
8× H100 SXM5 80GB · 640GB HBM3 · NVLink · NDR
Used / refurb · Ready now · Most active series
AI infrastructure market intelligence
Verified pricing from real deal flow. Live supply from vetted sources. The GPU server market: what it's actually clearing, not what sellers wish it would.
Weekly market read · Sourced from real deal flow.
Hardware catalog: GPU servers & workstation AI
Used and new HGX server systems plus workstation GPU, benchmarked against verified market intelligence. Full 8-GPU systems include chassis, NDR networking, host CPUs, and memory. Not bare GPU modules.

8× H100 SXM5 80GB · 640GB HBM3 · NVLink · NDR
Used / refurb · Ready now · Most active series

8× H200 SXM 141GB · 1.1TB HBM3e · NVLink · NDR
Used · Neocloud surplus · Tightening band

8× B200 SXM6 180GB · 1.4TB HBM3e · NVLink · NDR800
New · Factory order · 14–20 wk lead time

8× B300 SXM 288GB HBM3e · NVLink · NDR800 · liquid cooling
New · Allocation-basis only · 20+ wks

8× A100 SXM4 80GB · 640GB HBM2e · NVLink · HDR (EOL)
Used / refurb · Ready now

96GB GDDR7 · PCIe Gen5 · Blackwell arch · workstation AI / single card
New · In stock · No lead time
Full 8-GPU server system pricing: chassis, NDR networking, host CPUs, and memory included. RTX Pro 6000 priced per single card. Indicative market ranges, updated Jul 18, 2026.
Market intelligence: what we're seeing
We collect quotes from reputable sources across the ITAD and neocloud ecosystem and publish what we're actually seeing, anonymized, supplier-blind, methodology-driven. This is the intelligence; the listings below are a product of it.
Current supply: GPU server systems
Every lot benchmarked against current market medians, price shown relative to the Index, lead time on every card. Supply surfaces through our intelligence relationships, not a general marketplace.
8× H100 SXM5 80GB · 2× Xeon 8468 · 2TB DDR5 · NDR · burn-in report included
8× B200 SXM6 180GB · liquid-cooling ready · NDR800 fabric
8× A100 SXM4 80GB · 2× Xeon 8352Y · 1TB DDR4 · HDR · burn-in report
8× H100 SXM5 80GB · Dell chassis · full diagnostics · seller warranty
8× H200 SXM 141GB · 11mo in service · OEM warranty transfer negotiable
8× A100 SXM4 80GB · off-lease · burn-in reports · 6 systems available
8× B300 SXM 288GB HBM3e · NDR800 · liquid-cooling required · 20+ wk lead
96GB GDDR7 · PCIe Gen5 · Blackwell arch · workstation AI · single card
Tell us the spec and quantity: we source against the whole market, including supply that never gets listed.
Request custom supply →GPU server system lots: full 8-GPU systems, chassis included. Supply surfaces as inventory clears verification. As of Jul 18, 2026.
Rack Rate certification
Before a used or refurb system goes live, it's scored across five categories, 10 points each. The score ships with the listing itself, not just a badge, so you can see exactly what passed.
Applies to used and refurb HGX systems; new-build allocations ship OEM-sealed with manufacturer warranty instead. OEM warranty status and provenance are verified separately and shown alongside each listing's score.
Weekly market intelligence
The Rack Rate: weekly Index snapshot, lead time tracker, and what we're seeing across the HPC and AI-infrastructure ecosystem. Sourced from real deal flow.
Read by ITADs, neoclouds, and AI infra operators.
Why Rack Rate
We collect quotes from reputable sources, publish what the market is clearing, and surface verified inventory for discovery. Like CarGurus for AI infrastructure: the data serves the buyer, not the dealer.
01 · INTELLIGENCE
Anonymized quote tape, price history, and lead times built from real transaction flow, not scraped listings or futures indexes. Know what the market is actually clearing before you transact.
02 · DISCOVERY
Supply surfaces through our intelligence relationships with ITADs, neoclouds, and OEM channels. Listings appear because we know they exist, not because someone paid to be listed.
03 · TRANSACTION
Benchmarked pricing, verification support, and deal structure guidance from first inquiry through delivery. Independent: we don't take positions or earn per-transaction commissions.
FAQ
Rack Rate is the market intelligence platform for GPU server hardware. We collect verified pricing from real deal flow, publish anonymized quote data, and surface listings from vetted sources so buyers and sellers know what the market is actually clearing.
From direct relationships with ITADs, neoclouds, and OEM channels built over years of deal flow. Listings appear because we know they exist, not because someone paid to be listed. The published Index is supplier-blind, methodology-driven, and has no commercial relationship with any single seller.
HGX server systems across the NVIDIA stack: A100 (SXM4 80GB), H100 (SXM5 80GB), H200 (SXM 141GB), B200 (SXM6 180GB), and B300. Plus the RTX Pro 6000 Blackwell for workstation and edge AI. Complete 8-GPU server systems: chassis, networking, CPUs, and memory. Not bare GPU modules.
From real, anonymized wholesale quotes and closed deals. The published ranges reflect observed market medians, not listed ask prices or scraped data. Every series shows the number of sources (n) and is supplier-blind. We don't publish a series until we have N qualifying quotes behind it.
Used makes sense when you need capacity immediately and can accept a shorter warranty. New build is right for B200/B300 when you need the latest generation and can absorb 14–20+ week lead times. Each model page covers the trade-offs, depreciation curves, and scenario analysis.
Use the "List inventory" flow above: drop in what you have, condition, and quantity. We benchmark against the Index, verify the lot, and surface it to qualified buyers. Supplier-blind by default: your identity is never disclosed in market data.