Test report DSG-7453 · Rev F · tested October 30, 2026
AI Datacenter InfrastructureDevice under test
Core 4 Solutions Lists 300-Rack AI Supercomputer on eBay for $59.7 Million
Core 4 Solutions lists a 300-rack AI supercomputer on eBay for $59.7 million: 9,600 Nvidia A100 GPUs, 4,200 AMD EPYC Rome CPUs, 1.4 petabytes of memory, with free FedEx Ground shipping.
- Read
- 3 min
- Words
- 599
- Node
- 7nm
- Operator
- Priya Raman
Spec summary
- Core 4 Solutions lists a 300-rack AI supercomputer on eBay for $59.7 million.
- The cluster packs 9,600 Nvidia A100 40GB GPUs and 4,200 AMD EPYC Rome 64-core CPUs, totaling 268,800 cores and 1.4 petabytes of DDR4 memory.
- Core components are estimated at roughly $28.1 million: about $24 million for A100s, $3 million for EPYC chips, $1.1 million for DDR4.
- Shipping options include free local pickup, free FedEx Ground, $25 two-day, and $50 overnight delivery.
- 11 watchers and one buyer with the cluster in a cart had shown interest.

A 300-rack AI supercomputer is listed on eBay for $59.7 million. IT vendor Core 4 Solutions posted the decommissioned cluster on the marketplace, offering free local pickup, free FedEx Ground delivery, or paid options of $25 for two-day and $50 for overnight shipping. The listing includes 9,600 Nvidia A100 accelerators and 4,200 AMD EPYC Rome 64-core processors.
The sale reflects a secondary market forming around prior-generation AI hardware. Nvidia has moved to Blackwell and Vera Rubin architectures, and AMD to EPYC Turin. A100 (codenamed Ampere) and EPYC Rome parts remain capable for AI, machine learning, and scientific research workloads, however. Current-generation racks such as Nvidia's GB300 NVL72 and VR200 NVL72 cost up to $6.5 million and $7.8 million per unit respectively, and carry long waitlists plus supply chain delays. A100-based capacity avoids both problems. That is the pitch, at least. Nobody at Core 4 Solutions has disclosed why the cluster was decommissioned or how heavily the units were used.
What do the specifications include?
Each rack holds four compute nodes and three management nodes. Each compute node contains:
- Two AMD EPYC Rome 64-core processors
- Eight Nvidia A100 40GB SXM4 accelerators
- 1TB of DDR4-2933 memory
Each management node carries two EPYC 64-core chips and 192GB of DDR4 memory. Aggregated across the cluster, that totals 9,600 A100 accelerators and 4,200 EPYC Rome processors. With 64 Zen 2 execution cores per chip, the cluster reaches 268,800 cores. Total memory capacity reaches 1,401.6 TB, or roughly 1.4 petabytes.
Is the price justified?
A component-level estimate suggests the asking price exceeds the sum of its parts. Rough figures put the A100s at about $24 million, the EPYC Rome chips at about $3 million, and the DDR4 memory at about $1.1 million. That is roughly $28.1 million in core components. The remainder would need to cover racks, cooling, power supplies, Nvidia Mellanox ConnectX-6 SmartNICs (network interface cards with offload processing) and assembly. The listing does not itemize those costs, so the gap between $28.1 million and $59.7 million remains unexplained. Buyers will want an audit of hardware condition before committing.
Who buys a supercomputer on eBay?
The venue is unusual but not unprecedented. A broker once sold a 405-foot Gigayacht on eBay for $168 million. Precedent exists, in other words, for eight-figure marketplace transactions. As AI farms and data centers expand, the seller appears to be betting that an enterprise client will treat the listing as a bulk procurement channel rather than a curiosity.
Interest is reportedly modest so far. Eleven people are watching the listing, and one prospective buyer has placed the cluster in a cart. Watching is not bidding, and a cart is not a purchase. Whether the cluster sells at all, and at what price after negotiation, remains an open question.
What about delivery?
Moving 300 racks is a logistics undertaking the listing addresses only through standard shipping options. Core 4 Solutions offers free local pickup or FedEx Ground delivery, with $25 two-day and $50 overnight alternatives also available. Those flat rates sit awkwardly beside the physical reality of transporting hundreds of racked servers, which normally requires freight logistics, rigging, and site preparation. The listing does not describe who bears responsibility for damage, installation, or power commissioning.
The practical questions are the unanswered ones. There is no stated warranty, no disclosure of runtime hours, and no service agreement. For a cluster of this scale, those omissions matter more than the shipping fee.
via Tom's Hardware (Source)
More from Priya Raman
Show full bio
Correspondent covering money, funding and market shifts around hardware layer of artificial intelligence at Die Signal.
467 articles
Same lot · LOT-C204
- DSG-460814nmFoxconn Puts NVIDIA Vera Rubin AI Datacenter Cost at $47 Billion Per Gigawatt
- DSG-474845nmSpaceX Seeks $40 Billion to Buy Nvidia AI Chips
- DSG-986810nmAmazon Seeks to Offload $8bn of Nvidia Chips to Investors
- DSG-767865nmSpaceX in $40B debt talks for Nvidia AI chips: report
- DSG-823214nmUpscale AI launches platform to interconnect rival-vendor AI chips