Where GPU Prices Stand in Q2 2026
The GPU market in Q2 2026 looks markedly different from just twelve months ago. Enterprise buyers who delayed procurement decisions are now finding a more nuanced landscape — one where the window for smart buying may be opening, but caution is still warranted. Supply chain dynamics, next-generation hardware transitions, and persistent memory shortages are colliding to create both opportunities and risks for IT procurement managers sourcing at scale.
On the secondary and refurbished market — where companies like TBR Trade Group operate — NVIDIA H100 GPUs are now priced between $21,000 and $34,000 refurbished, compared to the $25,000–$40,000 range for new units. The A100 has dropped further: the 40GB variant now trades at $8,000–$12,000 (from a $15,000+ new price) and the 80GB at $12,000–$18,000. These are meaningful numbers for enterprise IT teams that need compute power without committing to the full new-unit price tag.
Supply Chain: Stabilizing — But Not Fully
The widespread disruptions that characterized the GPU supply chain in 2023 and 2024 have largely resolved. Manufacturers have invested heavily in new fab capacity, and logistics pipelines have normalized. However, a critical component constraint is now tightening: High Bandwidth Memory (HBM).
HBM shortages are running at 30–70% of demand across the industry, and DRAM costs have climbed approximately 15% compared to late 2025. This is not a minor footnote. HBM is essential to AI-focused GPUs like the H100, H200, and B200 — and tight HBM supply is acting as a ceiling on how quickly new inventory can enter the market.
- HBM supply shortfalls are projected to persist through Q3 2026, keeping new-unit prices elevated for premium AI GPUs
- DRAM price increases are flowing downstream into workstation and server-grade component pricing
- Regional pricing disparities remain significant — energy costs, local demand, and infrastructure availability all create price variation across geographies
- Long-lead procurement remains the safest strategy for Q3 and Q4 delivery slots
The B200 Transition: What It Means for Buyers
NVIDIA's B200 GPU is now available from select providers, with cloud spot pricing starting around $2.12/hr and new hardware listing at $30,000–$50,000 per unit. The DGX B300 system — NVIDIA's flagship cluster unit — is listing at $300,000–$350,000. These figures put B200-class hardware well out of reach for most enterprise buyers shopping on typical IT budgets.
The more relevant dynamic for enterprise procurement teams is what the B200 transition is doing to the H100 secondary market. Many organizations that locked in A100 and H100 GPUs under multi-year cloud contracts in 2023 and 2024 are seeing those contracts expire this year. As early adopters migrate to B200 and GB300 systems, significant H100 and A100 inventory is expected to return to the secondary market through 2026. This adds further downward price pressure — current forecasts suggest 10–20% additional downward movement on H100 secondary pricing once B200 availability broadens further.
Cloud vs. On-Premises: The Q2 Calculation
Enterprise GPU procurement is no longer a simple buy-vs-rent decision. Cloud GPU rental pricing has come down substantially from its 2024 peaks. H100 instances are now available from $1.03/hr (spot) to $2.50/hr (on-demand SXM5 configuration) across major providers. A100 80GB SXM4 instances have dropped to $1.07/hr on-demand.
However, cloud spot pricing has become unreliable as a planning input. Teams that relied on spot instances for development workloads in 2024 are finding that spot availability simply evaporates during peak demand windows — even at elevated pricing. For production workloads or dedicated AI training pipelines, spot is not a safe foundation.
For workloads with predictable compute requirements, on-premises or co-located owned hardware often delivers better total cost of ownership over a 3–4 year horizon. Refurbished H100 and A100 units — when sourced from a verified supplier with warranty coverage — remain compelling at current secondary market pricing.
What Enterprise Buyers Should Do in Q2 2026
1. Don't Wait for the Bottom
HBM constraints and the B200 transition are creating unusual pricing dynamics. Some categories will continue softening; others may tighten as new AI model releases drive sudden demand spikes. Trying to time the exact bottom is a losing strategy. If your workloads need compute now, act now.
2. Evaluate Refurbished Hardware Seriously
The refurbished GPU market has matured substantially. Reputable suppliers now conduct multi-stage testing, issue warranty certificates, and can provide performance documentation. At current price gaps — often 20–35% below new — certified refurbished H100s and A100s offer genuine value for enterprise deployments.
3. Cap Commitment Windows
For cloud reserved capacity, 12-month maximum commitment windows are advisable given the pace of hardware transitions. For on-premises purchases, ensure your supplier agreements include trade-in or upgrade provisions as the next GPU generation arrives.
4. Source From Verified Suppliers
The secondary GPU market includes both reputable and unreliable actors. Prioritize suppliers who offer documented testing records, warranty support, and transparent provenance. At TBR Trade Group, every unit is tested and validated before it ships — a standard enterprise buyers should demand from any GPU supplier.
Looking Ahead: Q3 2026
The second half of 2026 will likely bring further secondary market supply as B200 deployments accelerate and legacy contract inventory flows back into the ecosystem. For buyers with flexibility, waiting until Q3 could yield additional price advantages — particularly on H100 and A100 inventory. For buyers with near-term workload requirements, the Q2 2026 secondary market already offers solid value relative to peak 2024 pricing.
TBR Trade Group tracks GPU pricing and availability continuously. Contact our procurement team for current pricing on specific GPU models, bulk order availability, and enterprise warranty options.