MU Memory 会快过剩吗?
https://bbs.wenxuecity.com/cfzh/53419.html
I do not know the answer.
I used chatGPT to list new Western fabs (China not included). Normally memory use wafers is about 1/3.
Updated Fab Capacity with Approx. Monthly Wafer Throughput (WSPM) next 3 years
| Company / Fab | Location | Tech Focus | Approx. Monthly Capacity (WSPM) | Expected 300?mm Ops |
|---|---|---|---|---|
| TSMC – Fab?21 (Phase?1) | Arizona, USA | Logic (4?nm/5?nm) | ~20,000 → ~30,000 | 2025–2027 |
| TSMC – Fab?21 (Full campus) | Arizona, USA | Logic (3?nm/2?nm) | ~20,000–40,000+ | 2027–2030+ |
| TSMC – Taiwan Advanced Node Fabs | Taiwan (Hsinchu/Kaohsiung) | Logic (3?nm/2?nm/1.4?nm) | ~40,000–80,000+ | 2025–2028 |
| Samsung – Taylor | Texas, USA | Logic | ~20,000–30,000 | 2026–2028 |
| Texas Instruments – Sherman | Texas, USA | Analog & Power | ~20,000–30,000 | 2025–2028 |
| GlobalFoundries – Fab?8.2 | Malta, NY, USA | Specialty & RF | ~30,000–50,000 | 2026–2027 |
| Micron – Idaho (ID1) | Idaho, USA | DRAM | ~20,000–40,000 | 2027 |
| Micron – Idaho (ID2) | Idaho, USA | DRAM | ~20,000–40,000 | 2028 |
| Micron – Taiwan (P5 acquired) | Taiwan (Tongluo) | DRAM | ~30,000–50,000* | 2H 2027 |
| Micron – P5 (initial) | Taiwan (Tongluo) | DRAM | ~20,000–30,000 (early) | 2H 2027 |
| SK hynix – Yongin Fab?1 | Korea | DRAM/HBM | ~40,000–60,000 | 2027 |
| SK hynix – Cheongju (M15X) | Korea | DRAM | ~30,000–50,000 | 2025–2026 |
| Samsung – Pyeongtaek (P5) | Korea | Memory | ~40,000–60,000 | 2028 |
| SanDisk/Kioxia – Kitakami Fab?2 | Japan | NAND Flash | ~30,000–50,000 | 2025–2026 |
| TSMC / JASM Phase?2 | Japan | Logic | ~20,000–40,000 | 2027 |
| ESMC (TSMC+ Partners) | Germany | Specialty Logic | ~20,000–40,000 | 2027–2028 |
| Intel – Fab 52 & Extensions (Ocotillo) | Arizona, USA | Logic / Intel Foundry | ~15,000–25,000 (current throughput) | Operational today |
| Intel – New Albany “Ohio One” | Ohio, USA | Logic (future) | Not meaningful 2025–2028 | Delayed to 2030+ |

