We buy GPU clusters, nodes & AI servers at fleet scale.
HPC clusters, GPU nodes, multi-node clusters, GPU servers and AI servers — bought at chassis and cluster level by a team that prices the accelerators individually. NDA on request, data erasure, freight at our cost.
From a single node to a multi-rack cluster
If it accelerates compute and it is coming out of a rack, it is in scope. Mixed vendors and mixed generations are normal, not a complication.
HPC Clusters
Research and simulation clusters retired at end of grant or end of life. Head nodes, compute nodes, InfiniBand fabric and switches priced together, including the racks if you want them gone.
GPU Nodes
Single 2U/4U accelerated nodes — 4-way, 8-way, SXM or PCIe. Priced by accelerator, host platform and NVLink topology rather than as generic used server weight.
GPU Clusters
Full accelerated pods with their networking. We take the fabric, cabling, PDUs and management switches as part of the lot so you are not left with the unsellable half.
Multi-Node Clusters
Multi-rack deployments across several sites or cages. One quote, one BOL set, staged pickups if your decommission runs over weeks rather than a single day.
GPU Servers
Dell PowerEdge, HPE ProLiant, Supermicro, Lenovo ThinkSystem, Gigabyte and whitebox chassis. Populated, partially stripped, or as bare chassis and rails.
AI Servers
Training and inference platforms including DGX and HGX systems. We buy working fleets, mixed-condition fleets, and units pulled for RMA that never went back.
The silicon we are buying now
Current demand, strongest first. Anything not listed still gets a real number — this is where the volume is, not the limit of it.
NVIDIA Datacenter
- H100SXM5 and PCIe, 80GB — HGX baseboards bought whole
- H200141GB HBM3e, SXM and NVL
- B200Blackwell SXM, HGX B200 trays
- A10040GB and 80GB, SXM4 and PCIe
- V10016GB and 32GB, SXM2 and PCIe
- L40S48GB, the usual inference and Omniverse pull
- L4 Tensor Core24GB single-slot, dense inference fleets
- DGX SparkGB10 desktop AI systems, boxed or bare
AMD Instinct
- MI300XHBM3 OAM modules and full UBB trays
- MI300AAPU configurations, socketed nodes
- MI325X256GB HBM3e OAM, whole 8-way baseboards
- MI250XOAM, the Frontier-class HPC parts
- MI21064GB PCIe, dual-slot HPC and render nodes
- MI10032GB PCIe, earlier CDNA compute fleets
- MI50 / MI60Vega 20 compute, sold by the tray
Prosumer & Workstation
- RTX 509032GB Blackwell, render and dev workstations
- RTX 409024GB — A high volume card we buy
- RTX 508016GB Blackwell, mid-tier build refreshes
- RTX 4080 SUPER16GB, common in seat-based render farms
- RTX 4070 SUPER12GB, fleet workstation refreshes
- RTX A600048GB Ampere workstation, NVLink pairs
- RTX A5000 / A400024GB and 16GB, dense seat deployments
Selling one or two cards rather than a fleet? Get an instant offer from the catalog instead — it takes about ten seconds.
Manifest to payment in four steps
One dedicated buyer end-to-end. Your ops team stays hands-off.
1.Send your manifest
Model list, node counts, and condition. Spreadsheets, CSVs, or asset-management exports all work.
2.Firm offer in one business day
Line-item valuation at current market pricing, locked for 14 days. No auctions, no consignment.
3.We handle logistics
De-racking, packing, palletization, and insured freight — coordinated with your colo or facility team.
4.Paid on receipt
Wire or ACH within 3 business days of intake, with a serialized audit of everything received.
Send asset exports, rack elevations, or CSV inventories.
Drive erasure on request, optional NDA, and driveless quotes if the media never leaves your site.
Lots over $5,000 ship freight at our cost. We arrange palletization, BOL, insurance, and dock scheduling.
W-9, certificate of insurance, and purchase-order transactions supported. ACH or wire on your terms.
Need more help?
Have a question that wasn’t covered above? Send us an email and we’ll get back to you within one business day — often much sooner.