~$25Klatest observation · analyst estimate · 2026-08
4dated price observations on record
10sourced specs (all linked below)
Specifications
Sourced specs
| Spec | Value | Source |
|---|---|---|
| FP4 Tensor Core | 18 PFLOPS with sparsity / 9 PFLOPS dense per GPU | via nvidia.com ↗2026-08-27 |
| FP8/FP6 Tensor Core | 9 PFLOPS with sparsity / 4.5 PFLOPS dense per GPU | via nvidia.com ↗2026-08-27 |
| INT8 Tensor Core | 9 POPS with sparsity per GPU (72 POPS per 8-GPU HGX) | via nvidia.com ↗2026-08-27 |
| FP16/BF16 Tensor Core | 4.5 PFLOPS with sparsity / 2.25 PFLOPS dense per GPU | via nvidia.com ↗2026-08-27 |
| TF32 Tensor Core | 2.25 PFLOPS with sparsity / ~1.125 PFLOPS dense per GPU | via nvidia.com ↗2026-08-27 |
| FP32 / FP64 | ~75 TFLOPS FP32 / ~37 TFLOPS FP64 per GPU (600 / 296 TFLOPS per 8-GPU HGX) | via nvidia.com ↗2026-08-27 |
| Memory | 192GB HBM3e per GPU, 8 TB/s (1,440GB total at 64 TB/s per 8-GPU DGX/HGX system) | via resources.nvidia.com ↗2026-08-27 |
| Interconnect | NVLink 5 at 1.8 TB/s per GPU (14.4 TB/s per HGX); 5th-gen NVSwitch | via nvidia.com ↗2026-08-27 |
| Package / process | 208 billion transistors, custom TSMC 4NP, two reticle-limited dies joined by 10 TB/s chip-to-chip link | via nvidianews.nvidia.com ↗2024-03-18 |
| TDP | ~1000W (SXM module) | via techpowerup.com ↗2026-08 |
Pricing
Price history, newest first
Step chart of per-unit observations (system and order totals stay in the table). The line is drawn in USD at each observation year's average FX rate so mixed-currency histories read as one trend; hover any point for the price exactly as reported.
| Date | Price | Type | Source | Confidence |
|---|---|---|---|---|
| 2026-08 | ~$25K per GPUProduction cost ~$6,400 ($5.7k-7.3k) implies $25k-32k at 75-80% margin | Analyst estimate | via Epoch AI BOM estimate ↗ | low |
| 2025-07-14 | ~$30K per B200 SXM module$30k-40k for the module; DGX B200 8-GPU system ~$515k (via Wccftech); GB200 superchip (1 Grace + 2 B200) $60k-70k | Street price | via Modal (citing Tom's Hardware range) ↗ | low |
| 2025-07-14 | ~$4 per GPU-hour (cloud on-demand, 2025-07)On-demand rates $3.79-18.53/hr across Lambda, RunPod, Baseten, AWS, GCP | Market index | via Modal cloud price survey ↗ | medium |
| 2024-03-19 | ~$30K per GPUHuang on CNBC: Blackwell GPU to cost $30,000-40,000; later clarified pricing varies since Nvidia sells integrated systems | Reported contract | via CNBC (Jensen Huang statement); corroborated by Tom's Hardware and TechPowerUp ↗ | high |
vendor-official or multiple independent reports · single major report or several secondary · estimate or lone secondary source
Prices in USD, converted at each observation year's average exchange rate; the original figure and rate print under every converted price. How prices are sourced and graded ↗
Context
Notes
- No official Nvidia list price; the $30-40k figure is a CEO statement, widely treated as the reference price.
- Announced 2024-03-18 at GTC; partner systems available from late 2024.
Where it sits
In the stack
All specs and prices link to their original sources, dated as observed. Figures are market observations, not offers; treat each as of its date.