The first fully domestic cloud AI accelerator and complete inference servers, built on advanced domestic process technology.
China's first domestically developed 12nm large-model AI compute chip — co-developed with our R&D partners, optimized for training and inference workloads of large language models.
Concept Render
The first fully domestic cloud AI accelerator with a secure, controllable supply chain.
32Kx32K image codec · 40-ch FHD encode · Fully domestic supply chain
16 × J20 · PCIe 5.0 ×16 fully interconnected · Full-scale DeepSeek 671B on a single server
| Form Factor | 4U rack server |
| CPU | 2 × Intel® Xeon® 4th/5th Gen |
| Accelerators | 16 × Huajun J20 |
| Compute | FP16 2.24 P FLOPS · INT8 4.48 P TOPS |
| Memory | 4096G device memory · 32 × DDR4 slots |
| Topology | PCIe 5 ×16, 128G/s interconnect |
| Power | 2700W (TDP) · air cooled |
| Energy Efficiency | 1.87 TOPS/W |
R&D and manufacturing of domestic computing servers and AI accelerator hardware, building a collaborative ecosystem of R&D, manufacturing and market operations.