HBM share varies by methodology (bit-output vs revenue) and by snapshot vs full-year, and sources revise heavily — shown as a labeled table, not a trend line. The two 2025 rows below directly conflict on Samsung.
| Period | SK hynix | Samsung | Micron | |
|---|---|---|---|---|
| 2022 | 50% | 40% | 10% | |
| 2023 | 53% | 38% | 9% | |
| 2025 · point-in-timea | 62% | 17% | 21% | |
| 2025 · alt viewb | 46–49% | 42–45% | 4–6% | |
| 2026Ec | ~50% | ~28% | ~20% |
a — 2Q25 point-in-time snapshot (Micron passes Samsung for #2). b — alternate analyst compile, different methodology; shows much higher Samsung. The two 2025 rows are the source disagreement, shown deliberately. c — TrendForce 2026 bit-output projection (SK hynix down from ~59% in 2025). 2022/2023 = TrendForce annual share. Forecast/estimate-grade.
HBM3E is the 2026 mainstream; HBM4 mass production began end-1Q26 (Samsung Feb'26, SK hynix qual ~done, Micron ramps 2Q26); HBM4E targets sampling 2H26, mass production 2027. Allocation splits for NVIDIA Rubin still firming.
Design-in to NVIDIA accelerators drives HBM allocation and pricing power. Status is news-sourced and dated; the 2026 share column is a forecast.
| Supplier | HBM3E (Blackwell) | HBM4 (Rubin) | 2026 HBM share* | Note | |
|---|---|---|---|---|---|
| SK hynix | Shipping | Qual ~done | ~50% | Lead partner; >half of NVIDIA's 2026 HBM. Customer demand >3 yrs exceeds supply. | |
| Samsung | Qualified, shipping | Mass prod. Feb'26 | ~28% | HBM4 mass production began Feb'26 (11.7 Gb/s); leads Vera Rubin HBM4. Full NVIDIA volume allocation still firming. | |
| Micron | Shipping | Ramping 2Q26 | ~20% | HBM4 ramps 2Q26, targeting mid-tier/inference. Entire 2026 HBM sold out. |
* 2026 HBM share = TrendForce bit-output projection; shifts with final NVIDIA Rubin allocation.
Sequential contract-price change across nine periods (2Q24 → 2Q26E). 1Q26 and 2Q26E are forecasts; earlier periods are best-available figures from quarterly press releases (some revised intra-quarter). Plotted at range midpoints.
The full 9-period series behind the chart.
| Quarter | QoQ % | Basis | |
|---|---|---|---|
| 2Q24 | +13 to +18 | final / actual | |
| 3Q24 | +8 to +13 | forecast; exceeded expectations | |
| 4Q24 | 0 to +5 | forecast; reversed down intra-quarter | |
| 1Q25 | −8 to −13 | forecast (decline) | |
| 2Q25 | 0 to −5 | forecast (near-flat) | |
| 3Q25 | +45 to +50 | revised up from 10–15% as market tightened | |
| 4Q25 | +18 to +23 | revised up from 8–13% | |
| 1Q26 | +90 to +95 | forecast (record) | |
| 2Q26E | +58 to +63 | forecast |
1Q26 per-segment: PC DRAM >+100% (record), Server DRAM ~+90%, Mobile DRAM ~+90%, conventional blended +90–95% (TrendForce Feb'26).
Average NAND contract unless a segment is noted. 1Q24 and 3Q24 not isolated in the PRs reviewed.
| Quarter | QoQ % | |
|---|---|---|
| 2Q24 | +13 to +18 | |
| 4Q24 | −3 to −8 | |
| 1Q25 | over −10 | |
| 2Q25 | +10 to +15 (wafer) | |
| 3Q25 | +5 to +10 | |
| 4Q25 | +5 to +10 | |
| 1Q26 | +55 to +60 | |
| 2Q26E | +70 to +75 |
4Q25 avg modest but Nov monthly hikes ran +20% to >60% MoM (512Gb TLC >65%) as supply tightened. 1Q26 enterprise SSD +53–58% (record). 1Q26 / 2Q26E are forecasts.
Direction and scale, not a normalized series. Bases differ: Micron is clean fiscal-year capex (USD); SK hynix's 2025 figure bundles R&D; Samsung's figures are total semiconductor investment guidance (KRW).
| Vendor / period | Value | Basis | |
|---|---|---|---|
| Micron FY2025 | $13.80B | capex net, full fiscal year | |
| Micron FY2026 guide | >$25B | raised from ~$20B; ~$7B in FQ3 | |
| SK hynix 2025 | ₩36.64T | capex + R&D combined (not pure capex) | |
| SK hynix 2026 guide | ~₩50T | +20–30% YoY; M15X, Yongin, EUV | |
| Samsung semi 2025 | ~₩47.5T | semiconductor capex (approx) | |
| Samsung semi 2026 guide | >₩110T | total semi investment; ~doubling vs 2025 |
Not apples-to-apples — currency and capex-vs-investment scope differ per row. Use for scale/direction only. Micron FY2026, SK hynix 2026 and Samsung 2026 figures are guidance/projection.
What vendors and TrendForce actually state, plus the real wafer-mix figures. Hard inventory-weeks and fab-utilization % are not disclosed.
| Metric | Reading | Note | |
|---|---|---|---|
| HBM demand growth | +130% / +70% | YoY 2025 actual / 2026E forecast | |
| HBM share of DRAM wafers | 19% → 23% | 2025 → 2026E | |
| HBM wafer intensity | ~300% | 1 HBM bit needs ~300% more wafer than DDR5 | |
| HBM revenue / wafer | 3–5× | vs conventional DDR5 | |
| 2026 supply status | Sold out | Micron's entire 2026 HBM sold out; all three tight (1Q26 calls) | |
| SK hynix demand visibility | >3 yrs | customer requirements already exceed supply | |
| Shortage duration (guidance) | → 2027+ | Samsung & SK hynix warn | |
| 1Q26 bit color | ~flat / −10% | SK hynix DRAM bits ~flat QoQ; NAND bits −10% QoQ |
+70% HBM demand (2026), 19%→23% wafer share (2026E) and shortage-through-2027 are forecast/guidance. Per-vendor bit-growth % forecast series not disclosed (model-licensed).
| Company | Focus | Status (May'26) | Note | |
|---|---|---|---|---|
| CXMT / ChangXin | DRAM (DDR4/DDR5, early HBM) | Scaling fast | DRAM share 3.97%→7.67%; ~$4.3B STAR IPO cleared 5/27/26; 1Q26 net profit +1,688% YoY, revenue RMB 50.8B (>7× YoY). | |
| YMTC | NAND (some DRAM) | Expanding | 3rd Wuhan fab; new capacity ~50% DRAM, mass prod H2'26–2027. Listing reported/planned H2'26 — unconfirmed. | |
| Fujian Jinhua (JHICC) | DRAM | Dormant in DRAM | No commercial DRAM since 2018 US restriction; included for completeness. |