Core 4 Solutions, a used IT hardware seller based in Minnesota, recently listed an AI supercomputer-scale cluster on eBay for $59.7 million with free shipping. The listing is no longer active. The product page says the seller ended it early on Oct. 28 because the item was no longer available, but it does not say whether the system was actually sold.
The listing described 300 racks and 9,600 Nvidia A100 GPUs
According to the product title, the system included 300 racks, with 32 Nvidia A100 GPUs in each rack. Broken down further, each rack contained four compute nodes, and each server carried eight Nvidia A100 40GB cards. That adds up to 1,200 servers and 9,600 A100 GPUs in total. Nvidia introduced the A100 as a data center GPU in 2020.
On the CPU side, the compute nodes plus three management servers attached to each rack brought the total to 4,200 AMD EPYC processors, or about 269,000 compute cores. The racks were manufactured by ZT Systems. The hardware had previously been deployed in the data center of a hyperscale cloud operator and was later retired, though the seller did not identify the company.
Larger than the first phase of Meta’s RSC
By GPU count alone, the system was larger than the first phase of Meta’s AI Research SuperCluster, or RSC. Meta said in 2022 that the first phase of RSC used 6,080 A100 GPUs. The eBay listing showed 9,600.
Core 4 Solutions is located in Eden Prairie, Minnesota. On its website, the company says it sells new and refurbished servers and also handles IT asset disposition. Its eBay profile shows 21,217 reviews and a 99.9% positive rating.
How the $59.7 million price compares
Using the total asking price and GPU count as a rough guide, the listing works out to about $6,200 per A100. That figure is only a simple allocation across the full package. The sale was not for standalone GPUs, but for complete servers, networking cards, memory, and racks.
Tom’s Hardware estimated that the core components alone — A100 GPUs, EPYC processors, and DDR4 memory — were worth about $28.1 million. That estimate did not include the racks, cooling, power systems, or networking. On that basis, the asking price was about 2.1 times the estimated value of those core parts. Even so, Tom’s Hardware said the number looked reasonable for a complete cluster.
Older AI hardware may still find buyers
Tom’s Hardware also cited pricing for Nvidia’s newer GB300 NVL72 rack-scale systems, saying a single rack can cost as much as $6.5 million. At that level, $59.7 million would buy roughly nine racks. The comparison is stark: nine racks of the latest systems versus 300 racks built around chips first released in 2020.
As for who would want older GPUs, the report pointed to comments from CoreWeave Chief Executive Officer Mike Intrator. On the company’s August earnings call this year, he said CoreWeave had signed an A100 contract that runs through 2029. He also said average selling prices for older GPUs were flat to higher than a year earlier. If older cards can still be rented out, retired rack-scale systems can still attract buyers.
Tom’s Hardware said the A100 has become a budget option and can also help buyers avoid waiting lines and supply-chain delays tied to newer hardware.


