eBay used Intel Gaudi 2 HL-225(B) 96GB OAM module
various server liquidators / eBay sellers · United States (eBay.com / picclick)
Fields
id
supply/ebay-used-intel-gaudi2-hl225-oamKind
used-market
Vendor
various server liquidators / eBay sellers
Region
United States (eBay.com / picclick)
Channels
marketplace
Price (USD)
2,100
Price basis
USED, BARE OAM MODULE - no chassis, no fabric. $2,100 = midpoint of observed asks captured 2026-10-03..14: HL-225 $2,500, HL-225H $2,700, PC-07784-01-GP3 $1,499-1,745. THE RUMOURED $700-900 MARKET IS NOT SHOWING on eBay Oct-2026 - treat that figure as a stale 2025-era liquidation echo. 8x Gaudi2 Supermicro tray $11,995. ASKING price, not sold. PERISHABLE-FIGURE STAMP: OBSERVED 2026-10-14. RE-VERIFY BY: 2027-01-14.
Availability
in-stock
Lead time source
structural-na
Lead time basis
lead_time_weeks is null because this is a secondary market priced per listing.
Export controlled
no
Export control regime
EAR 3A001 or EAR99 + EAR 744.23 / 744.6 party screening
Export control note
Verify classification at purchase; general EAR plus party screening applies.
notes
TRAP for solo use - the module is cheap because the chassis isn't. THE REAL TRAP IS STRUCTURAL: HL-225 is OAM-only, has NO PCIe SKU, and needs the Gaudi UBB chassis (Supermicro 820GH-TNR2 ~$11,995 for the 8-module baseboard) plus a 24x 100GbE fabric to even light up. You cannot put one in a workstation. 96 GB HBM2e / 2460 GB/s, 600 W. SynapseAI/vLLM-fork works but the ecosystem is thin and Intel is winding the line down. MLPerf v4.0 8x aggregate 8,034.85 output tok/s is batch-1024 serving, not single-stream (record comm-gaudi2-8x-llama2-70b-fp8-mlperf).
Notes
TRAP for solo use - the module is cheap because the chassis isn't. THE REAL TRAP IS STRUCTURAL: HL-225 is OAM-only, has NO PCIe SKU, and needs the Gaudi UBB chassis (Supermicro 820GH-TNR2 ~$11,995 for the 8-module baseboard) plus a 24x 100GbE fabric to even light up. You cannot put one in a workstation. 96 GB HBM2e / 2460 GB/s, 600 W. SynapseAI/vLLM-fork works but the ecosystem is thin and Intel is winding the line down. MLPerf v4.0 8x aggregate 8,034.85 output tok/s is batch-1024 serving, not single-stream (record comm-gaudi2-8x-llama2-70b-fp8-mlperf).