~$9Klatest observation · analyst estimate · 2026-08
3dated price observations on record
6sourced specs (all linked below)
Specifications
Sourced specs
| Spec | Value | Source |
|---|---|---|
| Process node | 7nm, chiplet packaging | via tmtpost.com ↗2025-08 |
| Performance positioning | ~80-90% of NVIDIA A100 overall performance (vs 4nm H100, 1-2 generations behind) | via tmtpost.com ↗2025-08 |
| FP16 compute | ~256 TFLOPS (unofficial) | via mirrorfrog.com ↗2026-08 |
| INT8 compute | ~512 TOPS (unofficial) | via mirrorfrog.com ↗2026-08 |
| Memory | 48GB HBM2e (single secondary source) | via mirrorfrog.com ↗2026-08 |
| Interconnect | MLU-Link 3.0 (unofficial) | via mirrorfrog.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 | ~$9K per chipEstimated $8,000-10,000 (midpoint shown) | Analyst estimate | via Mirrorfrog AI card wiki ↗ | low |
| 2025-08 | ~$9.7K from ¥70,000 at 2025 avg 7.19 CNY/USD per chipReported selling price ~6万-8万元/枚 (midpoint shown); slightly below other mainstream domestic AI chips | Reported contract | via TMTPost 钛媒体 ↗ | medium |
| 2025-08 | ~$139.1K from ¥1,000,000 at 2025 avg 7.19 CNY/USD per 8-card server8-card server not more than ~100万元/台 (implies ≤~125,000 CNY/card including system) | Reported contract | via TMTPost 钛媒体 ↗ | medium |
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
- Cambricon does not officially disclose MLU590 specs or pricing; Baidu was an early customer with hundreds of units.
- Official compute and memory specs are not published — vendor never published a datasheet.
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.