Neither fab has disclosed its HBM split, but the displacement ratio is robust: a wafer moved to HBM4 loses ~70-80% of its commodity bit potential whichever ratio you believe
The question
"What share of first-phase wafers at SK hynix Yongin Y1 and Samsung P5 is allocated to HBM vs commodity DRAM, and how many DDR5 bits does each HBM4/4E wafer displace?"
This is the derivative of tripwire #3 in the memory capital-cycle framework. Tripwire #3 fires on greenfield wafer capacity coming online, but a wafer start is not a commodity bit. Converting the 2027-28 online dates into commodity bit supply needs two multipliers: the per-fab HBM share, and the HBM-to-DDR5 wafer-per-bit trade ratio. Sub-question (a) is a disclosure question. Sub-question (b) is an arithmetic question.
What we already know (from the vault)
- This is the vault's own unanswered follow-up, stated as such. [[2026-09-15-dram-greenfield-fab-online-dates]] closes with open follow-up #1: "What share of Y1 and P5 first-phase wafers goes to HBM vs commodity DRAM, and what is the HBM4/4E-to-DDR5 wafer-per-bit trade ratio? Without these, wafer starts cannot be converted into commodity bit supply." Its own register marks Y1's product mix "DRAM incl. HBM4/4E (mix undisclosed)" and P5's "Mixed: HBM, DRAM, NAND, maybe foundry."
- Y1 first-phase shape is sourced; its product mix is not. Cleanroom Feb-Mar 2027, tools Q2 2027, first-phase output ~H2 2027, six cleanroom phases, +60k wafers/month per cleanroom to 360k by H1 2030 (Korean trade press, secondary). P5 is later and softer: shell H1 2027, mass production late 2028, so it contributes zero bits inside the 2027 window ([[2026-09-15-dram-greenfield-fab-online-dates]], [[2026-10-06-memory-fab-subsidy-share-2027-28-wave]]).
- The vault already knows the mechanism qualitatively and has refused to quantify it. [[2026-09-15-dram-greenfield-fab-online-dates]] on the 2026 brownfield step: "it goes mostly to HBM. HBM uses far more wafer area per bit than DDR5, so commodity-bit relief from Step A is small. That is why prices kept rising through it."
- Only one defensible ratio exists in the vault, and one circulating figure set is already blacklisted. [[2026-10-06-memory-fab-subsidy-share-2027-28-wave]] keeps Micron's statement that HBM3E consumes ~3x the wafer supply of DDR5 per bit, widening with HBM4, and explicitly fails verification on the widely-quoted "HBM takes 22%/30% of wafer input while producing 9%/13% of bits" set: "The cited article contains none of those figures." That set must not be reused.
- Tripwire #3's literal threshold is wafer-count-based, not bit-based. "2+ vendors bring greenfield wafer capacity online (not node/packaging), OR Samsung hits ~250k HBM wafer/mo on/ahead of schedule" ([[2026-07-07-dram-hbm-phase2-phase3-early-signals]]). That wording is the reason this brief matters: the threshold counts wafers, and wafers are no longer a stable proxy for bits.
- HBM4 specs in the vault are thin and one is mis-unitted. [[2026-05-18-hbm3e-hbm4-capacity-timeline-phase-b-end]] has 12-Hi and 16-Hi HBM4/HBM4E, HBM4 mass production from ~Feb 2026, HBM4E on 1c in 2027, and "48 Gb per package vs HBM3E's 36 Gb." Those should read 48 GB and 36 GB; as written they are off by 8x. Core-die density, die area, and base-die node are absent from the vault entirely.
What the web says
- (a) Neither company has published a per-fab HBM-vs-commodity split. This is a genuine non-disclosure, not a search failure. TrendForce's Yongin piece, citing The Elec, gives the fab shape (six cleanrooms, equipment install pulled from May 2027 to February 2027, +360k wafers/month by H1 2030, SK hynix DRAM capacity ~550k wafers/month today to ~1M by 2030) and goes exactly as far as "all newly added capacity is currently designated for DRAM production" — no HBM share, and no figure for HBM's share of SK hynix DRAM capacity now or in 2027 (TrendForce, 2026-06-05). Secondary reporting that the first Yongin fab "focuses on next-generation HBM4 and HBM4E" exists but carries no percentage and no primary attribution; one version of it also attaches 321-layer NAND to Yongin, which is inconsistent with the DRAM-only designation above and marks the whole claim as low-grade.
- (a) The only bounded figures are company-level and industry-level, not fab-level. Samsung EVP Kim Taewoo, reported by Reuters: "HBM is expected to account for nearly 30% of the DRAM industry's total wafer capacity next year," up from about 20% currently (TrendForce, 2026-09-29). A separate secondary estimate puts SK hynix at roughly 30% of its DRAM wafer capacity on HBM today, with analysts expecting ~40% by 2027 (single source, uncorroborated).
- (a) Samsung's own HBM wafer trajectory has slipped against the vault's recorded target. The same TrendForce piece carries industry estimates via Seoul Economic Daily that Samsung's average monthly HBM wafer input "could rise nearly 40% from roughly 180,000 wafers this year to around 250,000 in 2027." The vault's figure, vintage December 2025, was 170k rising to 250k by end-2026 ([[2026-05-18-hbm3e-hbm4-capacity-timeline-phase-b-end]]). The 250k milestone has moved out by roughly a year, and the 2026 reading is ~180k, not ~250k.
- (b) Micron is the only producer to have put a number on the ratio. At Hot Chips 2026 Micron said the HBM wafer penalty is widening with every generation and that AI memory uses about 3x more silicon than DDR5 (Tom's Hardware; body text was truncated on fetch and the detail below comes from secondary summaries of it). The comparison is bits-for-bits: roughly three times the wafer supply for the same number of HBM3E and DDR5 bits, and Micron has said HBM4's ratio should be higher. The stated mechanism is bank count, with an HBM4 die carrying 256 banks against 32 on a DDR5 die.
- (b) TrendForce explicitly warns against applying the headline 3:1 uniformly. Its own wafer-revenue analysis incorporates die size, yield rates and per-gigabit pricing, and it says the headline ratio "should not be applied indiscriminately to HBM4, HBM4E, every DDR5 product or every supplier," while expecting the crowding-out effect to intensify as HBM4 adoption grows and die sizes increase.
- (b) The one circulating displacement multiple is trade press, and internally inconsistent. ijiwei, via TrendForce, is cited for "HBM consumes roughly three times as much wafer area as conventional DRAM" and "the wafer area occupied by a single HBM die is roughly equivalent to that required for 3GB of DDR5," from which TrendForce's own writer infers that "conventional DRAM supply could be displaced at two to three times that pace" as HBM goes 20% to 30%. The second quote does not survive contact with the first: an HBM4 core die at 24Gb is 3GB, so "one HBM die occupies the area of 3GB of DDR5" describes a 1:1 area-per-bit ratio, not 3:1, in the same breath as the 3:1 claim. Treat the 2-3x inference as unusable and the sentence as loose writing.
- (b) No vendor, JEDEC document or press source publishes an HBM4 core-die area. JEDEC's HBM4 spec allows 24Gb or 32Gb core dies at 4/8/12/16-Hi. Base dies move to logic: Samsung stacks 1c DRAM on its own 4nm logic base die; SK hynix's base-die node is reported as both in-house 4nm and TSMC 12nm by different outlets. A measured mm² figure would need a TechInsights teardown or an ISSCC paper. No DDR5 16Gb die-area figure surfaced either.
Convergences and contradictions
- Convergence, and it is the honest answer to (a): the split is not disclosed by anyone. The vault's register says "mix undisclosed" for Y1 and "mixed, maybe foundry" for P5; TrendForce/The Elec stops at "designated for DRAM"; no analyst house publishes a per-fab HBM share for free. Every number in circulation is either company-level (SK hynix ~30% of DRAM wafers on HBM) or industry-level (HBM 20% to ~30% of industry DRAM wafer capacity, 2026 to 2027). Anyone quoting a Y1 or P5 percentage today is interpolating from those, and so is the bounded range below.
- Contradiction worth escalating: tripwire #3's alternate leg is reading behind schedule, not ahead. The vault carries "Samsung hits ~250k HBM wafer/mo on/ahead of schedule" with a December 2025 vintage of 170k to 250k by end-2026, and flags it as "effectively untracked at the moment it matters most." The September 2026 estimate is ~180k in 2026 and ~250k in 2027. Same target, roughly a year later, with 2026 growth of ~6% against the ~47% the vault recorded. The leg is not merely untriggered; it has moved away from triggering.
- Contradiction on the ratio's direction of travel, which no source resolves. Micron says the penalty widens with HBM4. But the single largest structural change at HBM4 is that the base die leaves the DRAM fab for a logic foundry, so one of 13 dies in a 12-Hi stack stops consuming DRAM wafer capacity. "Widening in silicon" and "widening in DRAM wafers" are not the same statement, and Micron's framing ("3x more silicon") is the former. This is the main reason not to push the DRAM-wafer ratio much past 4x for HBM4.
The displacement arithmetic (Ray calculation, auditable)
The honest way to compute this is to anchor on the one company-guided quantity (the bits-for-bits ratio r) rather than on die areas nobody publishes.
Step 1 - bits per commodity DDR5 wafer (the reference unit). Assumptions, all Ray-chosen and all contestable: 300mm wafer (70,686 mm² gross), 16Gb monolithic DDR5 die on a 1b/1c-class node, die area 35 mm², 0.93 packing-and-edge-exclusion factor, 90% die yield.
- Gross dies = (70,686 / 35) x 0.93 ≈ 1,878
- Good dies = 1,878 x 0.90 ≈ 1,690
- Bits = 1,690 x 16Gb = 27,040 Gb ≈ 3.4 TB per wafer
- Band across die area 30-40 mm² and yield 85-95%: 2.8 to 4.2 TB.
Step 2 - bits per HBM4 core-die wafer. HBM bits per wafer = 3.4 TB / r. With r = 3.5 for HBM4 (Micron's ~3 for HBM3E, "higher" for HBM4, capped by the base-die offset above): ~1.0 TB per wafer. With r = 4.25 for HBM4E (32Gb core dies, 16-Hi, thinner dies, lower stack yield): ~0.8 TB per wafer.
Step 3 - the displacement answer. Converting one 300mm wafer from DDR5 to HBM4 core dies:
- DDR5 bits foregone: ~3.4 TB
- HBM4 bits delivered: ~1.0 TB
- Net DRAM bits removed from industry supply: ~2.4 TB per wafer, i.e. the wafer loses ~71% of its commodity bit potential.
- Per bit: each HBM4 bit costs ~3.5 commodity bits of foregone supply; each HBM4E bit ~4.25.
- Per stack, the intuitive form: a 36GB HBM4 12-Hi stack consumes the wafer area of ~126GB of DDR5; a 48GB HBM4E 16-Hi stack ~204GB.
Sensitivity, and the one genuinely robust finding here. The whole calculation rides on r, but the two outputs behave very differently. Across r = 3.0 to 5.0, HBM bits delivered swings from 1.13 to 0.68 TB per wafer (a 66% spread), while net commodity bits removed only moves from 2.25 to 2.70 TB (a 20% spread). A wafer moved to HBM loses 67-80% of its commodity bit potential across the entire plausible range of r. For tripwire purposes that is the number to carry, because it survives being wrong about r. Any claim about how much HBM you get does not.
Two further levers. Bits per commodity wafer drift up 15-20% a year with node shrink and the 24Gb monolithic die, so the absolute TB figures re-base annually while the ratios hold. And the base-die question above is unresolved in a direction that matters: if Micron's 3x already nets out the foundry base die, r for HBM4 in DRAM-wafer terms could be nearer 3.2 than 3.5.
Step 4 - mix translation (company-guided input, Ray calculation). On a fixed wafer base, commodity bits scale with (1 - HBM wafer share). Taking Kim Taewoo's 20% to "nearly 30%": 0.70 / 0.80 = commodity bit supply from the same wafers falls ~12.5% in 2027 on mix alone. Industry wafer capacity has to grow ~12.5% next year just to hold commodity bits flat before any node-shrink gain.
Step 5 - what Y1's first phase is actually worth in commodity bits. Y1 phase 1 at +60k wafers/month (secondary-source cadence), expressed in commodity-wafer-equivalents:
- If 100% HBM4: 60k / 3.5 ≈ 17k wpm-equivalent
- If it mirrors SK hynix's company mix at ~35% HBM: 60k x (0.65 + 0.35/3.5) = 60k x 0.75 ≈ 45k wpm-equivalent The undisclosed split is therefore worth a 2.6x range on the first real step of tripwire #3. P5 contributes nothing to the 2027 window at all (mass production late 2028), and its "HBM, DRAM, NAND, maybe foundry" designation will not support a split estimate in either direction.
Synthesis for RDCO
The answer to (a) is that it is not disclosed, and the correct move is to stop asking for a percentage and start carrying a range. Neither SK hynix nor Samsung has published an HBM-versus-commodity allocation for Yongin Y1 or Pyeongtaek P5, and the free analyst layer does not price it per-fab either. The usable bounds are company-level and industry-level: SK hynix at ~30% of DRAM wafers on HBM today with ~40% a 2027 analyst guess, and HBM at 20% of industry DRAM wafer capacity rising to nearly 30% in 2027 on a Samsung executive's own guidance. Y1's first cleanroom is reported as HBM4/HBM4E-leaning, which argues the top of that band, but the only attribution strong enough to cite says "designated for DRAM" and nothing more. That bounds Y1 phase 1 somewhere between 17k and 45k commodity-wafer-equivalents per month, and leaves P5 outside the window entirely.
The answer to (b) is more useful than it first looks, because the robust half of it is the half the tripwire needs. Anchored on Micron's bits-for-bits ratio, one wafer converted to HBM4 gives up roughly 3.4 TB of DDR5 bits and returns roughly 1.0 TB of HBM, so the net removal is about 2.4 TB per wafer, around 71% of the wafer's commodity potential. Run r from 3.0 to 5.0 and the HBM bits delivered move by two-thirds while the commodity bits removed move by less than a fifth. So a framework that only needs to know how much commodity supply a wafer stops producing is on firm ground; a framework that needs to know how much HBM arrives is not. That asymmetry should shape which claims we are willing to make.
For the memory-cycle tripwire list, three concrete changes follow. First, tripwire #3's primary leg counts wafers and should count commodity-bit-equivalent wafers, because an HBM-dedicated greenfield phase delivers roughly a quarter of the commodity relief its wafer count implies. "2+ vendors bring greenfield wafer capacity online" can be satisfied in 2027 by capacity that moves commodity bits almost not at all, which is exactly the trap the brownfield Step A already sprang in 2026. Second, tripwire #3's alternate leg should be rescored as moving away from trigger: the Samsung 250k wafers/month milestone the vault recorded for end-2026 now reads ~180k in 2026 with 250k pushed to 2027, which is a slip of about a year and a 2026 growth rate of ~6% against the ~47% on file. Third, the mix shift itself is a quietly large supply fact that the list does not instrument at all: HBM going 20% to ~30% of industry wafer capacity subtracts ~12.5% from commodity bit supply at constant wafers, which is the same order of magnitude as a year of capacity growth and cuts the other way.
The net effect of all of this is to push the commodity-bit-relief date later rather than earlier, consistent with the vault's existing re-specification of #3 to "first wafer out at 2 of 3 makers, no earlier than Q3-Q4 2027, material bits around Q1-Q2 2028." Nothing here is a position recommendation and nothing here changes a position. It is a measurement change: the tripwire should be read in bits, and the one number worth hard-coding is that an HBM-allocated wafer is worth roughly 0.2 to 0.33 of a commodity wafer in bit terms, about 0.29 at the HBM4 central case.
Why this is in the vault
It closes open follow-up #1 of [[2026-09-15-dram-greenfield-fab-online-dates]], which the memory-cycle framework named as the missing multiplier that prevents greenfield wafer-start dates from being converted into commodity bit supply, and it supplies the unit conversion that tripwire #3 in [[2026-07-07-dram-hbm-phase2-phase3-early-signals]] needs in order to be scored in bits rather than raw wafers. It also delivers a fresh reading on that tripwire's alternate leg, which the vault had flagged as untracked since December 2025.
Open follow-ups
- What is the measured HBM4 core-die area in mm² from a TechInsights teardown or an ISSCC/IEDM paper, and does it corroborate r ≈ 3.5, or push it toward 3.0 or 4.5?
- Does Micron's "~3x the wafer supply, widening with HBM4" count the HBM4 foundry base die inside the ratio, or is it DRAM wafers only? This is the single largest unresolved input and it moves r in the opposite direction to die-size growth.
- What is Samsung's actual HBM wafer input as of Q4 2026 against the 250k/month target? Tripwire #3's alternate leg still has no reading from a primary source, and the best secondary estimate now implies a roughly one-year slip.
- Does any SK hynix primary source (newsroom post, earnings call, regulatory filing) carry an HBM4/HBM4E earmark for Yongin Y1's first cleanroom, as opposed to Korean trade press? That would move Y1's split from "interpolated" to "bounded."
- When does the 24Gb monolithic DDR5 die become the volume commodity die, and what is its bits-per-wafer figure? It re-bases every absolute TB/wafer number in this brief without touching the ratios.
- Should tripwire #3 be formally rescored in commodity-bit-equivalent wafers, and if so what replaces "2+ vendors bring greenfield wafer capacity online" as the threshold? This is a framework amendment, not a research question, and it needs the founder's call.
Related
- [[2026-09-15-dram-greenfield-fab-online-dates]] — parent; the fab-by-fab online-date register and the source of open follow-up #1 that this brief answers
- [[2026-05-18-hbm3e-hbm4-capacity-timeline-phase-b-end]] — parent; HBM3E/HBM4 capacity timeline, the Samsung 170k-to-250k figure now superseded, and open follow-up #2 on the conversion ratio
- [[2026-07-07-dram-hbm-phase2-phase3-early-signals]] — the five-row tripwire list whose #3 this brief re-units
- [[2026-10-06-memory-fab-subsidy-share-2027-28-wave]] — the Micron 3x anchor and the failed verification of the 22%/30% vs 9%/13% figure set
- [[2026-09-25-hbm-dram-phase-2-3-transition-q3-2026]] — the current six-row rescored scoreboard that tripwire #3 lives in
- [[2026-09-09-hbm-plateau-hypothesis-backtest-variant]] — HBM wafer-area penalty as one of the structural-break mechanisms
- [[2026-06-28-chip-memory-cycle-phase2-phase3-indicators]] — original indicator set; "capex composition not level"
- [[2026-09-06-vendor-capex-composition-memory-anchor]] — the tool-move-in leading observable that dates the first greenfield rung
- [[2026-11-01-hbm-fixed-price-offtake-primary-filings]] — HBM offtake structure; the demand-side counterpart to this supply-side split
Sources
Vault:
~/rdco-vault/06-reference/research/2026-09-15-dram-greenfield-fab-online-dates.md~/rdco-vault/06-reference/research/2026-05-18-hbm3e-hbm4-capacity-timeline-phase-b-end.md~/rdco-vault/06-reference/research/2026-07-07-dram-hbm-phase2-phase3-early-signals.md~/rdco-vault/06-reference/research/2026-10-06-memory-fab-subsidy-share-2027-28-wave.md~/rdco-vault/06-reference/research/2026-09-25-hbm-dram-phase-2-3-transition-q3-2026.md~/rdco-vault/06-reference/research/2026-09-09-hbm-plateau-hypothesis-backtest-variant.md~/rdco-vault/06-reference/research/2026-06-28-chip-memory-cycle-phase2-phase3-indicators.md~/rdco-vault/06-reference/research/2026-09-06-vendor-capex-composition-memory-anchor.md~/rdco-vault/06-reference/research/2026-11-01-hbm-fixed-price-offtake-primary-filings.md~/rdco-vault/01-projects/investing/theses/2026-05-17-memory-cycle-v1.md— contains the misstatement corrected below
Web:
- TrendForce, 2026-09-29, "Samsung Sees HBM Taking Nearly 30% of Industry DRAM Capacity in 2027, Up From 20% Currently" — https://www.trendforce.com/news/2026/09/29/news-samsung-sees-hbm-taking-nearly-30-of-industry-dram-capacity-in-2027-up-from-20-currently/ (Kim Taewoo via Reuters; Samsung HBM wafer input via Seoul Economic Daily; ijiwei wafer-area claim)
- TrendForce, 2026-06-05, "SK hynix Reportedly to Double DRAM Capacity to 1M Monthly Wafers by 2030, Speeds Yongin Expansion" — 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/ (attributed to The Elec)
- Tom's Hardware, Hot Chips 2026, "Micron warns HBM wafer penalty is widening with every generation" — https://www.tomshardware.com/tech-industry/semiconductors/micron-says-the-silicon-gap-between-hbm-and-ddr5-is-widening-with-every-generation (body text truncated on fetch; ratio detail taken from secondary summaries, not from the article body)
- JEDEC HBM4 standard coverage (24Gb/32Gb core dies, 4/8/12/16-Hi) — https://www.allaboutcircuits.com/news/jedec-officially-releases-hbm4-memory-standard/
- Samsung HBM4 logic base die on 4nm — https://www.vlsi.kr/en/samsung-hbm4-logic-base-die-en/
- Not consulted / flagged: Digitimes Nov-Dec 2025 P4 allocation reporting is paywalled beyond the headline and was not retried. TrendForce's reconciled wafer-capacity and wafer-revenue series are paid products; the HBM wafer-input-share versus bit-output-share split remains unavailable without a subscription.
Correction logged
[[2026-07-07-dram-hbm-phase2-phase3-early-signals]] is fine, but the memory-cycle thesis stated: "Every HBM wafer also crowds out ~3 commodity DRAM wafers (Micron's stated conversion ratio)." The sentence sat in BOTH live theses -- [[2026-05-18-memory-cycle-v1.1]] (line 32) and [[2026-05-18-nvidia-supply-chain-v1]] (line 36), both at status: framing-pending-founder-deploy-gate -- not only in [[2026-05-17-memory-cycle-v1]], which is superseded-by-v1.1 and audit-trail only. That is not what Micron said and it is not dimensionally possible — one wafer cannot crowd out three wafers. Micron's statement is bits-for-bits: the same bit count needs ~3x the wafer supply in HBM. The correct restatement is "every HBM bit crowds out ~3 commodity bits," equivalently "an HBM wafer yields ~1/3 the bits a commodity wafer would." The error overstates HBM's supply drag by roughly 3x if read literally. CORRECTED 2026-11-09 in both live theses plus the two downstream repetitions ([[2026-07-03-cxmt-dram-capacity-memory-oligopoly-risk]], [[2026-05-15-jane-street-dwarkesh-tour-ai-datacenter]]); the superseded v1 was deliberately left unedited as an audit record.