November 2026 · Refreshed monthly · Every line carries its source and date
Observed estimates from published market sources. These are not quotes, offers, or bids — RackMind never posts its own prices. Final prices are set in private, negotiated deals.
the workhorse card of the last AI wave
Published buyer band: $5,500 – $7,500 per module — a national ITAD firm's published AI hardware buying bands, November 2026
Index line: none — the October 2026 price index does not publish a line for this card
Index reference: $105,160 per 8-GPU server, used (≈ $13,145 per GPU) — a published institutional GPU price index, October 2026 edition
Published buyer band: $9,000 – $12,000 per module — a national ITAD firm's published AI hardware buying bands, November 2026
Index reference: $217,603 per 8-GPU server, used (≈ $27,200 per GPU) — index, October 2026 edition
Buyer band: $18,000 – $26,000 per module — ITAD bands, November 2026
Index reference: $321,483 per 8-GPU server, used (≈ $40,185 per GPU) — index, October 2026 edition
Buyer band: $28,000 – $36,000 per module — ITAD bands, November 2026
Index reference: $508,484 per 8-GPU server, NEW (≈ $63,561 per GPU) — index, October 2026 edition
Used band: none published by either source yet — used supply of this generation barely exists
Two honest sources rarely agree — one is an index of reference values, the other is a buyer's published band. The gap between them is real, and it is exactly why placement matters: the same fleet is worth different money to different buyers.
Month over month, per the published institutional GPU price index, October 2026 edition:
Heating up: H200 and H100 reference values rose fastest.
Cooling: the GB300 rack reference eased.
This page is the market note — refreshed monthly, dated, and sourced. When the bands move, this page moves.
A GPU fleet's value is not the badge on the box. It is the card, the memory, the shape (SXM or PCIe) — and, more than most sellers expect, the paperwork.
The numbers on this page are real: an A100 80GB module trades in a published band of $9,000–$12,000, and a full 8-GPU A100 server carried an October 2026 index reference of $105,160. The A100 40GB, the workhorse of the last AI wave, trades at $5,500–$7,500 a module.
Why documented fleets sell for more: buyers price unknowns as risk. A fleet that arrives with full configurations, test results, photos, and serials (released after a signed agreement) removes the gamble — so buyers bid on the hardware instead of the fear.
We have watched the same effect in plain servers: a Dell R440 sold at an Ohio county auction for $210 with no spec attached, while spec'd R440s typically bring about $1,114. Same badge. The difference was documentation.
If you are retiring a fleet, document it before you take it to market. That is what the RackMind Certification Report is for.
