The parts of a private AI build, one datasheet at a time.
Every part GPU Smith specifies into a bill of materials, documented the way we use it: measured specifications, the facility requirements it imposes, and what procurement actually looks like. Each sheet cites its sources and carries a verification date.
Datacenter GPU modules and PCIe accelerators: compute, memory, interconnect, power and procurement, stated per part.
NVIDIA H100 SXM
NVIDIA H200 SXM
NVIDIA B200
NVIDIA B300
NVIDIA RTX PRO 6000 Blackwell Server Edition
NVIDIA L40S
Systems & racks
HGX and DGX platforms and rack-scale NVL systems: node-level specifications and the facility requirements they impose.
NVIDIA DGX B200
NVIDIA DGX B300
NVIDIA DGX H200
NVIDIA HGX B200
NVIDIA HGX B300
NVIDIA GB200 NVL72
NVIDIA GB300 NVL72
NVIDIA DGX Spark
InfiniBand and Spectrum-X Ethernet switches, adapters and DPUs: the fabric layer of a validated cluster.
NVIDIA Quantum-2 QM9700
NVIDIA Quantum-X800
NVIDIA Spectrum-X SN5600
NVIDIA ConnectX-7
Research
The pricing data and sizing analysis behind these datasheets, published as sourced articles.
GPU Price Index 2026: H100, H200, B200, GB200 Street Prices
NVIDIA Data Center GPU Pricing: H100 to GB200 Cost Guide
GPU Cloud Rental Prices 2026: H100 vs H200 Cost Comparison
Own vs Rent GPUs: On-Prem vs Cloud AI Cost Comparison 2026
Blackwell vs Hopper: NVIDIA GPU Architecture Comparison 2026
LLM Inference Hardware Sizing: Enterprise GPU Guide 2026
NOTE — These sheets are engineering references, not offers. Pricing and lead times move; the assessment quotes them for a defined build. GPU Smith is an independent firm and is not affiliated with NVIDIA.