gpus · available · Listed 2026-09-22

NVIDIA A100 40GB SXM4 Complete GPU Compute Racks (32 GPUs per rack)

Request current availability and a lot-specific quote.

Lot details

Complete 48U NVIDIA A100 compute racks, each built as a Microsoft WCS / ZT Systems generation-A100 rack with four HGX-8 GPU servers and three utility servers. Each GPU server carries eight NVIDIA A100 40GB SXM4 GPUs, two AMD EPYC Rome 64-core processors, 1 TB of DDR4-2933 and eight ConnectX-6 200GbE adapters, giving each rack 32 A100 GPUs, 1.28 TB of GPU memory, 896 CPU cores and roughly 4.67 TB of system RAM. Utility servers add dual 64-core Rome CPUs, 192 GB of RAM and 1.92 TB of NVMe each. Top-of-rack switching (Arista 7060CX-32S plus Ingrasys management switches), WCS rack and row managers, Geist power distribution, rails, cabling and spare fans and power supplies are included as listed in the original bill of materials. Racks are offered FOB Texas at $155,000 per complete rack. The source states they are staged in truckload lots of 20 racks with additional quantity potentially available; current confirmed inventory, single-rack purchase, serial lists, power input requirements and tested status are provided on request. Photos show seller-supplied racks and video stills; they do not establish working condition. Request the original BOM and confirmation of the technical package. The 200GbE NIC rating does not establish a complete 200GbE fabric: the listed Arista 7060CX-32S has 100GbE ports and the inter-rack topology remains unconfirmed.

Quantity

Confirm with us

Reported condition

Used complete racks per supplied BOM; operational status and serial inventory to be confirmed before sale

Location

Texas, United States (FOB)

Manufacturer

ZT Systems (Microsoft WCS design)

Model

GPU server MSF-053000 (4 per rack) and utility server MSF-052999 (3 per rack) in a 48U ZT cabinet

Supplied specifications

Per rack: 32 x NVIDIA A100 40GB SXM4 (HGX-8 A100 assemblies), 14 x AMD EPYC Rome 64-core CPUs, 896 CPU cores, about 4.67 TB DDR4-2933 system RAM (1 TB per GPU server, 192 GB per utility server), 1.28 TB aggregate GPU memory, 32 x NVIDIA ConnectX-6 200GbE PCIe adapters, 2 x Samsung PM983 960 GB NVMe per utility server, 1 x Arista DCS-7060CX-32S, 3 x Ingrasys management/network switches, Microsoft WCS rack and row management hardware, 4 x Geist rack power distribution assemblies, rails, blanking panels, cabling, spare fans and spare power supplies per the supplied bill of materials

Testing / data handling

Not yet tested by Arch Point Recovery. Working status, firmware/BMC state, GPU-server SSD capacity and exact CPU SKU are subject to confirmation with the source.

Pickup / shipping

Seller states FOB Texas and 20 racks per truckload; exact pickup location, quantity and single-rack availability require confirmation.

Asking price

$155,000 per complete rack (asking, subject to confirmation)

Reference: lot-37bf0c202ad0 · Listed 2026-09-22 · Updated 2026-09-22. Details above are supplied lot information; request supporting records for your purchase.

Before purchase

Confirm the exact configuration, included accessories, available quantity, test results, data-sanitization status where applicable, and freight arrangements. Photos alone do not establish testing, warranty coverage or fitness for your application.