What is my used server worth?

The four things that set the value

When we price a server lot, we weigh:

• Generation & model. Newer generations hold more, but popular older models (e.g. Dell R740, HPE DL380 Gen9/10) stay liquid for years — sometimes longer than the OEM keeps supporting them.
• Configuration. CPUs, memory, drives, GPUs and NICs inside the chassis often carry more value than the bare server itself. A stripped chassis and a loaded one can differ by 5-10x.
• Condition. Working and complete is worth most; for-parts and incomplete units still have value at the component level, sometimes a surprising amount if the CPUs or GPUs are current.
• Demand. Global secondary demand for the specific model on the day — which is why values move. AI training demand pulled H100 prices in one direction; a hyperscaler dumping 10,000 R740s can pull rack-server prices in the other.

Real-world examples — what typical models fetch

Nothing communicates how wide the spread is like real ranges. These are the indicative bulk-lot bands we quote today (firm on inspection, not offers, condition and quantity move them):

• Dell PowerEdge R640 (14th-gen 1U, 2017) — refurbished base units ~$375–$575, loaded configs materially more. A liquid, easy-to-move model.
• Dell PowerEdge R650 (15th-gen 1U, 2021) — configured units ~$5,000–$18,000 depending on CPU SKU, DIMM population and NVMe.
• HPE ProLiant DL360 Gen10 — $300–$700 base, $700–$1,800 mid-spec, $1,800–$3,500+ for loaded dual-CPU builds.
• NVIDIA H100 PCIe/SXM5 — commonly $15,000–$28,000 per card; the single biggest value on any decomm most companies will ever do.
• Cisco Nexus 93180YC-EX — $1,500–$5,000 depending on licensing and optics.

Full per-model bands live on each model page and in the server resale value index, GPU resale value index, and storage & networking index. They're updated as the market moves.

Why there's no fixed price

Anyonj — we take title and resel, so the offer reflects what the lot will actually move for, not a marketing number pulled out of a template.

How a trading desk actually prices your lot

Under the hood, three inputs meet:

1. Live comparable transactions. Recent sales of the same or a peer model in the same condition band, weighted toward the most recent. A model that sold 40 units last month at a tight range prices very differently from one that hasn't traded in 90 days.
2. Configured value roll-up. For servers, we back out component value — CPU, DIMM population, drive stack, GPUs, HBA/NIC/RAID — against current spot prices for each, then add a chassis premium (or discount) for how well the config moves as a unit.
3. Lot friction. Consistent lots (100 of the same SKU) price better than mixed lots (12 different models) because they resell as a single line. If your lot is very mixed, expect a small discount for the sorting work.

The output is a range — wide when we're pricing off a limited list, narrow when the list is detailed. Detail is your friend.

Timing — the depreciation curve is not flat

Server value drops most steeply in the first two to three years after launch — often 30-40% per year — then flattens as the model becomes a workhorse of the secondary market. A rough shape for a mainstream 2U rack:

• Launch year: 100% of a benchmark configured price.
• +1 year: ~60-65%.
• +2 years: ~40-45%.
• +3 years: ~25-30%.
• +4-5 years: ~15-25%, then a long tail as the model becomes established on the secondary market.

End-of-support (EOS) is often the right moment to sell, not the wrong one. Global parts and secondary deployment demand for popular EOS models stays strong for years. See EOL vs EOS explained.

The value floor — even for-parts has a bid

Servers rarely go to zero. A truly dead unit still carries component value: CPUs and DIMMs are the biggest recoverable pieces, then RAID cards, HBAs, PSUs, backplanes and rails. A for-parts Dell R740 with dual Xeon Gold and 384GB RAM inside can carry hundreds of dollars in components even if the board is bad. Include for-parts units in the lot rather than sending them to shred — the delta between component recovery and scrap value is often meaningful.

Five valuation mistakes that leave money on the table

1. Assuming warranty = value. Active OEM warranty adds marginal resale value; most secondary buyers self-support or use third-party maintenance. Don't hold a lot back waiting for warranty to expire.
2. Ignoring the GPU premium. A single H100 or A100 in a server can be worth more than the entire chassis. If you have GPUs, list them separately with quantity and form factor (SXM vs PCIe).
3. Pulling drives before pickup. Unnecessary. A reputable buyer wipes to NIST SP 800-88 after pickup. Pulling drives destroys caddy pairing and reduces the offer.
4. Missing rails, caddies, or bezels. Small parts that materially add value. Include everything with the unit.
5. Selling one server at a time. Bulk lots move faster and price better. Group the entire decomm into a single offer.

How to get a fast, firm number

Send a spreadsheet with model numbers, quantities and condition. For servers, include CPU SKU, RAM total, drive count/size, and any GPUs. The clearer the list, the faster the firm bulk offer — most lots price in under 24 hours once the list arrives.

Related guides

More on selling surplus, used and end-of-life IT hardware in bulk: How to sell surplus IT hardware in bulk · Server EOL and EOS — and why it still sells · Data wiping explained: NIST SP 800-88 · Used H100 vs A100 vs H200 — resale value (2026) · What determines your used server's value · What is your data-center hardware worth at decommission? · Buyback desk, marketplace, or recycler — where should you sell?. When you are ready, run the instant value estimator for an indicative range, or send your asset list to get a firm bulk offer — one buyer, the whole lot, drives wiped to NIST SP 800-88.

ServerBuyback is a USA & Canada wholesale buyback desk: we buy surplus, used and end-of-life IT hardware and electronics in bulk, take title, and resell through a global B2B channel. Questions about your specific lot? Talk to the desk →