← All hardware
China's flagship AI accelerator: 96GB HBM, 1.8 TB/s
Pros
- Top non-Western AI accelerator
- Strong memory capacity
- Available where NVIDIA is restricted
- Rapidly scaling supply
Cons
- Limited availability outside China
- CANN ecosystem less mature than CUDA
- Yield/process constraints
- Sanctions complicate access
✓ Where it shines / best for
- Chinese enterprises and cloud providers needing domestic AI silicon
- Large-scale LLM training/inference where NVIDIA is restricted
- Rack-scale cluster deployments via CloudMatrix
✕ Not the best fit for
- Buyers outside China (export and availability constraints)
- Teams reliant on the CUDA ecosystem
- Workloads needing broad third-party software/tooling maturity
Features
- ✓ Data-center scale
- ✓ LLM
- ✓ Inference
- ✓ Training
- ✓ Rack Scale
- ✓ Hbm
- ✓ Cann
- ✓ NPU
Pricing
| Plan | Price | Billing | Notes |
|---|---|---|---|
| Hardware purchase | Custom / regional quote | one-time | Sold to enterprises/cloud operators primarily in China; estimated ~$28,000+ per accelerator in reported deals. Not publicly listed; export-restricted outside China. |
| CloudMatrix / cloud access | Custom quote | contract | Available via Huawei Cloud and CloudMatrix 384 rack-scale systems by contract. |
Pricing verified from the official source. Prices change often — confirm on the vendor's site before buying.
Specifications
| use | Data center training and inference |
| power | ~400W+ class |
| memory | 96GB HBM2e, ~1.8 TB/s |
| performance | ~800 TFLOPS FP16 (est.) |
| architecture | Da Vinci (SMIC 7nm) |
Sponsored
A full review is being generated for this product and will appear here shortly.