
$59.7M AI Supercomputer With 9,600 Nvidia A100 GPUs on eBay
Core 4 Solutions is offering a 300-rack eBay supercomputer with 9,600 Nvidia A100 GPUs and 4,200 AMD EPYC Rome CPUs for $59.7 million, plus free shipping.
- By
- Grace Kim
- Filed
- Channel
- AI & Compute
- Read
- 3 min read
By Grace Kim
Core 4 Solutions has listed a 300-rack AI supercomputer on eBay for $59.7 million, with 9,600 Nvidia A100 40GB SXM4 accelerators and 4,200 AMD EPYC Rome processors.
The offer includes free local pickup and free FedEx Ground delivery. Buyers can pay $25 for two-day service or $50 for overnight shipping, including delivery directly to a home.
What is inside the 300-rack system?
Each rack contains four compute nodes and three management nodes. The installation therefore contains 1,200 compute nodes and 900 management nodes.
Each compute node includes:
- Two AMD EPYC Rome 64-core processors
- Eight Nvidia A100 40GB SXM4 accelerators
- 1 TB of DDR4-2933 memory
Each management node includes two EPYC 64-core processors and 192 GB of DDR4 memory. Rome uses the Zen 2 architecture, with 64 execution cores per processor, for a reported total of 268,800 cores.
The report gives total memory as up to 1,401.6 TB, or 1.4 petabytes. Supporting hardware includes Nvidia Mellanox ConnectX-6 SmartNICs, racks, cooling equipment and power supplies.
The eBay headline describes the A100 accelerators as decommissioned, making this installed-capacity sale rather than a new-hardware order.
Why would buyers consider A100 and Rome?
Nvidia and AMD have since moved to Blackwell, Vera Rubin and EPYC Turin. Yet A100, code-named Ampere, and EPYC Rome remain relevant for AI, machine learning and scientific research.
Tom's Hardware editor Zhiye Liu called the A100 “a legitimate budget alternative” to newer Nvidia hardware. He said it could also avoid long waitlists and supply-chain delays affecting newer AI systems.
Nvidia's GB300 NVL72 and VR200 NVL72 racks, based on Blackwell and Vera Rubin respectively, cost up to $6.5 million and $7.8 million per unit. Those figures provide a limited comparison because the A100 cluster mixes compute and management nodes across all 300 racks.
How does the asking price compare with component value?
A rough estimate in the report assigns $24 million to the A100 accelerators, $3 million to the EPYC Rome processors and $1.1 million to DDR4 memory.
That produces a $28.1 million component subtotal before racks, cooling, power supplies and Mellanox ConnectX-6 SmartNICs. The figure is an estimate, not a confirmed valuation, while $59.7 million remains the seller's asking price.
The report does not itemize the difference between components and the full asking price. It also identifies no completed transaction, negotiated discount or confirmed buyer.
Can eBay move a supercomputer at this scale?
The shipment requirement is the clearest commercial complication. Core 4 Solutions offers free local pickup, free FedEx Ground delivery, or paid express delivery to the buyer's home, making physical logistics part of the transaction.
A broker previously sold a 405-foot luxury yacht on eBay for $168 million. That precedent shows the platform has handled an exceptionally large asset, but it does not establish demand for a used AI cluster.
The report expects some expanding AI farms and data centers to see value in the listing. It said 11 people reportedly watched the eBay page, with one prospective buyer already leaving the cluster in an eBay cart.
The report identified no final buyer or sale price. The distinction matters because the $59.7 million figure sets the listing ask, not evidence of a completed market transaction for 9,600 A100 accelerators.
What could determine the outcome?
Zhiye Liu wrote: “The real question is not whether it will sell, but whether the buyer will choose $50 overnight shipping.” His framing puts the unusual delivery choice at the center of the listing's appeal.
A completed sale would transfer 9,600 A100 accelerators at once, giving the buyer a large installed base for AI, machine learning or research. The price test is whether that capacity attracts enterprise buyers despite long waits and supply delays for newer Blackwell and Vera Rubin systems.
Source: Tom's Hardware
More from Grace Kim
Show full bio
Market editor covering people, launches and day-to-day practice in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.
734 articles
Related articles
idc-gpu-accelerated-servers-hit-87-4b-in-q2-up-52-yoy-b59f6f0f
IDC: GPU-Accelerated Servers Hit $87.4B in Q2 2026, Up 52% YoY
amd-outlines-venice-cpus-mi455x-gpus-and-helios-rack-roadmap-383b3463
AMD Outlines Venice CPUs, MI455X GPUs, and Helios Rack Roadmap
carla-latin-america-squeezes-more-from-the-machines-it-has-da7be648
CARLA 2026: Latin America Squeezes More From the Machines It Has
microsoft-commits-billions-to-amd-helios-racks-for-azure-inference-1f4c04ad
Microsoft Commits Billions To AMD Helios Racks For Azure Inference



