Rack Report
GPU Resale Index
Resale data for GPUs, memory, storage, server CPUs, and networking.
Issues
Each monthly issue contains seven pages. Contents include a market overview, price and volume trends, benchmark assets, asking vs. realized prices, and much more.
Current market view
A snippet from the October 2026 issue. The full report contains much more.
Indexed to H1 2025 = 100.
| Component | Index | MoM | YoY | vs. depreciation | Liquidity | 90-day |
|---|---|---|---|---|---|---|
| Memory | 491 | +19.3% | +224.7% | +492% | B | ↘ |
| Enterprise SSD | 214 | -4.3% | +116.8% | +123% | B | ↗ |
| GPU | 75 | -12.3% | -22.1% | −14% | B | ↘ |
| Server CPU | 48 | -1.0% | -36.1% | −52% | B | ↘ |
| Network Cards / DPU | 100 | +2.4% | -0.5% | −9% | B | ↗ |
Questions
Where does the data come from?
Every number is built from realized resale transactions contributed by brokers, resellers, and ITADs, alongside our own procurement work. Dealer asking prices are tracked separately, only to measure the gap between asks and what actually closes.
What does the index mean?
Each component class has a price index anchored to a baseline of H1 2025 = 100. A reading of 200 means realized prices are about double their first-half 2025 average, and a reading below 100 means they sit under it. Indexing puts memory, drives, GPUs, CPUs, and networking on one comparable scale.
How do the liquidity and confidence grades work?
Every series carries two letter grades from A to D. Liquidity reflects how quickly and consistently the asset trades in the secondary market. Confidence reflects how well supported the number is, given the volume and consistency of the transactions behind it. Grades are refreshed with every issue, and the methodology page explains the thresholds.
How many transactions is the benchmark built on?
A large and growing set of completed transactions (hundreds every month) across all five component classes, expanding monthly as data partners contribute. Each issue publishes per-series sample sizes on the methodology page, so you can judge the support behind every number.
How often is a new issue released, and is it free?
One issue a month. Current issues are free by email while the benchmark is in open release. Future issues will be split into a free edition and a premium edition (Rack Report Pro) with transaction-level detail, segment cuts, and spreadsheet delivery. Subscribers hear first.
I trade hardware. How do I contribute data?
ITADs, brokers, and resellers are invited as data partners. Contributed transactions are anonymized and published only in aggregate. In return, partners get early access to every issue and a private read on their realized prices vs. the market. Use the contact form below to get started.
Can you value my specific portfolio?
Yes. Paid portfolio evaluations mark a fleet, collateral book, or lease portfolio to realized comps, asset by asset, with liquidity and confidence grades. Reach out through the contact form below.
Still have questions?
Services
Leading financial institutions source data center compute deals and manage their risk with American Compute.
Discuss a transaction →For lenders, lessors, and investors
- If you need deals: Deal flow. Qualified GPU projects seeking financing, for lenders, lessors, and investors.
- If the equipment is overpriced: Appraisals & due diligence. We check the price against resale data, and the site's power and cooling.
- If the equipment is damaged: Standard insurance. Property damage, transit and breakdown, with you as loss payee.
- If the operator is unavailable: Backup operator. We step in and keep the project running, so the offtaker keeps paying.
- If the offtaker stops paying: Lost revenue cover. A bond pays part of the missed revenue for 30 days, giving you time to find a new offtaker.
- If you need a certain resale value: Recovery floor (20%+), RVI. Recovery, remarketing coordination, and residual value insurance for a minimum resale value.
For operators and owners
- GPU Lenders: Insanely fast GPU financing for clusters under $20M.
- Managed GPUs: Your own GPU cluster, deployed and run for you.