Datacenter GPU Market Signals
NVIDIA Hopper, Blackwell, Ampere, Ada Lovelace · AMD Instinct MI-series · Intel Gaudi. Training and inference accelerators aggregated across resellers, brokers, and surplus channels.
Data as of 2026-08-29
▶What do these badges mean?
▶What do these badges mean?
Architecture generations determine software compatibility and performance tier. Hopper (H100/H200) is the current-generation datacenter AI platform. Blackwell (B100/B200/GB200) is the 2025 successor — still ramping in supply. Ampere (A100) remains the highest-volume recertified GPU in the market. Ada Lovelace (L4/L40S) uses GDDR6 and targets inference and visualization workloads. Volta and Turing are legacy but still actively traded in the recert market.
Form factor determines system compatibility. SXM accelerator modules require compatible SXM/HGX platforms and are not interchangeable with standard PCIe accelerator cards. PCIe accelerator cards use a standardized interface, but server compatibility still depends on physical dimensions, power, cooling, firmware/BIOS support, and OEM/platform qualification. OAM modules (AMD MI-series, Intel Gaudi) likewise require a compatible OAM baseboard system. Rows tagged DIRECTIONAL ONLY on SXM or OAM entries indicate fewer than 5 observed listings — treat as directional signal only.