Huawei's full-stack AI chip with 640 TFLOPS FP16 and 64GB HBM — the primary domestic choice for large-model training and inference.
A complete domestic chain from silicon to software stack.
Chip, board and CANN/MindSpore all native.
Meets trustworthy-computing and industry compliance.
MindSpore and mainstream framework support growing.
Domestic supply improves total cost of ownership.
Training, inference and industry deployment with high sovereignty and compliance needs.
Yes via adaptation layers; migration cost is low.
GigaX supports heterogeneous scheduling across Ascend and NVIDIA.
Yes, in-region and designated-region private deployment.