Google Tensor Processing Unit / variant
Currently-shipping Broadcom-designed training TPU; 192 GB HBM3e per chip, 4.6 PFLOPS FP8, 9216-chip superpod [8].
區間 + 信心水準 + 來源；不同分析師、不同時點的預估會並列。
| 期間 | 出貨量 (low–mid–high) | ASP (USD) | 信心 | 來源 |
|---|---|---|---|---|
| 2026 FY | 3000k–3500k–4000k | 0.01M | medium | TheNextWeb 2026-04-22 + Trendforce 2025-11-07 |
每行 = 單一機台的零組件 slot。share % 是該供應商在此 component 的估計份額。
GPU / CPU dies — the headline silicon.
Broadcom-designed; TSMC N3.
HBM stacks attached to the compute die.
192 GB per chip; 8-high stacks.
Advanced packaging (CoWoS-L) + ABF substrate.
