Hardware: GPU servers
HGX H200 SXM: 8-GPU Server System
141 GB HBM3e per GPU. The memory upgrade over H100 that wins on long-context inference and large-model serving.
Live lots: available now
Currently available HGX H200 systems
Limited secondary supply · Neocloud surplus only. Prices shown against the Rack Rate Index median, deal score is automatic.
8×H200 SXM HGX — neocloud surplus
8× H200 SXM 141GB · 11mo in service · OEM warranty transfer negotiable
Market pricing: Rack Rate Index
What HGX H200 systems trade at
| Series | Min | Median | Max | N | MoM |
|---|---|---|---|---|---|
| Used, tested | $345,000 | $362,000 | $375,000 | 6 | +3.1% |
| New build | Quote only; contact for availability | — | — | ||
Full server system pricing: chassis, NDR networking, host CPUs, and memory included. Indicative market ranges, updated Jul 18, 2026. Verified series publish once they clear our N-quote threshold: median of verified quotes, supplier-blind.
System specifications
HGX H200: full system specs
| GPUs | 8× NVIDIA H200 SXM |
| GPU memory | 141 GB HBM3e per GPU · 1.1 TB total |
| Memory bandwidth | 4.8 TB/s per GPU · 38.4 TB/s total |
| GPU interconnect | NVLink 4 · 900 GB/s bisection · NVSwitch |
| FP8 (system) | ~16 PFLOPS dense |
| BF16 (system) | ~8 PFLOPS dense |
| GPU TDP | 700 W per GPU |
| System power draw | ~10.5 kW · plan 13 kW with PUE |
| Host platform | 2× Xeon SP or EPYC · 2 TB DDR5 · PCIe Gen 5 |
| Common chassis | Supermicro SYS-821GE-TNHR2 · Dell XE9680 |
| Network fabric | 8× ConnectX-7 400G NDR InfiniBand |
| Form factor | 10U air-cooled (same chassis as H100) |
Buyer decision guide
Should you buy a HGX H200?
Buy HGX H200 if
You're running large-context inference
70B+ models at long context windows fit on fewer nodes with H200's 141 GB. The $/token math favors H200 over H100 clusters for memory-bound workloads.
Buy HGX H200 if
Same DC footprint, more memory
Drop-in for H100 racks: same chassis, same power envelope. No DC re-design needed.
Consider alternatives if
Consider H100 if budget is the constraint
H100 is ~44% cheaper and has far deeper secondary market liquidity. Right for most inference and fine-tuning workloads.
HGX H100 pricing →Used vs refurb: what the premium buys
| Factor | Used, tested | Refurb, warranted |
|---|---|---|
| Warranty | None / seller DOA window | 90-day seller warranty + DOA |
| Burn-in report | Available (ask for it) | Included and certified |
| Price premium | — | +8–12% over used median |
| Right for | Buyers who can evaluate burn-in logs | Teams who want turnkey assurance |
Pre-purchase verification
What to verify before buying
At this price point, skipping due diligence is expensive. Every item below should have a documented answer before you wire funds.
- Same checks as H100: GPU ECC error history is critical
- HBM3e stacking defects: confirm vendor ran full memory stress test
- Neocloud decommission lots: verify service records and runtime hours
- Bid/ask spreads are wide: compare any quote to current Index before committing
Export control notice
H200 carries the same EAR export control classification as H100. End-user screening required for all cross-border transactions.
Looking to buy HGX H200 systems?
Submit your requirements: quantity, condition, timeline, and budget. We benchmark against the Index and match you with verified supply.
Have HGX H200 inventory to sell?
List your lot: describe what you have, condition, and quantity. We benchmark against the Index and surface it to qualified buyers.
Compare models