GPU Server Buyback

Models and families covered

NVIDIA DGX H100, DGX A100, DGX-2, DGX Station. NVIDIA HGX-baseboard systems (H100/A100 8-GPU). Dell PowerEdge XE9680, XE8640, R760xa, R750xa. HPE Apollo 6500 Gen10/Gen11 with H100/A100. HPE Cray XD with NVIDIA. Supermicro AS-4124GS-TNR, AS-2124GQ-NART. Lenovo ThinkSystem SR670 V2, SR675 V3.

GPU server pipeline characteristics

GPU servers are higher-residual, higher-volatility, and more compliance-heavy than general-purpose servers. Buyback engagements typically come from AI startup refresh cycles (H100s replaced by H200/B100); BFSI ML platform refreshes; sovereign-fund-backed lab refreshes; and hyperscale tenant exits. The secondary buyer base is concentrated (large AI labs, regional cloud operators, well-funded research institutions, cross-border remarketers) and price-discovery is opaque — Maxicom's quote pricing reflects live remarketing inputs rather than published guidance.

Per-asset wipe protocol for GPU servers

Server-level sanitisation: NIST SP 800-88 Purge for HDDs, IEEE 2883-2022 Sanitize for SSDs/NVMe, BIOS/iDRAC/iLO/IPMI configuration cleared. Accelerator-level sanitisation: NIST SP 800-88 Cryptographic Erase via NVIDIA driver stack on H100/A100 with confidential computing keys destroyed; HBM overwritten via on-board test pattern; per-accelerator certificate. NVLink and InfiniBand fabric configurations cleared. Per-server Certificate of Destruction issued naming the standard, the accelerator state, and the chassis/baseboard sanitisation steps.

Cross-border export and licensing

GPU servers shipping internationally require per-rule export classification (US BIS for NVIDIA-equipped systems; equivalent for AMD MI-series). Maxicom handles classification for the routing path and confirms that the destination is not on a restricted-party list before resale.

Settlement and engagement mechanics

Settlement is in CAD (CAD) against your purchase order, line-item per asset, payment terms agreed in the SOW. Programme engagements run on milestone-based settlement against the rolling pickup schedule with monthly true-up. Multi-site engagements (where assets route between your Canadian sites) are consolidated to your single CAD ledger; the customer-facing transaction is single-currency. The SOW is structured with Maxicom Inc. (Canada); GST/HST treatment is handled per Canadian tax law. Quote validity follows the asset class — 14 days for steady-state enterprise hardware, 5 business days for AI accelerators where the secondary market re-prices weekly, 30 days for memory and components. We re-quote without penalty where the validity has lapsed and the customer is ready to transact.

Audit defensibility and certificate format

Every asset routed through this engagement receives a per-asset Certificate of Destruction with eleven required fields: serial number, make/model/capacity, data classification at retirement, sanitisation method (Clear/Purge/Destroy under NIST SP 800-88 Rev. 2, with the specific technique cited), particle size or field strength or encryption algorithm where applicable, sanitisation tool + version + verification response, UTC timestamp + facility location, operator name + ID + signature, witness signature where present, chain-of-custody reference back to the pickup manifest, and the destruction reason where Reuse-First triage was overridden. Certificates are admissible against OSFI B-13, PIPEDA, NIST SP 800-88 Rev. 2, IEEE 2883-2022, and (where contractually specified) DoD 5220.22-M and NAID-aligned process — one certificate covers all simultaneously. Certificate retention is 7 years default, 8+ years for BFSI engagements, longer where the master service agreement specifies.

Cross-border resale routing under NDA

Where local market depth in Canada cannot absorb the retiring volume at fair refurb pricing, working assets route cross-border through Maxicom's remarketing channel — US sub-tier markets, US sub-tier markets + North America, CA → US sub-tier markets and North America, SG → North America. The routing decision is made per asset-class at engagement scoping; the customer sees the routing on the SOW and can opt out where channel-respect or data-residency rules require. NDA discipline is standard. Surplus does not return to your own market's primary channel without explicit consent. Export classification (US BIS for AI accelerators; equivalent local regimes for other restricted-class hardware) is handled before the trade closes; restricted-party screening is part of every cross-border transaction.

Reuse-First disposition KPIs reported back to you

Programme-level engagements receive quarterly business reviews covering: total tonnage processed, Reuse-First reuse rate (% refurbished and redeployed vs % destroyed by media class), residual value recovered in CAD, embodied-carbon-recovered estimate (CO₂e avoided by keeping working assets in service rather than replacing them with newly-manufactured hardware), diversion-from-landfill percentage, material-recovery breakdown, and exception reporting. The reporting format is mapped to your sustainability reporting framework — ISSB IFRS S1/S2, GRI 301/305/306, SASB IT services standards. Single-event engagements receive the same data as a per-engagement summary attached to the consolidated certificate. The reuse-rate metric is the most informative KPI: our blended cohort typically runs around two-thirds reuse rate (indicative); programme engagements typically improve year-over-year as the engagement learns the asset mix.

Key models in our pipeline

NVIDIA DGX H100 · DGX A100 · Dell PowerEdge XE9680 · XE8640 · HPE Apollo 6500 · Supermicro AS-4124GS · Lenovo SR670 V2