HBM3E / HBM4 capacity-online timeline through 2026-2027 — and the milestones that end Phase B
The question
What's the precise expected capacity-online timeline for SK Hynix HBM3E/HBM4 (M16 fab), Samsung HBM, and Micron HBM through 2026-2027 — and which production milestones would mark the end of Phase B (cycle-peak) for the memory trade?
Top-scored item in the 2026-05-18 backlog sweep (14/15: fit 5, spec 5, signal 4). Directly load-bearing for the active memory-cycle thesis paper trades (MU + SMH deployed, SNDK + INTC added via smart-money-mirror v1).
What we already know (from the vault)
- [[01-projects/investing/theses/2026-05-17-memory-cycle-v1]] — locks the 3-anchor framework (hyperscaler capex direction / DRAM spot 30d / HBM capacity-online cadence), names MU + SMH as initial positions, and explicitly lists
HBM4 Q2 2026, HBM4E samples H2 2026, HBM4E mass prod 2027as the bullish-cadence baseline. The disqualifying condition reads: ">2-quarter delay across 2+ vendors" on HBM capacity cadence = bearish anchor flip. - [[01-projects/investing/anchors/dram-spot/2026-05-17-trendforce-backfill-notes]] — TrendForce scrape (252 rows, 2021-02 to 2026-04) confirms HBM3E is the highest-magnitude pricing tier in primary-source data (avg ~25%+ on quantified rows vs. low-single-digit DDR4). 2026-YTD has the densest reporting cadence (9.3 articles/month) — supercycle coverage is in full swing, which is itself a late-cycle attention signal.
- [[06-reference/2026-05-16-a16z-charts-of-the-week-memory-ai-adoption]] — names "sextuple operating income in 2026" as the bull claim and flags it as a peak-earnings forward call. The fact that memory-stocks-to-the-moon hit a16z chart-of-the-week is a late-cycle attention marker. Vault carries the cyclicality discipline: 2017-18 NAND glut produced 60-70% drawdowns in 12-18 months, 2022 DRAM did similar.
- Micron Q2 FY26 (Feb 2026) earnings already in thesis: $23.86B revenue +196% YoY, Q3 guide $33.5B at 81% gross margin, capex raised to $20B, 2026 HBM supply fully pre-sold including HBM4 (TrendForce Source 4 in thesis).
- Bear-case operational signals already in vault: SK Hynix lost ~200 engineers to competing venture early 2026; 50,000-worker strike threatening 5-day supply disruption; Samsung beat Hynix to HBM4 shipment (oligopoly discipline cracking).
What the web says
- SK Hynix M16 HBM4 mass production starts February 2026 at M16 (Icheon) and M15X (Cheongju), 4 months ahead of original schedule, opening at ~10,000 wafers/month [Introl/SammyFans/Digitimes Dec 2025]. HBM4E samples ship H2 2026, mass production 2027 on 1c (10nm-class) node [Seoul Economic Daily Apr 2026]. Yongin cluster first fab (next-gen HBM4 / HBM4E) operations begin 2027.
- BUT SK Hynix HBM4 capacity ramp postponed Q2 → Q3 2026 per a Dec 2025 update [Digitimes 2025-12-08, TechPowerUp]. Hynix is also simultaneously slowing the HBM4 ramp to prepare 300+ layer NAND — a hedge that implies internal demand uncertainty.
- Samsung HBM4 mass production starts February 2026 at Pyeongtaek, achieving 11.7 Gbps in internal evaluations. Most allocated to NVIDIA's Vera Rubin (H2 2026 launch). Samsung targets 250,000 wafers/month by end-2026 (+47% vs current 170k) — the most aggressive expansion of the three [TrendForce Dec 2025, KEDGlobal]. Already cleared NVIDIA 12-Hi HBM3E qualification September 2025; HBM4 NVIDIA qualification expected mid-to-late Q1 2026.
- Samsung sold out 2026 HBM supply after starting NVIDIA Q3 2025 shipments; near deal for >30% of NVIDIA's 2026 HBM4 [KEDGlobal Oct 2025, Digitimes Dec 2025].
- Micron HBM 2026 capacity fully booked; HBM4 ramps Q2 2026 per Micron Q2 FY26 earnings. Tom's Hardware roadmap: 12-Hi and 16-Hi HBM4 and HBM4E offerings, 48 Gb per package vs HBM3E's 36 Gb. HBM4E targeted late 2027.
- Micron Hiroshima HBM fab: ¥1.5T ($9.6B), construction starts May 2026, first shipments not until 2028 [Tom's Hardware Nov 2025, DCD]. Taiwan back-end is the HBM bottleneck through 2026; Hiroshima is a 2028+ capacity addition, not a 2026-2027 lever.
- NVIDIA is pushing 16-Hi HBM4 orders and all three suppliers are competing — Samsung's aggressive capacity (+47% wafer growth) is a defensive land-grab for share, which historically marks the end of oligopoly pricing discipline [TweakTown].
- HBM3E and HBM4 prices hiked ~20% ahead of 2026 shift [Digitimes Dec 2025] — pricing power confirmed for entry to 2026, but the price-hike-into-capacity-ramp pattern is the textbook cycle peak signature.
Convergences and contradictions
Strong convergence (vault ↔ web):
- All three makers begin HBM4 mass production approximately February 2026 (Hynix M16/M15X, Samsung Pyeongtaek, Micron Q2). The vault thesis's "HBM4 Q2 2026" baseline is matched at the leading edge (Hynix/Samsung actually 1 month earlier than baseline) but is on-schedule for Micron.
- HBM4E mass production lands in 2027 across all three (Hynix on 1c node, Micron late 2027, Samsung 1γ 6th-gen). The vault thesis's "HBM4E mass prod 2027" baseline holds.
- 2026 capacity is sold out at all three. Confirms the bull-case demand-anchor in the thesis.
Contradiction / new information:
- The vault thesis under-weighted the Samsung capacity-expansion magnitude. Samsung's +47% wafer ramp (170k → 250k/mo by year-end 2026) is the largest capacity addition by any single memory maker in this cycle and the strongest single signal that supply is about to catch demand. The vault thesis names "any major memory maker announces capacity expansion that closes the bottleneck in <18 months" as a disqualifying condition. Samsung's 250k target announced Dec 2025 → by end-2026 = 12-15 months from announcement to capacity online. This is borderline-triggering on the existing disqualifying condition.
- SK Hynix Q2 → Q3 2026 ramp postponement is a 1-quarter delay at one vendor. The thesis disqualifying threshold is ">2-quarter delay across 2+ vendors" — so this is not a trip yet. But pairing it with the engineer-poaching and labor-strike signals already in vault, Hynix execution is materially fragile.
- Micron Hiroshima at 2028 not 2026-2027 is the right read for sizing — that fab does not help the 2026-2027 supply picture. It's a 2028+ supply event, which is exactly when the bear case predicts a glut.
Synthesis for RDCO
When does Phase B (cycle-peak) end?
Working framework: 4 phases — (1) demand surge, (2) capacity announcement, (3) capacity online, (4) down cycle. The vault thesis maps these to its own 3-state phase-marker scheme (Tightness / Plateau / Rollover). Phase B in the founder's framework = "cycle-peak" = the window where (2) is mostly priced in and (3) is starting to be visible in shipped wafers. Phase B ends when (3) — capacity online — passes the demand-growth rate.
Best-evidence Phase B end window: Q3 2026 - Q2 2027.
Three converging signals point to this window:
- Samsung's +80,000 wafer/month of new 1c capacity comes online progressively across 2026; by Q3-Q4 2026 a meaningful slug is shipping. Samsung-led capacity ramps historically inflect pricing within 6 months of online (2018 cycle precedent).
- HBM4 ramps at all three makers concurrent in Q1-Q2 2026 — first year of HBM4 is supply-constrained (Phase B-confirming); second year (2027) is when yields normalize and effective capacity steps up sharply.
- NVIDIA Rubin launches H2 2026 with HBM4 — the demand pull peaks here. Whatever incremental demand Rubin adds is then matched against the Samsung-led capacity expansion landing in late 2026 / early 2027.
The cleanest single milestone that would mark Phase B end: any one of the three makers reports HBM ASP -10% QoQ AND DRAM spot -15% over 90 days in the same earnings cycle, OR Samsung reports its 250k wafer/month target was hit early. Hitting the wafer target early = supply landed faster than the demand curve absorbed it = textbook cycle top.
Concrete production milestones to watch (in priority order)
| Milestone | Vendor | Date | Phase B end signal? |
|---|---|---|---|
| HBM4 mass production starts | Hynix M16 / Samsung Pyeongtaek | Feb 2026 (actual) | No — confirms Phase B in progress |
| Samsung HBM4 NVIDIA qualification | Samsung | Mid-late Q1 2026 | No — confirms three-supplier reality (mild bear pressure) |
| HBM4 Hynix Q3 2026 ramp (delayed from Q2) | SK Hynix | Q3 2026 | Yes if it slips further OR if Q2 2026 earnings show Hynix HBM ASP -10% QoQ |
| Samsung 250k wafer/month target hit | Samsung | End-2026 target | Yes — primary cycle-top trigger if hit early or fully on time |
| HBM4E sampling | All three | H2 2026 | No — confirms HBM4 supply is normalizing |
| NVIDIA Rubin H2 2026 launch | NVIDIA pull | H2 2026 | Indirect — peak demand event |
| Micron HBM Q2/Q3 FY27 ASP report | Micron | Sep-Dec 2026 earnings | Yes if HBM ASP -10% QoQ paired with DRAM spot -15%/90d |
| HBM4E mass production | Hynix / Samsung / Micron | 2027 | Yes if it arrives early — implies supply over-run |
| Yongin cluster first fab online | SK Hynix | 2027 | Yes — adds capacity post-Phase-B-window |
| Micron Hiroshima first shipments | Micron | 2028 | Post-Phase B (down-cycle territory) |
What does this mean for current paper positions?
- MU + SMH paper trades remain on the right side of Phase B for now — HBM4 just ramped (Feb 2026), 2026 capacity is sold out, NVIDIA Rubin H2 2026 is still pulling demand forward. Continue the tranche schedule.
- SNDK + INTC (smart-money-mirror v1) are exposed to the same cycle but at lower beta — SNDK is NAND (later-cycle leverage; benefits from the HBM-crowds-out-commodity-DRAM-and-NAND dynamic); INTC has no meaningful HBM exposure but does benefit from broader semis sentiment via foundry / Gaudi positioning. Phase B end hits these later than MU but they should also be in the quarterly review.
- Tighten anchor watch in Q3 2026 specifically. Samsung's capacity target lands progressively through 2026, and the most likely first sign of Phase B end is either Samsung reporting wafer-target-hit-early OR Hynix HBM ASP turn in Q2-Q3 2026 earnings. Run the memory-cycle-v1 quarterly review on the standard cadence (end of Sep 2026) with extra attention to Samsung's Q3 2026 capex+capacity callout and TrendForce ASP commentary.
- The single fact that would force capitulation to the bull case (long-dated): any hyperscaler signing a 3+ year HBM offtake at fixed price. That has not happened. Until it does, the 2026-2027 window remains a position-trade with a hard exit, not a 5-year hold.
Cross-domain analog (for founder's frame)
This is structurally Snowflake 2022 mid-cycle, not Snowflake 2020 early-cycle. Snowflake's customer-credit-consumption peaked Q4 2022 ~6 months after gross margins hit their max — the analog here is HBM gross margins peaking Q2-Q3 2026, with the cycle top trailing by 1-2 quarters into Q4 2026 / Q1 2027. The phData experience: the contracts-signed signal was the cycle top, not the runway. The same pattern is rhyming in HBM right now (all 2026 capacity pre-sold = peak demand commitment = cycle-top precondition).
Open follow-ups
- What is CXMT's actual HBM3 / HBM3E volume trajectory in 2026, and at what point does it become a meaningful secondary supply source (vs. the 2022-2024 "China DRAM threat" narrative that did not materialize at HBM tiers)?
- What is the actual HBM-to-DDR5 wafer conversion ratio across the three makers in 2026 (the thesis cites Micron's stated 3:1; is this matched at Hynix and Samsung)? Material for commodity-DRAM spot price modeling.
- How does NVIDIA Rubin's per-GPU HBM stack count compare to Blackwell's — and what does the 16-Hi HBM4 push imply for total wafer demand per GPU?
- What does the 2018 cycle look like when re-keyed to "Samsung capacity-expansion announcement → ASP rollover lag" — that historical interval is the single best calibration point for the Q3 2026 - Q2 2027 Phase B end window.
- Are there observable hyperscaler-side optimization signals (MoE adoption rate, KV-cache compression deployment, ternary-weight model count) that could compress HBM-per-FLOP and accelerate Phase B end?
Sources
Vault
- [[01-projects/investing/theses/2026-05-17-memory-cycle-v1]]
- [[01-projects/investing/anchors/dram-spot/2026-05-17-trendforce-backfill-notes]]
- [[06-reference/2026-05-16-a16z-charts-of-the-week-memory-ai-adoption]]
- [[06-reference/2026-05-12-diamandis-innermost-loop-ai-infrastructure-thesis]]
Web (primary and tier-1 trade press)
- Samsung HBM4 February 2026 mass production (SammyFans, Dec 2025)
- Samsung + SK Hynix accelerate HBM4 to early 2026 (Digitimes, Dec 2025)
- SK Hynix completes world-first HBM4 development (SK Hynix newsroom)
- SK Hynix to ship HBM4E samples H2, mass production 2027 (Seoul Economic Daily, Apr 2026)
- South Korea's HBM4 Moment (Introl Blog)
- SK Hynix slows HBM4 ramp, prepares 300+ layer NAND (TechPowerUp)
- SK Hynix delays HBM4 mass production amid NVIDIA Rubin launch (Digitimes, Dec 2025)
- Samsung 50% HBM capacity surge in 2026 (TrendForce, Dec 2025)
- Samsung sells out 2026 HBM supply after NVIDIA Q3 2025 shipments (KEDGlobal)
- Samsung clears NVIDIA 12-layer HBM3E qualification (KEDGlobal, Sep 2025)
- Samsung HBM4 NVIDIA early 2026 validation (TrendForce, Nov 2025)
- Samsung near deal to supply >30% of NVIDIA's HBM4 in 2026 (Digitimes, Dec 2025)
- HBM roadmaps for Micron, Samsung, SK Hynix to HBM4 and beyond (Tom's Hardware)
- Micron plans $9.6B HBM fab in Japan (Tom's Hardware)
- Micron $9.6bn HBM fab at Hiroshima, first shipments 2028 (DCD)
- SK Hynix, Samsung, Micron compete for NVIDIA 16-Hi HBM4 orders (TweakTown)
- Samsung + SK Hynix hike HBM3E prices ~20% ahead of 2026 (Digitimes, Dec 2025)