Complete guide to NVIDIA GPU surplus values in 2026. Current market prices for H100, H200, A100, and L40S GPUs. Where to sell surplus AI accelerators for top dollar.
The secondary market for NVIDIA data center GPUs has become one of the most dynamic and valuable segments in the IT asset disposition industry. As AI training facilities undergo rapid hardware refresh cycles — moving from A100 to H100 to H200 and beyond — enormous quantities of high-value GPU accelerators are entering the surplus market. In 2026, a single NVIDIA H100 SXM GPU can sell for $15,000 to $25,000 on the secondary market, making GPU decommissioning one of the highest-value asset recovery opportunities available.
The pace of AI hardware evolution means that GPUs which were cutting-edge just 18-24 months ago are now being replaced by newer architectures. However, these displaced GPUs remain extremely capable for inference workloads, fine-tuning, and smaller-scale training tasks. This creates a robust secondary market where organizations with less demanding workloads can acquire proven hardware at a fraction of new equipment cost.
DataCenterDecommissioning.ai has emerged as a leading buyer of surplus NVIDIA GPU infrastructure, purchasing individual GPUs, complete servers, and entire AI training clusters. We understand the unique handling requirements, value drivers, and market dynamics of GPU equipment.
NVIDIA H100 SXM5 80GB GPUs currently sell for $15,000 to $25,000 each on the secondary market, depending on condition, warranty status, and quantity. Complete DGX H100 systems (8x H100 SXM) command $120,000 to $280,000. The H100 remains the workhorse of large-scale AI training and retains exceptional value due to continued strong demand.
NVIDIA H200 GPUs, with their expanded 141GB HBM3e memory, are beginning to enter the secondary market as organizations upgrade to Blackwell architecture. Early surplus H200 units are selling for $20,000 to $35,000 each, reflecting their newer status and enhanced capabilities for large language model inference.
NVIDIA A100 80GB SXM GPUs have stabilized at $5,000 to $10,000 each, down from their peak but still representing significant value. Complete DGX A100 systems sell for $35,000 to $110,000 depending on the 40GB vs 80GB variant. The A100 remains popular for inference and fine-tuning workloads.
NVIDIA L40S GPUs sell for $4,000 to $8,000 each. These inference-optimized accelerators are in strong demand from organizations deploying AI services that don't require the full training capabilities of H100-class hardware. Their lower power consumption and PCIe form factor make them versatile for many deployment scenarios.
NVIDIA A40 and A30 GPUs sell for $2,000-$4,500 and $1,500-$3,500 respectively. While these older-generation cards carry lower individual values, large quantities from data center refreshes can represent substantial total recovery.
Memory variant is the primary value driver. The 80GB versions of both A100 and H100 command significant premiums over their lower-memory counterparts. For the A100, the 80GB SXM variant is worth roughly 50-70% more than the 40GB version.
Form factor matters significantly. SXM variants (designed for NVLink-connected systems like DGX) are generally worth more than PCIe variants because they offer higher memory bandwidth and support NVLink interconnects. However, PCIe variants have broader compatibility, which supports their value for diverse deployment scenarios.
Quantity and packaging affect pricing. Individual GPUs sell at retail secondary market prices, but large lots (100+ units) may sell at slight discounts due to the capital required for bulk purchases. Conversely, complete systems (DGX, HGX baseboard) often command premiums over the sum of their individual GPU values because they include tested, validated configurations.
Warranty and provenance impact buyer confidence. GPUs with transferable NVIDIA warranty remaining, or those from known enterprise environments with documented handling, sell for premiums over units of uncertain origin. We provide full documentation of provenance for all equipment we purchase.
The GPU market moves fast. Values can shift 10-20% in a single month based on new product announcements, supply changes, and demand fluctuations. If you have surplus GPUs from a data center refresh or decommissioning project, acting quickly maximizes your return.
DataCenterDecommissioning.ai purchases NVIDIA GPUs in all quantities — from individual cards to complete AI training clusters with thousands of GPUs. We provide firm cash offers within 24 hours and can arrange secure pickup with full chain-of-custody documentation.
For large GPU lots, we offer on-site assessment where our team catalogs every card by serial number, verifies functionality, and provides a comprehensive offer. We handle all packaging, shipping, and insurance for these high-value assets.
Contact us at 651-946-0136 or email surplus713@gmail.com with your GPU inventory details. Include the model (H100/H200/A100/L40S), memory size, form factor (SXM vs PCIe), quantity, and whether they are individual cards or installed in servers/systems.
Get a firm cash offer within 24 hours. Free assessment, free rigging, free freight. We buy nationwide.
Get My Free Quote Call 651-946-0136