NVIDIA DSX AI Data Center / variant
Transportable DSX AI-factory concept: a compact container shell carrying DSX-ready compute racks, power intake/conditioning, liquid-cooling loop, and DSX Exchange telemetry for grid-aware operations. NVIDIA sources describe DSX as a full-stack AI-factory reference design and DSX Flex / Exchange as the grid and operations integration layers; the container form factor is modeled for /explore visualization. [1][2][5]
區間 + 信心水準 + 來源；不同分析師、不同時點的預估會並列。
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.
High-density accelerated compute rack payload following NVIDIA DSX Reference Design coverage of compute, networking, storage, facilities infrastructure, and hardware cluster design. NVIDIA Taiwan names Dell, HPE, Lenovo, Supermicro, ASUS, Foxconn, Gigabyte, Pegatron, QCT, Wistron, and Wiwynn as builders of NVIDIA DSX-ready systems. [1][5]
Cold plate, CDU, hoses, manifolds.
Liquid-cooling distribution and plant interface for DSX AI factories. NVIDIA describes MaxLPS as combining 45C liquid cooling and in-rack performance/Watt to run up to 40% more GPUs at an efficient operating point. [4][5]
PSUs, busbars, BBU.
Power intake, conversion, and control path connected to DSX Flex, which bridges AI factories to grid signals, utility sources, renewables, storage, load shedding, demand response, and pricing events. [1][5]
NICs, switches, cables, CPO modules.
Operations layer integrating compute, network, energy, power, cooling-plant, IT/OT, and operations-system signals for DSX AI factories. [1][5]
Rack, sleds, trays, brackets, heatsink fins.
Transportable enclosure, doors, rack rails, cable/piping chases, and external utility hook-up modeled for /explore. Public DSX sources describe the AI-factory stack and partner infrastructure; the container shell is a visualization packaging assumption.
