~$10Klatest observation · analyst estimate · 2026-08
3dated price observations on record
9sourced specs (all linked below)
Specifications
Sourced specs
| Spec | Value | Source |
|---|---|---|
| FP64 compute | 81.7 TFLOPS vector / 163.4 TFLOPS matrix | via amd.com ↗2026-08-27 |
| FP32 compute | 163.4 TFLOPS | via amd.com ↗2026-08-27 |
| TF32 matrix | 653.7 TFLOPS dense / 1,307.4 with sparsity | via amd.com ↗2026-08-27 |
| FP16/BF16 matrix | 1,307.4 TFLOPS dense / 2,614.9 with sparsity | via amd.com ↗2026-08-27 |
| FP8 matrix | 2,614.9 TFLOPS dense / 5,229.8 with sparsity | via amd.com ↗2026-08-27 |
| INT8 matrix | 2,614.9 TOPS dense / 5,229.8 with sparsity | via amd.com ↗2026-08-27 |
| Memory | 192GB HBM3, 5.3 TB/s, 8192-bit | via amd.com ↗2026-08-27 |
| Interconnect | Infinity Fabric: 8 links at 128 GB/s; PCIe Gen5 | via amd.com ↗2026-08-27 |
| Package / TDP | 153B transistors, 19,456 SPs / 304 CUs @ 2100 MHz, 750W peak TBP (OAM) | via amd.com ↗2026-08-27 |
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 | ~$10K per cardEpoch continues to model MI300X at $10k-15k | Analyst estimate | via Epoch AI dataset range ↗ | low |
| 2024-01 | ~$10K per cardMI300X chips cost $10,000-15,000 apiece; TechPowerUp later cited $15k-17k | Street price | via Tom's Hardware (+ TechPowerUp analyst piece) ↗ | medium |
| 2023-11 | ~$10K per cardCiti: Microsoft paying ~$10k per MI300X | Reported contract | via Epoch AI (citing Citi note on Microsoft) ↗ | low |
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
- AMD does not disclose Instinct pricing publicly (an AMD representative confirmed this to Tom's Hardware).
- 2025-2026: MI300X remains widely deployed in cloud fleets (Oracle, Microsoft); China-specific MI308 variant was hit by the April 2025 US license requirement (AMD took an ~$800M charge).
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.