The NVIDIA A100 line listed on 26 August has sold through. The desk can still source NVIDIA A100 off-page: send the quantity, condition and delivery window and we will check the wider market. Details below are the August listing, kept for reference.
An eight-GPU node for less than the price of a single flagship card. Provenance documented and burn-in results available before commitment.

The price above is ex-works. Between the supplier door and your rack sit freight and insurance at node weights, export screening, customs classification and duties, import VAT or GST and its recovery, and the last mile into the facility. We price all of it up front so two offers can be compared honestly. The mechanics are on the import and IOR desk.
Full specification and where this SKU fits is on the NVIDIA A100 page. Everything else verified today sits on the live inventory. If this is not the right SKU for your workload, we will say so.
The A100 stopped being an allocation problem some time ago. What reaches the secondary market today is fleet turnover: hyperscale and large-lab estates retiring HGX nodes as newer silicon lands, with condition verified before resale. Current listed reference: used HGX nodes at $125,000 per 8-GPU node ex-works US, 7 days to hand over.
This changes a buyer's position. You are not queueing behind anyone. When a fleet rotates, whole racks land at once, and sellers want them gone before the next tranche arrives. That gives a prepared buyer with clean KYC and fast sign-off genuine leverage on price and terms, provided they can commit inside the window.
The trap is waiting for a better batch. Rotation is lumpy, so a strong lot can be followed by weeks of nothing. We run KYC before any introduction and contract you directly with the supplier, which is why our first sourcing pass inside one business day matters on this SKU more than most.
The people taking A100 nodes now are cost-driven. Inference shops serving mid-sized models, university and corporate research groups fine-tuning open weights, and teams that need proven CUDA maturity without paying for current-generation headroom. Two generations back, the platform wins on cost per token for those workloads, and the software stack has had years to settle.
Go up a generation if your workload is training-heavy or memory-bandwidth bound: the newer platforms pay for themselves there. Go below the A100 only for light inference or batch jobs where interconnect barely matters. In the middle band, sustained inference and research at scale, the A100's economics are hard to argue with right now.
If that describes your workload, the sensible move is to price a verified used node against your current per-token spend before deciding anything else. We acknowledge within one hour and source across OEM, ODM and distributor channels, covering new, refurbished and cancelled-order stock.
Fleet-retired A100 nodes are generally strong stock: they ran in controlled environments and the units we source have verified condition before any introduction. KYC runs on both sides first, and you contract directly with the supplier, so warranty and condition terms sit in your contract, not in a middle layer.
Used HGX A100 nodes are currently listed at 7 days, ex-works US, at $125,000 per 8-GPU node. Because this stock comes from fleet turnover rather than a production queue, lead time is mostly logistics, not allocation. We acknowledge enquiries within one hour and run a first sourcing pass within one business day.
This line quotes 7 days from commitment, United States-located. Lead times only mean something when the allocation behind them is verified, and every line we publish is confirmed with the holder before it is listed.
Indicative $125,000 per 8-GPU node, ex-works USD. That excludes freight, insurance, importer-of-record, duties and local charges, all of which we price into a landed number before you commit. Volume, term and configuration move it.
Condition is stated plainly and verified before listing. Refurbished and used lines come with documented provenance and burn-in evidence, and a warranty position confirmed for your jurisdiction. If a line cannot be verified, it does not get offered.
Yes. Freight, insurance, customs and importer-of-record are handled by the same desk, subject to export screening on the destination and end user. We price the landed cost up front rather than discovering it at the border.
Twenty minutes with the desk, no pitch and no quote at the end of it. Tell us roughly what you need and we will come back within one business day.
Your enquiry has landed with the desk. Acknowledged within one hour.