Core 4 Solutions Lists 9,600 A100 GPUs on eBay for $59.7M
A decommissioned 300-rack cluster of 9,600 NVIDIA A100 GPUs is priced at $59.7M, about $6,200 per GPU, a data point for used-GPU values behind neocloud debt.
2 min readAI-assisted
NVIDIA's A100, the accelerator listed in the cluster. The photo shows the product, not the seller's hardware.
Image: NVIDIA
Source · Press image from the NVIDIA website, used for news reporting.
Core 4 Solutions has listed a decommissioned 300-rack AI cluster on eBay for $59.7 million, with up to 9,600 NVIDIA A100 GPUs and 4,200 AMD EPYC Rome processors, according to Tom's Hardware on Oct. 28.
A screenshot of the listing posted on X on Oct. 27 shows the price as $59,700,000 or best offer. Tom's Hardware reported that the seller offers free local pickup or FedEx Ground delivery, with $25 two-day and $50 overnight options. It also reported that 11 people were watching the listing and one buyer had it in a cart. The listing terms here rest on Tom's Hardware's reading of the eBay page.
The ask is a vendor listing; the component values are Tom's Hardware's rough estimates, before racks, cooling and networking.
Illustration (AI-generated)
On Voltgrid's arithmetic, $59.7 million over 9,600 GPUs comes to about $6,200 per accelerator, with the servers, memory and racks included. Tom's Hardware put a rough value of $24 million on the A100s alone, $3 million on the EPYC chips and $1.1 million on the DDR4 memory. That is about $28.1 million in components before racks, cooling, power supplies and networking cards, and the outlet called the ask reasonable.
What is in the 300 racks
Each rack holds four compute nodes and three management nodes, Tom's Hardware reported. Every compute node carries two 64-core EPYC Rome processors, eight A100 40GB SXM4 accelerators and 1 TB of DDR4-2933 memory. Each management node has two 64-core EPYC chips and 192 GB of DDR4. The outlet totals the cluster at 268,800 Zen 2 cores.
The processor count follows from that layout. Across 300 racks, the 1,200 compute nodes use 2,400 EPYC chips and the 900 management nodes use another 1,800, which gives 4,200.
Why it matters for GPU financing
The A100 launched in 2020, and the listing is a rare public asking price for a whole used cluster. The question is how fast used accelerators lose value, which sits under the debt that GPU-cloud operators raise against their fleets. As Voltgrid reported, Nebius secured a $3.5 billion facility for data centers and GPUs, and the SpaceX $40 billion GPU debt raise shows the scale of that borrowing.
One ask is not a market price. It is an unaccepted offer from a single seller, and the listing does not say how old the hardware is or what condition it is in. Earlier data points the other way. CoreWeave's CEO Mike Intrator said on the company's second-quarter earnings call in August that it had signed an A100 contract running into 2029, and that pricing for prior-generation SKUs is at or above where it was years ago, Tom's Hardware reported.
Power and facility angle
Tom's Hardware noted that an air-cooled NVIDIA DGX A100 system draws 6.5 kW at maximum load and fits halls designed for roughly 20 kW per rack. It put current GB200 and GB300 NVL72 racks at 120 kW to 140 kW with direct-to-chip liquid cooling. The listing gives no power figure for this cluster, so a buyer would need to size the electrical and cooling load before bidding.
What is not confirmed yet
The reports Voltgrid read carry no comment from NVIDIA, AMD or eBay, and the sale price is unknown. The listing says up to 9,600 GPUs, so the delivered count and the working condition of each node are not independently verified. Whether the cluster sells at, near or well below the ask will be the real signal for used-GPU values.
