The GPU Market in 2026: State of Play
The GPU market in 2026 is shaped by two opposing forces: insatiable AI demand driving prices up, and expanding supply from NVIDIA and AMD pushing prices down. The result is a market that is stabilizing after the extreme volatility of 2023–2024.
Several dynamics define the current market:
Supply is catching up — slowly. NVIDIA increased H100 production throughout 2025, and Blackwell (B200) GPUs are now shipping in growing quantities. But HBM memory remains constrained, limiting total production. Wait times for bulk H100 orders have shortened from 6+ months to 2–3 months.
The secondary market is maturing. Used data center GPUs (A100, V100) are actively traded. As organizations upgrade to Hopper and Blackwell, their Ampere-generation hardware enters the secondary market, creating more supply for budget-conscious buyers.
Consumer GPUs follow the gaming cycle. The RTX 5090 launched in early 2025, pushing RTX 4090 prices down as gamers upgrade. This creates buying opportunities for AI developers who use consumer GPUs for inference and fine-tuning.
Current GPU Market Prices (New and Used)
| GPU | New Price | Used Price | Used Price Trend | Demand Level |
|---|---|---|---|---|
| H100 80GB SXM | $25,000–$35,000 | $18,000–$22,000 | Stable | Very High |
| A100 80GB | Discontinued (new) | $12,000–$18,000 | Declining slowly | High |
| A100 40GB | Discontinued (new) | $7,000–$10,000 | Declining | Medium |
| RTX 4090 | $1,600–$2,000 | $900–$1,400 | Declining (RTX 5090 effect) | Very High |
| RTX 4080 | $1,000–$1,300 | $600–$900 | Declining | Medium |
| RTX 3090 | Discontinued (new) | $500–$800 | Stable-low | Medium |
| RTX 3080 | Discontinued (new) | $300–$500 | Low, stable | Low-Medium |
Market context: H100 prices remain elevated due to continued AI training demand. A100 prices are in gradual decline as organizations upgrade to Hopper/Blackwell. Consumer GPUs (RTX 40-series) are dropping as the RTX 50-series pulls demand to the new generation.
Buying Tips: How to Get the Best Deal
For Data Center GPUs (A100, H100)
Verify thoroughly. Data center GPUs have often been running under sustained compute loads for months or years. Request:
- Serial number (verify with NVIDIA for warranty status)
- Runtime hours or deployment history
- Thermal throttling history (if available)
- Original purchase documentation
- Error/RMA history
Check VRAM health. Run nvidia-smi diagnostics and a sustained memory test. HBM errors can develop under heavy use and may not be immediately apparent. A GPU with VRAM errors is worth significantly less.
Negotiate based on generation. Ampere (A100) GPUs will continue declining as Hopper and Blackwell expand. Factor in 6–12 months of additional depreciation when offering on used A100s. H100s hold value better in the near term but will depreciate sharply when Rubin-architecture GPUs arrive (late 2026–2027).
Buy from reputable sources. Authorized NVIDIA partners, established GPU resellers, and data center liquidation platforms offer buyer protections. Avoid unverified sellers on consumer marketplaces for enterprise hardware.
For Consumer GPUs (RTX 4090, 3090)
Inspect for mining/compute history. GPUs used for cryptocurrency mining or continuous rental workloads have accelerated wear on fans, thermal paste, and power delivery. Ask the seller directly — reputable sellers will disclose usage history.
Compare used vs. new pricing. When used pricing is within 10–15% of new, buying new gives you full warranty and known history. The sweet spot for used consumer GPUs is 50–70% of new pricing — enough savings to justify the warranty trade-off.
Time your purchase. Consumer GPU prices drop fastest in the 3–6 months following a new generation launch. The RTX 5090 launch means RTX 4090 prices will continue declining through mid-2026. If you can wait, prices will be lower.
Test before buying. Run a sustained GPU benchmark (3DMark Time Spy, FurMark) for at least 30 minutes. Check for thermal throttling, artifacts, or crashes. A GPU that appears fine at idle may have issues under load.
Selling Tips: Maximize Your GPU’s Resale Value
Documentation and Proof
Buyers pay more for confidence. Provide:
- Benchmarks: Run 3DMark or GPU-Z and include timestamped screenshots
- Thermal data: Show GPU temperatures under load (under 80°C is reassuring)
- Photos: Clear images of the card, serial number, and any cosmetic details
- Purchase history: Original receipt or invoice if available
- Usage description: Honest characterization of use (gaming, rendering, AI training)
Pricing Strategy
Check comps on 3+ platforms. eBay, Hardware Swap forums, and specialized GPU resellers each have different pricing dynamics. Average the recent sold prices across platforms to find the market rate.
Price 5–10% below market for quick sale. The GPU market moves fast — a card listed at market price may sit for weeks while depreciation continues. Pricing slightly below market ensures a quicker sale and protects against further price drops.
Sell timing matters. GPU resale values follow predictable patterns:
Data center GPUs retain value longer (AI demand sustains pricing) but experience a sharp drop when the next generation becomes widely available. The optimal sell window is 12–18 months after purchase, before the next architecture launches.
Consumer GPUs depreciate faster, especially in the 3–6 months after a successor launches. The RTX 4090 lost roughly 30% of its value within 4 months of the RTX 5090 announcement. If you plan to sell, do it before or immediately after the next generation is announced — not after.
Sell vs Rent: A Data-Driven Comparison
The most important decision for GPU owners is whether to sell hardware outright or rent it out for recurring income. Here is the math:
Scenario: RTX 4090 (current used value ~$1,100)
Option A: Sell now
- Receive: $1,100 (one-time)
- Net after fees (10% marketplace fee): $990
- Done. Cash in hand.
Option B: Rent for 18 months, then sell
- Monthly net rental profit: ~$52/month (from our rental income guide)
- 18 months of rental: $936
- Estimated resale value after 18 months: ~$500–$700
- Net after marketplace fees: $450–$630
- Total: $1,386–$1,566
Option B generates 40–58% more total value — but spread over 18 months with ongoing effort (monitoring, maintenance, electricity costs).
Scenario: A100 80GB (current used value ~$15,000)
Option A: Sell now
- Receive: $15,000 (one-time)
- Net after fees: ~$13,500
Option B: Rent for 18 months, then sell
- Monthly net rental profit: ~$412/month
- 18 months of rental: $7,416
- Estimated resale value after 18 months: ~$6,000–$8,000
- Net after fees: $5,400–$7,200
- Total: $12,816–$14,616
For the A100, the comparison is closer. Renting generates slightly less total value than selling now — because the A100’s depreciation is steeper (Hopper and Blackwell are replacing it). The older the GPU, the more selling now makes sense due to accelerating depreciation.
Decision Framework
| Factor | Favors Selling | Favors Renting |
|---|---|---|
| GPU age | Older generation (A100, V100) | Current generation (H100, RTX 4090) |
| Depreciation timeline | Next gen launching soon | Next gen 12+ months away |
| Your electricity cost | High ($0.25+/kWh) | Low ($0.12 or less/kWh) |
| Maintenance willingness | Prefer zero effort | Willing to manage hardware |
| Cash needs | Need capital now | Prefer ongoing income |
| Time horizon | Short (under 12 months) | Long (18+ months) |
Key insight: For current-generation GPUs with 18+ months before the next architecture, renting almost always generates more total value. For older-generation GPUs with imminent replacement, selling now captures value before depreciation accelerates.
GPU Depreciation: Understanding the Curve
GPU hardware follows a distinctive depreciation pattern driven by architecture cycles rather than simple age:
Months 0–12: Slow depreciation. The GPU is current-generation, demand is strong, and supply is still constrained. Value retention: 70–90%.
Months 12–18: Moderate depreciation. Next-generation GPUs are announced or entering early production. Buyers start waiting. Value retention: 50–70%.
Months 18–24: Sharp depreciation. Next-generation GPUs are widely available and offer 2–4x better performance. The used market floods with current-gen hardware as data centers upgrade. Value retention: 30–50%.
Months 24–36: Stabilization at low level. The GPU finds its niche in the used market — budget AI, inference, education. Value retention: 20–35%.
For detailed analysis of GPU cloud rental economics — including whether renting is better than owning — see our rent vs. buy comparison. For understanding which GPU models deliver the best value at current prices, see our best GPU for AI guide.
Frequently Asked Questions
Is now a good time to buy a GPU?
For consumer GPUs (RTX 4090): prices are declining as the RTX 5090 takes over. Wait 2–3 months for further drops, or buy now if you need immediate compute. For data center GPUs (H100): prices are stabilizing. If you need H100 compute, buying makes sense if utilization will exceed 60% for 18+ months. Otherwise, rent from a marketplace.
Where should I buy used GPUs?
For consumer GPUs: eBay, Reddit Hardware Swap, local marketplaces, and specialized GPU resellers. For data center GPUs: authorized NVIDIA partners, data center liquidation platforms, and enterprise resellers. Always verify serial numbers and request benchmark documentation.
How fast do GPUs lose value?
Consumer GPUs lose 40–50% of value within 12 months of a successor launch. Data center GPUs retain value longer (AI demand) but drop 50–60% within 6 months of next-generation widespread availability. The key variable is the architecture cycle — sell before the next generation becomes easily accessible.
Should I sell my mining GPUs?
If you have GPUs from the mining era (RTX 3070, 3080, 3090), sell sooner rather than later. These GPUs have experienced heavy wear and their market value continues to decline. The longer you hold, the less they are worth. Exception: if electricity is cheap and the GPU still functions reliably, renting it out may extract a few more months of value before selling.
Is it worth buying a GPU specifically for rental income?
It can be, but the ROI timeline is long. An RTX 4090 at $1,600 generating $52/month net takes 31 months to break even — assuming stable rental rates, which is not guaranteed. An A100 at $15,000 generating $412/month takes 36 months. GPU rental is better viewed as monetizing hardware you already own, rather than a pure investment. If you want GPU compute exposure, renting from marketplaces (as a customer) avoids ownership risk entirely. For a full breakdown of this decision, see our rent vs. buy analysis.
Alternatively, if you want to earn from GPU compute without owning or managing hardware, you can buy a GPU package on GPUnex starting at $59 and receive daily revenue from compute demand.