Issue 01 / GPU Market ReportJune 2026

The secondary GPU market, split in two

Where Hopper, Ada, and Ampere are actually trading, and why who you buy from matters more than the model.

By Ryan Hammett, ClustermeltData as of June 1, 2026Issue 01
Market read

What the market is doing

Into the Blackwell ramp, the secondary market is splitting in two. Hopper holds a firm price floor that sorts genuine inventory from counterfeit risk, while Ampere has become the abundant, deeply discounted workhorse of the channel. A handful of inference-class cards are resisting depreciation entirely. The result is a market where the right number depends less on the model than on who you buy it from.

active catalog/ researched ranges, single source class

Generation bands

Hopper

2 models

$22,000to$35,000

Ada

4 models

$2,300to$8,000

Ampere

6 models

$700to$15,000

$700$35,000

$700

Entry point

$35K

Top of range

12

Models priced

Headline findings

What stood out this month

The signal behind the numbers. The full table and the by-generation read sit in the report below.

01 /

Authenticity sets the floor

Genuine Hopper inventory holds a hard price floor. Anything advertised well beneath the going range is the counterfeit or misrepresentation risk a broker is paid to screen out, not a deal. On the highest-value cards, the floor is a feature: it is the line that separates real supply from the rest.

02 /

Ampere is the abundant, discounted workhorse

The A100 and the wider Ampere line are the most available hardware in the secondary channel and carry the deepest discounts to new. For teams buying cost-efficient training and inference capacity right now, this is where the value is hiding.

03 /

Some cards resist the depreciation curve

Inference-class parts are not following the usual used-discount pattern. Production constraints and steady demand keep the RTX 4090 trading above its launch price and the L4 holding near new, while older datacenter cards fall away. Generation alone does not predict the number.

04 /

Condition and documentation move the price

Across every model, the spread between the low and high of a range is mostly condition and provenance, not the chip. Documented, tested units sit at the top of the range; unknown-history cluster pulls sit at the bottom. This is exactly the gap a brokered acceptance window is built to close.

The full report

Get the complete read

The full per-model price table, the by-generation breakdown, and the methodology. One email, and it opens right here.

Get the report

Read the full report

The complete table and breakdown for this issue, plus future monthly issues if you want them.

* required

Get started

Need a number you can trust?

Tell us what you need or what you have. We handle the rest, from first conversation to final payment.