$60 million AI rack cluster listed on eBay, with 1.4 petabytes of memory and almost 10,000 Nvidia AI GPUs

Who said those big AI companies have to get all the gear? If you had a cool $60 million hanging around yesterday, you could have picked up an AI supercomputer for yourself, complete with thousands of Nvidia A100 GPUs and AMD EPYC CPUs split over 300 racks. And if you went cheap, you could even get shipping for free.

Core 4 Solutions, a company that sells IT parts to companies like Amazon, Lucasfilm, Microsoft, Meta and even the US Department of Justice, listed the 300-rack cluster on its page this week (via Tom's Hardware).

That listing says each rack includes four compute nodes, complete with eight Nvidia A100 40 GB GPUs, eight Nvidia Connect network interface cards, and two AMD EPYC 64-core CPUs. Each rack also comes with 1 TB of DDR4 memory and three 1U management nodes, complete with two EPYC CPUs. As pointed out by Tom's Hardware, this equates to 9,600 GPUs and 4200 CPUs.

The Nvidia A100 GPU is a now-discontinued enterprise-level GPU built on the Ampere architecture, and it's specifically designed for AI use cases and data analytics. You are getting 6912 CUDA cores per GPU (A little more than the RTX 5070 has), though it's not something one would use for gaming.

The $59.7 million listing ended today, October 28, but it's not clear if the 300 racks were actually sold. If you happen to live in the United States, you could even get the racks sent free shipping if you chose the FedEx Ground or FedEx Home Delivery options. What a bargain.

Core 4 Solutions is selling some servers on its website, but the most expensive of these only fetches $87K. The 300 racks on eBay are almost 700 times more expensive.

Impressively, the $59.7 million rack cluster has a whopping 1.4 petabytes of memory, which is enough for just under 90,000 relatively impressive gaming rigs (assuming 16 GB is enough).

So, if you're wondering why your RAM is so costly, you can thank AI servers (and AI companies, by extension) for sucking up all that sweet, sweet memory. It really puts into perspective just how much AI companies are spending on all their gear—and each other.