NVIDIA datacenter GPU buyback — V100 to H200.
NVIDIA datacenter GPUs command the strongest residuals in IT secondary markets — sustained demand from AI training and inference customers globally. We buy back A100 (40GB / 80GB SXM4 / PCIe), H100 (80GB SXM5 / PCIe), H200 (141GB), L40S, L4, A40, A30, plus older V100 (16GB / 32GB SXM2). Indicative INR residuals: A100 80GB SXM4 Rs250,000–500,000; H100 80GB SXM5 Rs500,000–800,000+; L40S Rs120,000–180,000.
Common NVIDIA Datacenter GPUs (A100 / H100 / L40S / L4) configurations
- ♦ H200 (141GB) — flagship, Rs600,000–1,200,000.
- ♦ H100 80GB SXM5 — Rs500,000–800,000.
- ♦ H100 80GB PCIe — Rs350,000–600,000.
- ♦ A100 80GB SXM4 — Rs250,000–500,000.
- ♦ A100 80GB PCIe — Rs180,000–350,000.
- ♦ A100 40GB SXM4 — Rs150,000–280,000.
- ♦ L40S — Rs120,000–180,000.
- ♦ L4 (24GB) — Rs60,000–95,000.
- ♦ A40 (48GB) — Rs100,000–160,000.
- ♦ A30 (24GB) — Rs70,000–110,000.
- ♦ V100 32GB SXM2 — Rs50,000–90,000.
- ♦ V100 16GB SXM2 — Rs35,000–65,000.
- ♦ T4 (16GB) — Rs25,000–45,000.
NVIDIA-specific drivers of per-unit INR value.
NVIDIA datacenter GPUs are the strongest-residual asset class we handle. The AI training / inference market sustains global demand; secondary-market GPU pricing in India closely tracks the global market with a 5–10% India-specific arbitrage based on GST and import-duty considerations.
Configuration premiums: SXM form factor (A100 SXM4, H100 SXM5) commands 30–50% over PCIe equivalent because of NVLink bandwidth. HBM tier matters — H100 80GB carries 50–80% over H100 SXM 60GB-config equivalent. SerialPro firmware state (GPU-Direct RDMA, MIG configuration) documented on quote.
GPU memory has data potential — NVMe-style training-data caches on the HBM. Pre-decom: nvidia-smi reset and driver-stack uninstall via your normal change management. Per-GPU serial logged on the Certificate of Destruction. Where the customer is BFSI / DPDPA-sensitive, full GPU-to-host RDMA flush is documented.
Per-job Certificate of Destruction. Serial-level chain of custody.
Every drive that leaves your premises is sanitised to NIST SP 800-88 Rev. 1 (Clear / Purge / Destroy) and IEEE 2883-2022 storage-sanitisation methods, with DoD 5220.22-M overwrite available on request for legacy audit-compatibility. Each engagement ships with a per-job Certificate of Destruction listing serial numbers, sanitisation method per device, operator initials, and a SHA-256 hash of the supporting log file — the evidence pack that goes into your DPDPA 2023 § 8 (Reasonable Security Safeguards) and RBI IT-Risk Management Master Direction file.
Serial-level chain of custody is recorded from the moment a device leaves your premises to the moment it is sanitised. We carry professional indemnity, public-liability, and transit insurance, all INR-denominated. Right-to-audit clauses in our Statements of Work are standard for our BFSI customers — RBI-regulated banks, IRDAI-regulated insurers, and SEBI-regulated capital-markets participants.
How a Maxicom India buyback runs from asset list to settlement.
Step 1 — Asset list. You send a CSV / spreadsheet with serial numbers, makes, models, and configurations (CPU SKU tier, RAM total, drive count and type). Within 2 working hours we issue a written, INR-denominated, per-line indicative quote. Larger jobs (full-rack data-centre decom) get a site survey scheduled within 24 hours.
Step 2 — Statement of Work. We sign an SoW that specifies the sanitisation method (NIST 800-88 Clear or Purge, or physical destruction where required), the chain-of-custody handoff points, the right-to-audit clause, and the settlement timeline (typically 5 business days from disposition).
Step 3 — Pickup. Locked, GPS-tracked transit. Two-operator + witness sign-off discipline at packout. On-site witnessed destruction available where data-bearing media cannot leave the premises (common for some hospitals, certain bank trading floors, and government-cleared facilities).
Step 4 — Destruction + settlement. Drives sanitised to NIST 800-88; non-data hardware refurbished and remarketed through domestic India + SAARC trader and reseller channels. Per-job Certificate of Destruction issued on completion. INR settlement to your nominated account.
Maxicom India — frequently asked
Indicative INR for an NVIDIA H100 80GB SXM5 in 2026?
Rs500,000–800,000 working/tested for an H100 80GB SXM5 (HGX-H100 or equivalent reference platform integration). The range reflects GPU board state (with-without baseboard heat sink), SerialPro firmware version, NVLink connector condition, and current global secondary-market AI-training demand.
Do you handle 8-GPU HGX-H100 / HGX-A100 reference platforms?
Yes — entire HGX modules quoted as a unit. An HGX-H100 (8× H100 80GB SXM5) typically Rs5,000,000–7,500,000+ depending on host server platform (Dell R760xa, HPE DL380a Gen11, Supermicro AS-4124GS) and condition. Larger compute-cluster pulls (32-GPU, 64-GPU) coordinated as multi-week engagements.
How is GPU memory data destroyed?
GPU HBM (High-Bandwidth Memory) is volatile DRAM — power-off destroys the data. We document a verified power-off and idle-period at packout (typical 10-minute power-off-and-cool before pickup). Per-GPU serial logged on the Certificate. For BFSI / DPDPA-sensitive customers requiring belt-and-braces sanitisation, we coordinate an additional NIST 800-88 Clear cycle on the GPU controller before packout.
Can you buy back GPUs that have been used for crypto mining?
Yes — see /crypto-mining-server-buyback-in-india/. Crypto-used GPUs typically discount 15–25% vs equivalent never-mined units because of thermal cycling concerns, but H100 / A100 / L40S still commands strong residuals. We document mining-state on the quote where disclosed.
Are India + global pricing for AI hardware aligned?
Within 5–10%. The AI-training market is global and arbitrage is fast — significant pricing dislocations close within weeks. India-specific factors: GST 18% on import, customs duty on enterprise GPUs, and INR-USD exchange rate. We quote in INR with reference to current-week global secondary-market pricing.
Do you handle NVIDIA NVLink switch / NVSwitch hardware?
Yes — NVSwitch and NVLink-bridge boards quoted separately as part of a multi-component engagement. Demand strong for HGX-H100 / HGX-A100 customers building inference clusters. Quote on a per-engagement basis given the niche.
How quickly can I get an INR quote for these assets?
Within 2 working hours of receiving your asset list (CSV with serial numbers, makes, models, configurations). Larger jobs requiring a site survey: survey within 24 hours, quote within 1 working day after the survey.
Do you cover multi-location pickups across India metros?
Yes. Mumbai, Bengaluru, Hyderabad, Chennai, Pune, Gurugram, Noida — same-week pickup is standard. Tier-2 cities (Jaipur, Kolkata, Chandigarh-Mohali, Ahmedabad, Indore, Coimbatore, Kochi) covered with 5–10 business-day lead time. One Statement of Work, one project manager, one consolidated INR settlement.
Are your buyback engagements DPDPA 2023-aligned?
Yes. Per-job Certificate of Destruction citing NIST SP 800-88 Rev. 1 + IEEE 2883-2022 standards. Serial-level chain of custody. Documentation slots into your DPDPA 2023 § 8 (Reasonable Security Safeguards) evidence file. The Data Protection Board of India does not certify ITAD vendors, so what we offer is alignment, not certification.
Do you handle BFSI engagements (RBI / SEBI / IRDAI)?
Yes. RBI IT-Risk Management Master Direction-aware destruction documentation for banks and NBFCs. SEBI Cyber Security and Cyber Resilience Framework-aware for capital-markets participants. IRDAI Information and Cyber Security Guidelines-aware for insurers. Right-to-audit clauses standard in our SoWs.
Can on-site destruction be arranged at our facility?
Yes. Two formats: mobile shred unit (truck-mounted industrial shredder, particle size <2mm) or witnessed-wipe operator (NIST 800-88 Clear / Purge with member of your team witnessing each device). Documentation pack identical to in-facility destruction. Cost typically 15–25% higher; lead time 5–10 business days.