An IT hardware reseller has listed a cluster of 300 decommissioned server racks holding 9,600 Nvidia A100 GPUs on eBay for $59.7 million, Tom’s Hardware reported on 28 October 2026. The seller is Core 4 Solutions, and a screenshot of the listing posted on X on 27 October shows the price as “US $59,700,000.00 or Best Offer”.
In plain words, that is about $6,220 per GPU by our arithmetic, and the price also covers the servers, processors, memory and racks around the chips. Nvidia began shipping the A100 in May 2020. On Silicon Data’s rental index, the same chip rented for $1.56 per GPU-hour on 28 October.
Short explainer · Nvidia A100 cluster brief
A vertical news Short narrated from the published headlines and opening lines of this piece and three related FourWeekMBA pieces on GPU working life and GPU rental prices.
Business Pill · Lead Time
An evergreen Business Engineer explainer on lead time: why the wait for something is part of its price. It explains the general idea only and does not discuss this listing.
The key insight: As we read it, the listing puts a public asking price on a whole cluster of six-year-old Nvidia chips: about $6,220 per GPU including the systems around it, or about five and a half months of round-the-clock rent at Silicon Data’s A100 index, by our arithmetic.
The $59.7M asking price works out to about $6,220 per Nvidia A100 GPU including the systems around it, against about $13,670 for a year of round-the-clock rent at Silicon Data’s $1.56 per GPU-hour A100 index: about 5.5 months of gross rent. Our arithmetic. Sources: Tom’s Hardware, 28 Oct 2026; Silicon Data, 28 Oct 2026.
The listing’s title, as shown in the screenshot, is “300x Rack Cluster of 32x NVIDIA A100 40GB GPU Compute Node Racks, Decommissioned”, and its condition is given as “Used”. The seller appears as “Core 4 Solutions IT Hardware (21163)” with “99.9% positive” feedback, and an image banner on the listing reads “In stock and ready to ship from our MN warehouse”.
Each rack. Four compute nodes and three management nodes, according to Tom’s Hardware.
Each compute node. Two AMD EPYC Rome 64-core processors, eight Nvidia A100 40GB SXM4 accelerators and 1TB of DDR4 memory, Tom’s Hardware reported.
The whole cluster. Up to 9,600 A100 accelerators, 4,200 AMD EPYC Rome processors with 268,800 cores, and up to 1,401.6 TB of memory, by Tom’s Hardware’s count.
Delivery. Free local pickup or FedEx Ground, with $25 two-day or $50 overnight options, Tom’s Hardware reported.
Tom’s Hardware reported that 11 people were watching the listing and that one prospective buyer had it in their cart. The 27 October screenshot shows it “IN 1 CART”.
What the $59.7M asking price works out to per Nvidia A100 GPU, against Silicon Data’s A100 rental index. Per-GPU figures are our arithmetic: $59.7M / 9,600 GPUs (including servers, CPUs, memory and racks), and Tom’s Hardware’s rough $24M estimate for the A100s / 9,600. Rent-equivalent days = per-GPU price / $1.56 per GPU-hour / 24, at round-the-clock use before running costs. Sources: Tom’s Hardware, 28 Oct 2026; Silicon Data, 28 Oct 2026.
What Was Posted on X
300x … 32x A100 Racks for sale for a cool 59.7 million USD..
— Shannon ๐คโ ๏ธ (@cantworkitout) Oct 27 04:45
An X user shared a link to the eBay listing on 27 October, writing that 300 racks of 32 A100s each were for sale for 59.7 million US dollars, and that someone had it in their cart.
A 300-rack AI supercomputer powered by 9,600 Nvidia A100 GPUs is listed on eBay for $59.7M.
Thatโs 1,200 GPU compute servers with 384TB of combined GPU memory and roughly 1.4 petabytes of system RAM.
Each compute server connects eight A100 GPUs through $NVDA’s NVSwitch technology.
More than six years after A100 launched, The seller is offering these systems refurbished for new AI training and inference deployments.
Wall St Engine posted that a 300-rack AI supercomputer with 9,600 Nvidia A100 GPUs is listed on eBay for $59.7M, describing 1,200 GPU compute servers with 384TB of combined GPU memory and roughly 1.4 petabytes of system RAM.
What the Price Works Out To
By our arithmetic, $59.7 million across 9,600 GPUs is about $6,220 per GPU, and across 300 racks it is $199,000 per rack. Those figures include everything in the racks, not only the chips.
Tom’s Hardware gave its own rough estimate of the parts: about $24 million for the A100s, $3 million for the EPYC Rome chips and about $1.1 million for the memory, or about $28.1 million in total before racks, cooling, power supplies and network cards. That estimate implies about $2,500 per A100, and leaves about $31.6 million of the asking price for everything else, by our arithmetic.
For scale, Tom’s Hardware wrote that Nvidia’s newer GB300 NVL72 and VR200 NVL72 racks cost up to $6.5 million and $7.8 million per unit. Those racks hold newer chips and are not a like-for-like comparison. Tom’s Hardware called the A100 “a legitimate budget alternative that can also circumvent the long waitlists and supply chain delays on Nvidia’s newer AI hardware”.
What the Same Chip Rents For
Silicon Data’s A100 Rental Price Index read $1.56 per GPU-hour on 28 October 2026, for its neo-cloud segment. The company describes the A100 as “the workhorse GPU for inference pipelines, fine-tuning, and cost-sensitive training workloads”. On the same page, its H100 index read $2.89 and its B200 index read $6.47.
By our arithmetic, the listing’s price per GPU equals about 3,990 GPU-hours at that rate, or about 166 days, roughly five and a half months, of rent if a chip were rented around the clock. Rented around the clock at $1.56, 9,600 GPUs would bring in about $131 million a year before power, hosting, networking, staff and idle time. Real fleets do not run at full use, and the index prices single GPU-hours from neo-clouds and marketplaces, not whole clusters.
Connecting the Coverage
Nvidia chief executive Jensen Huang said on 21 October that “what people thought used to be a four-year life cycle really is probably eight to maybe more”, as we reported on 21 October. The A100 went into full production and began shipping in May 2020, according to Nvidia. What’s new: a whole A100 cluster now carries a public asking price.
Rental prices for newer chips have been rising. Silicon Data’s H100 index rose from $2.00 in December 2025 to $2.88 on 24 October, as we covered on 24 October, and its chief executive said its GPU forward curve has turned upward this year, as we reported on 25 October. On the Ornn index, B200 rent rose to $9.70 per GPU-hour in two weeks, as we covered on 27 October.
The Structural Read
As we read it, the listing is an asking price with a best-offer option, not a completed sale, so it marks what one reseller hopes to get rather than what the market has paid.
Silicon Data says its index prices single GPU-hours from neo-cloud providers and marketplaces; a buyer of 300 racks would also pay for power, hosting and staff, which the rent comparison leaves out.
Tom’s Hardware’s own parts estimate covers about $28.1 million of the $59.7 million, which leaves about $31.6 million for racks, cooling, power supplies, network cards and the rest, by our arithmetic.
Silicon Data, A100 Rental Price Index page
“The A100 remains the workhorse GPU for inference pipelines, fine-tuning, and cost-sensitive training workloads.”
Three Implications
FOR BUYERS OF COMPUTE Tom’s Hardware described the A100 as a way to avoid waitlists and supply delays on Nvidia’s newer hardware.
FOR GPU LENDERS AND OWNERS Huang said on 21 October that a chip’s working life is “probably eight to maybe more” years; this listing puts a public asking price on chips that began shipping in May 2020.
FOR NEO-CLOUDS Silicon Data’s A100 index read $1.56 per GPU-hour on 28 October, against $2.89 for the H100 and $6.47 for the B200 on the same page.
The framework puts it as “Inference is a revenue engine: continuous, distributed across millions of applications and agents, incurring marginal cost with every query.”
As we read it, that is why an older chip can still carry a price: Silicon Data describes the A100 as a workhorse for inference and fine-tuning, and its index still prices it at $1.56 an hour. The framework was published in March 2026 and does not discuss this listing.
What Is Not Established
We could not read the eBay listing itself: eBay returned an error page and a sign-in page to our fetches. Its details here come from Tom’s Hardware and from the 27 October screenshot. The screenshot also shows “2 available”, which the visible text does not explain. The listing accepts best offers, so the final price, if any, is not known, and neither the listing text we saw nor Tom’s Hardware says who used the cluster before.
Silicon Data’s page also carries an undated answer giving $1.29 per GPU-hour; we use its dated 28 October reading. We did not contact Core 4 Solutions.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
A reseller is asking $59.7 million, or best offer, for 9,600 used Nvidia A100 GPUs in 300 racks, about $6,220 per GPU including the systems around them, by our arithmetic. At Silicon Data’s $1.56 per GPU-hour, that price equals about 166 days of round-the-clock rent per chip, before running costs.
A note on sourcing. The listing details come from Tom’s Hardware’s 28 October 2026 report and a screenshot of the eBay listing posted on X on 27 October; we could not open the listing itself. Rental prices come from Silicon Data’s index page, read on 28 October. Per-GPU, per-rack and rent-equivalent figures are our arithmetic. We did not contact Core 4 Solutions.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
