanroom June 2029), SK Hynix M17 (₩19.1T, cleanroom December 2028) and Samsung P5/P6 land. The 48D step-up to 35%, with its elective-payment cash treatment and its hard 2026-12-31 construction-start gate, is a strong incentive to break ground now on capacity that produces at the end of the decade. So private discipline should still be the operative force on the Korean makers through the 2027-28 down-leg, which keeps the existing physical-anchor tripwire set in [[2026-09-25-hbm-dram-phase-2-3-transition-q3-2026]] valid for dating the turn; the subsidy-insulation risk should be re-sited as a depth-of-the-down-leg and shape-of-the-2030-recovery variable rather than a Phase-3 trigger. This is the same correction [[2026-09-26-financing-led-capital-cycle-phase3-precedent]] made about financing features: right feature, wrong socket.
Calibration. None of the ratios above deserve more than one significant figure. The denominator is paywalled and its two free fragments disagree by 17%; US per-fab capacity is undisclosed; the Korean infrastructure leg has no won figure attached to Yongin; project-level CHIPS disbursement is not published by anyone. The ratios are built from announced ceilings, statutory rates and a bottom-up capacity stack assembled from secondary trade press. The one thing here that is primary-sourced and high-confidence is the mechanism: the 35% ITC is the dominant US subsidy and it was entirely absent from the vault's prior treatment.
Comparison table
Capex in USD unless marked; ₩ converted at ~1,420 KRW/USD (back-solved from sources' own conversions, ±4%). "Govt share" is of announced program capex, not of in-window spend. Confidence applies to the row's weakest load-bearing figure.
| Project | Total announced capex | Announced / awarded government share | Maker-funded | 2027-28-allocated capacity | Tier / conf |
|---|---|---|---|---|---|
| Micron Boise ID1 (DRAM) | Not disclosed per fab | Inside the $6.1B definitive CHIPS award (2024-12-09, amended 2025-06-11) covering ID1+ID2+2 NY fabs — no per-fab split published; plus 35% 48D ITC on qualified capex | Remainder; Micron in $24.4B net cash, no raise disclosed, capex quoted net of anticipated incentives | First wafer 2H CY2027, "no meaningful output until 2028" → partial ramp only; nameplate undisclosed | PRIMARY (award, timing) / high; capacity not disclosed |
| Micron Boise ID2 (DRAM) | Not disclosed | Added to the same $6.1B ceiling 2025-06-11 (reallocation, ceiling unchanged) + 48D | Remainder | Guided late CY2028 → ~0 effective bits in window | PRIMARY / high |
| Micron Clay NY (DRAM, 4 fabs) | up to $100B over 20+ yrs; first phase $20B by ~2030 | Inside the same $6.1B + up to $5.5B NY state, explicitly NON-BINDING term sheet over 20+ yrs + 48D | Remainder | First production ~late 2030 → ZERO in window | 10-K PRIMARY for program shape; $100B + 2030 slip SECONDARY / med-high |
| Samsung Taylor TX | "over $37B" Central Texas | PMT ~$6.4B (Apr 2024) cut to ~$4.745B final (~2024-12-20) | Remainder | LOGIC FOUNDRY — zero memory bits. Excluded. Customer production 2027 | SECONDARY (commerce.gov 403) / med |
| Samsung Pyeongtaek P4 Ph2/Ph4 (DRAM, brownfield) | Ph4 construction contract ₩4.2T (~$3.0B) | 20% K-Chips facility credit; no project-specific grant or loan identified | Overwhelmingly maker; 2026 plan >₩110T (~$73B) | Completion pulled to Apr 2027, tools Q3 2027; >200k wpm of 1c by YE2026 → in-window adds ~60-120k wpm | SECONDARY / med |
| Samsung Pyeongtaek P5 | "minimum ₩50T" (weak) | 20% K-Chips credit | Maker | Mass production late 2028 → ~0 effective by end-2028 | SECONDARY / low-med |
| SK Hynix Yongin Y1 | Cluster ₩600T current plan (was ₩120T/4 fabs in 2019); ~₩120T per fab revised | 20% K-Chips credit; share of ₩6.5T combined drawn policy-loan balance at ~100bp discount; infrastructure 50-100% publicly fundable by statute but no Yongin won figure | Maker: debt fell to ₩18.59T while capex +70%; ~$44.8B net cash; $26.5B ADR raise | Cleanroom Feb-Mar 2027, +60k wpm per cleanroom, +60k every 6 mo, 360k by H1 2030 → ~120-180k wpm effective by end-2028 (my derivation from the sourced cadence) | SECONDARY / med |
| SK Hynix Yongin Y2 | ₩35.2T (board-approved 2026-08-07) | as above | Maker | Construction July 2027, cleanroom June 2029 → ZERO in window | PRIMARY (SK newsroom) / high |
| SK Hynix M15X Cheongju | >₩20T (~$14.6B) | 20% credit; ₩500B KDB low-interest approval Dec 2025 (~$340M) | Maker | ~40k wpm 2026 → ~80-90k wpm 2027 → in-window adds ~40-50k wpm | SECONDARY / med; sources disagree ~2x on the 2026 slope |
| SK Hynix M17 Cheongju | ₩19.1T | as above | Maker | Cleanroom Dec 2028 → ~0 in window | PRIMARY (SK newsroom) / high |
| CXMT (China, DRAM) | Not disclosed | ~77% capex intensity on state capital (vault) | ~23% | ~320k wpm now → ~420k by 2027 → in-window adds ~100k wpm | SECONDARY / low-med |
| YMTC (China, NAND) | Not disclosed | State-backed | — | ~200k wpm → ~300k by early 2027 → adds ~100k wpm | SECONDARY / low-med |
| Context: Micron Hiroshima +25k wpm late 2027 (HBM); Nanya 5A ~60k; Winbond 15k | included in the denominator | SECONDARY / med |
Bottom line. Insulated fraction of 2027-28 incremental memory wafer capacity: 20-35%, of which 15-30 points are Chinese state capital and only ~2-5 points are CHIPS-funded, with the KDB/Korean policy-loan contribution ≈0. Low end: Chinese ramps at the low end of reported targets, global additions at the high end of my 435-600k wpm bottom-up, Micron ID1 contributing 1-2 points. High end: CXMT to 420k and YMTC to 300k, global additions at 435k, ID1 at the top of its ramp. Subsidy share of capex behind in-window capacity is a different and larger number: ~30-45% for Micron's US fabs (ITC-dominated) and ~15-28% for the Korean clusters (credit plus an unpriced infrastructure leg). The band cannot honestly be narrowed without the paid capacity series and a won figure for Yongin infrastructure.
Why this is in the vault
This closes open follow-up #2 of [[2026-07-21-memory-maker-fab-financing-supply-side]] and amends the memory-cycle risk register: the "subsidy-insulated capacity" tail should be re-labelled as predominantly Chinese and re-dated to 2029-31 rather than 2027-28, which means the existing physical-anchor tripwire set in [[2026-09-25-hbm-dram-phase-2-3-transition-q3-2026]] stays valid for dating the Phase 2→3 turn and does not need a subsidy overlay. It also puts the 35% Section 48D credit into the vault for the first time — the largest single subsidy mechanism in the stack, previously absent from every memory-cycle brief.
Open follow-ups
- What is the won-denominated government commitment to Yongin cluster infrastructure specifically, now that the Special Act enforcement decree (in force 2026-08-11) permits state and local government to bear 50-100% of cluster power/water/road cost while directing priority to non-capital-region sites? This is the largest unpriced leg in the Korean column and it is the 2-10 point swing factor in the 15-28% band. Answerable from MOTIE/MOEF budget documents and Gyeonggi/LH disclosures.
- Korea's 2026 tax reform appears to add a domestic-production tax credit for semiconductors distinct from the K-Chips facility investment credit. How large is it, and does it accrue to Samsung and SK Hynix on leading-edge memory output? If it is material it could exceed the 20% investment credit and would raise the Korean subsidy-share band meaningfully.
Related
- [[2026-07-21-memory-maker-fab-financing-supply-side]] — parent brief; this answers its open follow-up #2
- [[2026-09-15-dram-greenfield-fab-online-dates]] — the fab-by-fab online-date register that constrains the 2027-28 window
- [[2026-09-25-hbm-dram-phase-2-3-transition-q3-2026]] — the tripwire scoreboard this brief leaves intact
- [[2026-09-26-financing-led-capital-cycle-phase3-precedent]] — the "right feature, wrong socket" correction this brief parallels
- [[2026-09-06-vendor-capex-composition-memory-anchor]] — Micron PP&E/CIP detail and the "no vendor discloses capex by purpose" caveat
- [[2026-07-15-capex-financing-layer-as-memory-cycle-phase-marker]] — the financing-layer-as-phase-marker test
- [[2026-09-08-cxmt-ymtc-bit-share-phase3-threshold]] — the Chinese state-funded leg this brief finds is the real insulated tail
- [[2026-07-03-cxmt-dram-capacity-memory-oligopoly-risk]] — CXMT state-funded capacity and oligopoly risk
- [[2026-09-29-phase3-memory-cycle-rotation-basket]] — where a re-dated subsidy tail would change instrument selection
- [[2026-05-19-dpa-title-iii-critical-minerals-beneficiaries]] — the vault's method model for a dollar-tagged industrial-policy beneficiary register
Sources
Vault:
- /Users/ray/rdco-vault/06-reference/research/2026-07-21-memory-maker-fab-financing-supply-side.md
- /Users/ray/rdco-vault/06-reference/research/2026-09-15-dram-greenfield-fab-online-dates.md
- /Users/ray/rdco-vault/06-reference/research/2026-07-15-capex-financing-layer-as-memory-cycle-phase-marker.md
- /Users/ray/rdco-vault/06-reference/research/2026-09-06-vendor-capex-composition-memory-anchor.md
- /Users/ray/rdco-vault/06-reference/research/2026-09-26-financing-led-capital-cycle-phase3-precedent.md
- /Users/ray/rdco-vault/06-reference/research/2026-09-25-hbm-dram-phase-2-3-transition-q3-2026.md
- /Users/ray/rdco-vault/06-reference/research/2026-09-08-cxmt-ymtc-bit-share-phase3-threshold.md
- /Users/ray/rdco-vault/06-reference/research/2026-07-03-cxmt-dram-capacity-memory-oligopoly-risk.md
- /Users/ray/rdco-vault/06-reference/research/2026-09-29-phase3-memory-cycle-rotation-basket.md
- /Users/ray/rdco-vault/06-reference/research/2026-05-19-dpa-title-iii-critical-minerals-beneficiaries.md
Primary (US):
- Micron FY2025 Form 10-K, FY ended 2025-08-28: https://www.sec.gov/Archives/edgar/data/723125/000072312525000028/mu-20250828.htm
- 26 U.S.C. 48D: https://www.law.cornell.edu/uscode/text/26/48D
- Micron 8-K EX-99.1 (US fab projects): https://www.sec.gov/Archives/edgar/data/0000723125/000110465925058741/tm2517778d1_ex99-1.htm
- NOT REACHED, flagged: Micron FY2026 Q3 10-Q https://www.sec.gov/Archives/edgar/data/723125/000072312526000015/mu-20260528.htm; Micron 8-K dated 2026-09-30; any 2026 Commerce/NIST release. commerce.gov returned 403 to WebFetch and to curl with a browser User-Agent.
Primary (Korea):
- MOEF/MOFE semiconductor support package release: https://english.mofe.go.kr/pc/selectTbPressCenterDtl.do?boardCd=N0001&seq=5899
- K-Chips Act credit rates: https://www.korea.kr/news/policyNewsView.do?newsId=148939772
- Restriction of Special Taxation Act text: https://www.law.go.kr/LSW/lsLawLinkInfo.do?lsJoLnkSeq=1017558849
- SK Hynix 2026 fab/facility investment: https://news.skhynix.com/en/fab-facility-investment-2026/
- Samsung 2025 capex public disclosure: https://www.samsung.com/global/ir/reports-disclosures/public-disclosure-view.84547/
- NOT OPENED, flagged as the highest-value unexploited primary source: SK Hynix SEC F-1/A (CIK 2120882) https://www.sec.gov/Archives/edgar/data/2120882/000119312526295501/d32785df1a.htm
Secondary (labelled as such throughout):
- Policy-loan drawn balances: https://www.businesskorea.co.kr/news/articleView.html?idxno=262456 ; https://www.businesskorea.co.kr/news/articleView.html?idxno=224911
- SK Hynix KDB approval Dec 2025: https://www.digitimes.com/news/a20251217PD206/sk-hynix-production-hbm-loan-financing.html
- Yongin cadence and SK Hynix 1M wpm target: https://www.trendforce.com/news/2026/06/05/news-sk-hynix-reportedly-to-double-dram-capacity-to-1m-monthly-wafers-by-2030-speeds-yongin-expansion/ ; https://www.techtimes.com/articles/317859/20260606/sk-hynix-dram-capacity-roadmap-revealed-yongin-alone-adds-360k-wafers-monthly.htm ; https://www.thelec.net/news/articleView.html?idxno=11059
- Micron NY delay to late 2030: https://www.trendforce.com/news/2025/11/10/news-microns-new-york-megafabs-reportedly-face-2-3-year-delay-first-now-set-for-late-2030/ ; US investment target raise: https://www.trendforce.com/news/2026/07/10/news-micron-raises-u-s-investment-target-to-250b-through-2035-begins-concrete-work-on-new-yorks-first-fab/
- Micron ID1 timing: https://www.tomshardware.com/pc-components/dram/micron-details-new-u-s-fab-projects-idaho-fab-1-comes-online-in-2h-2027-new-york-fabs-come-later-hbm-assembly-in-the-u-s ; "no meaningful output until 2028": https://overclock3d.net/news/memory/why-micron-doesnt-expect-the-dram-shortage-to-end-until-2028/
- Micron NY $100B announcement (2022-10-04): https://finance.yahoo.com/news/micron-announces-historic-investment-100-133900295.html
- Samsung Taylor is logic, 2027 customer production: https://electrek.co/2026/07/13/samsung-taylor-fab-tesla-ai5-chip-2nm/ ; https://www.techtimes.com/articles/317402/20260531/samsung-taylor-fab-production-confirmed-2027-sf2p-2nm-process-goes.htm
- HBM ~3x wafer per bit vs DDR5 (Micron statement, reported secondhand): https://mlq.ai/news/micron-says-hbm-uses-about-three-times-the-wafer-supply-of-ddr5-and-the-gap-is-widening/
- CXMT/YMTC scale: https://www.tomshardware.com/pc-components/dram/cxmt-close-to-matching-microns-memory-capacity-in-2026-research-claims-would-put-china-on-track-to-become-worlds-second-largest-dram-producer ; https://www.trendforce.com/news/2026/09/24/news-cxmt-ymtc-ramp-memory-capacity-but-chinas-ai-cloud-boom-could-soak-up-new-supply-through-2027/
- Korean cluster infrastructure decree: https://zdnet.co.kr/view/?no=20260803173932 ; https://www.newspim.com/news/view/20260804000342 ; https://www.energy-news.co.kr/news/articleView.html?idxno=208146
- ₩957T private-investment denominator: https://www.seoul.co.kr/news/economy/2026/07/13/20260713500265
- SK Hynix Q2 2026 capex/margin/net cash: https://www.investing.com/news/company-news/sk-hynix-q2-2026-slides-record-revenue-76-operating-margin-93CH-4818489 ; ADR raise: https://futurumgroup.com/insights/will-sk-hynixs-record-265bn-adr-issuance-help-close-its-capex-intensity-gap/
- Samsung 2026 capex plan: https://www.bloomberg.com/news/articles/2026-03-19/samsung-to-spend-73-billion-on-chip-expansion-research-in-2026
- 48D 35% rate and 2026-12-31 construction gate (secondary summary): https://www.frazierdeeter.com/insights/article/how-the-one-big-beautiful-bill-act-impacts-project-financing-for-infrastructure-manufacturing/
- SEMI 300mm memory equipment 2026 (semi.org 403; reachable mirror): https://www.techpowerup.com/350397/semi-projects-300-mm-memory-equipment-investment-to-surpass-usd-50-billion-in-2026
Explicitly NOT publicly available without a paid subscription, and therefore the reason the bottom-line band is wide: reconciled global installed DRAM/NAND capacity in kwspm (two free fragments disagree ~17%); per-fab nameplate kwspm for Micron ID1/ID2/Clay (Micron discloses cleanroom square footage and dollars only); year-by-year incremental DRAM additions in kwspm for 2026/2027/2028; regional share of 2027-28 additions; and HBM share of wafer input vs share of bits by year from a named publisher.