NVIDIA Vera Rubin / variant
Prefill-optimized rack with 144 GPUs. 8 EFLOPS, 100 TB fast memory, 1.7 PB/s aggregate memory bandwidth. Paired with NVL72 racks for decode.
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 |
|---|---|---|---|---|
| 2027 H1 | 500–1.2k–2.0k | 5.50M | rumor | Inferred from NVIDIA roadmap |
| 2027 FY | 1.5k–3.0k–5.0k | 5.30M | rumor | Inferred from NVIDIA roadmap |
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.
Tuned Rubin variant with prefill-tuned compute / memory ratio.
88-core Vera CPU.
HBM stacks attached to the compute die.
Higher capacity HBM4 for prefill (16-Hi 36-48 GB stacks).
Advanced packaging (CoWoS-L) + ABF substrate.
Cold plate, CDU, hoses, manifolds.
PSUs, busbars, BBU.
Driver + 上/下橋 MOSFET 整合單封裝；GPU 板端 Vcore 多相電源，板面積約離散方案 1/4、毛利 40–50%+。
800VDC 前端高壓開關，搭配高壓 superjunction；HVDC 電源側櫃 / PSU 前級整流升壓。
PSU / 板端整流、保護用二極體陣列；安世(Nexperia)斷料疑慮下台廠二供轉單標的。
NICs, switches, cables, CPO modules.
Rack, sleds, trays, brackets, heatsink fins.
機櫃滑軌組（tool-less rail / slide），承載運算與交換托盤抽拉維護。
Motherboard, backplane, BMC, signal integrity.
