Core 4 Solutions Lists 300-Rack AI Supercomputer on eBay for $59.7 Million
A supplier is offering a 300-rack computing system with thousands of Nvidia and AMD chips on eBay for about $59.7 million, according to TechRadar.
According to TechRadar, Core 4 Solutions has listed a large computing installation on eBay with an asking price of $59.7 million. The listing describes 300 server racks filled with graphics accelerators, enterprise processors and supporting equipment for large-scale computing work.
TechRadar reports that the configuration includes 9,600 Nvidia A100 accelerators and 4,200 AMD EPYC processors from the Rome generation. The outlet says each compute node pairs eight A100 GPUs with two 64-core EPYC chips. The listing reportedly advertises free shipping, a detail the outlet highlighted in its headline.
The reporting does not give a full breakdown of the remaining equipment, including networking, storage and power infrastructure. It also does not say what the system was used for or whether it is still running.
The reporting presents Core 4 Solutions as the seller and does not describe the company's history, location or other customers. No buyer or sale has been reported.
The listing puts a very large piece of data center hardware in front of ordinary online shoppers, which is an unusual sight. It offers a view into how surplus AI infrastructure may be sold as newer chips arrive.
Background
The Nvidia A100 is a data center graphics processor introduced in 2020 as part of the company's Ampere architecture. It became a workhorse for training and running machine learning models, and many cloud providers and research groups built large clusters around it. The AMD EPYC Rome line, which arrived in 2019, is a family of server processors based on the Zen 2 design, known for high core counts and wide memory access.
Hardware in this class is typically sold in complete systems rather than as loose parts. A rack-scale installation needs more than chips. It requires power delivery, cooling, high-speed networking between servers, storage, and software to manage the workloads. Running a facility like this can cost far more over time than the purchase price, particularly in electricity and maintenance.
When newer generations of accelerators come to market, older systems often lose value quickly for the largest buyers, who upgrade to get better performance per watt. That creates a secondhand market where older equipment may be sold to smaller operators, universities, startups or companies that need capacity without buying new. Resale prices depend on condition, warranty status, support contracts and whether the buyer can actually operate the hardware.
Export rules and licensing also affect advanced chips in some markets, which can shape who is able to buy and move equipment. Buyers of used high-end hardware typically need to check ownership history, documentation and compliance before taking possession.
Why it matters
The listing is a concrete example of how quickly the AI hardware cycle moves. Equipment that was among the most sought-after accelerators only a few years ago is now being offered in bulk, which signals how fast the industry's baseline is shifting. For buyers, that can mean lower entry costs. For sellers, it highlights the challenge of finding customers for systems that are large, expensive to run and specialized.
The story also raises questions about transparency in online marketplaces. A listing of this size depends on verifiable specifications, clear title and honest condition reporting. Whether a deal of this scale is completed through the platform, or handled separately, may matter to both sides. The case also shows how computing capacity can become a tradable asset, with real consequences for who gets access to it.
Finally, the price itself invites comparison. Buyers will weigh the asking price against the cost of new equipment, cloud rental rates and the expense of building or leasing a facility to house it. That comparison can help readers understand what AI computing actually costs at scale.
What to watch
Watch whether the listing is updated, removed or sold, and whether any buyer or completed transaction is reported. Details on the full system, including networking, storage, warranty status and the origin of the equipment, would clarify what is actually being offered. Independent verification of the component counts would also help readers judge the listing.
Also watch how the secondhand market for data center accelerators develops. If more large installations appear for resale, that could signal broader upgrade cycles among major operators. Any regulatory attention to the sale or movement of advanced chips would be important to follow as well.
Viewpoints
Commentary from our columns: three angles on this story.
The Skeptic open questions and weak points
A listing with a $59.7 million price tag and a free shipping badge is an invitation to ask hard questions. The first is whether the numbers describe one coherent system. Nine thousand six hundred accelerators and 4,200 processors should map cleanly onto the node design described, and it is fair to ask whether they do. Buyers of hardware this size should expect detailed documentation before anything else changes hands. What is the condition of the equipment? How old is it, and has it been run continuously or stored? Who maintained it, and what does a warranty cover at this scale? Shipping a few racks is one thing, but moving 300 of them is a logistics project, not a click. Nothing here suggests bad faith, and the seller deserves a chance to answer. Still, a listing alone cannot tell a serious buyer what they are getting. Skepticism here is not cynicism. It is simply the normal due diligence that any multimillion-dollar purchase requires, especially one involving equipment that may be years old.
The Ledger the business angle
Used data center hardware is a market with its own economics. Older accelerators lose value as newer generations arrive, which is why a system built around A100s can be offered at a fraction of what it once cost. For a buyer with cheap power, a suitable building and a clear workload, that discount may be real. For everyone else, the purchase price is only the start. Electricity, cooling, staffing and networking can cost far more over the life of the equipment, and older chips draw more power for the work they do than newer ones. The depreciation curve is steep, so the question is how long the hardware remains economically useful. Sellers, meanwhile, face a narrow pool of qualified buyers for something this large and specialized. An eBay listing is a notable choice for that reason. It may reach buyers who would otherwise never see such an offer, but it also makes the deal look more like a commodity than the engineering project it really is.
The Builder the engineer's view
From an engineering standpoint, the listing is interesting because the node design is recognizable. Eight GPUs paired with two server CPUs is a familiar pattern, and the real challenge is what surrounds it. The chips are the easy part to describe. The hard parts are the interconnect that lets nodes train large models together, the storage that feeds them data, the software stack that keeps the cluster healthy, and the facility that powers and cools it. A buyer would need to know the networking topology, firmware versions and whether the nodes can be reimaged and updated. Older accelerators can still do useful work for many workloads, especially inference, smaller training runs and research. Whether this installation is practical depends on the details that the summary does not mention. Anyone evaluating it should ask for a full inventory, a sample of benchmark results and a clear plan for running it. Good hardware without that operational context is just a very expensive pile of racks.