A decommissioned supercomputer is listed on eBay: 9,600 A100 GPUs, $60 million, free shipping included.
That works out to $6,250 per accelerator.
The listing is why the link spreads through founder group chats at 1 a.m., because hardware built for a scale most companies never reach now sits behind a Buy It Now button.
The price covers the accelerators.
A machine of that size also needs the power feed, the cooling loop, the floor space, the spares, and the people who keep all of it alive.
Ownership carries a second clock.
At $0.15/hr for an RTX 3090, the same $60 million buys 400 million GPU hours.
Keep all 9,600 cards busy and that budget is gone in under five years of continuous running, before a single kilowatt-hour of electricity is paid for.
So the $60 million buys a wager: that your workload stays large, predictable and fully used for years.
Most small teams live the opposite life.
A two week experiment, a quiet month, then a surge the week before a launch.
Affordable compute reverses the order.
You buy hours when there is work to run, and the spending stops the moment the run stops.
Buy Now Pay Later makes it zero upfront, with payment after one month.
What is the oldest GPU still doing real work in your stack?
https://ai.bhkcloud.com/?utm_source=linkedin&utm_medium=social&utm_campaign=bhk_social_2026w44&utm_content=evening
Originally posted on linkedin_personal