DeepSeek-Huawei AI Deal: 9 Ways China is Building an Alternative to Nvidia’s Ecosystem
Poulami Saha
DeepSeek-Huawei Partnership: DeepSeek and Huawei are developing programming infrastructure designed specifically for Huawei’s Ascend AI accelerators.
Open-Source AI Tools: DeepSeek has open-sourced software covering programming, computation and communication to support broader Ascend development.
TileLang Programming Language: TileLang provides a high-level programming layer intended to simplify development and help optimise workloads for AI chips.
Challenge to Nvidia CUDA: The initiative targets a major Nvidia advantage: its established CUDA software ecosystem supporting AI development and optimisation.
128-Chip Supernode System: DeepSeek and Huawei jointly advanced a supernode using 128 Ascend 950 chips, optimising computation and chip-to-chip communication.
Six Software Components: DeepSeek released six Ascend-focused components spanning programming, computation, communication, attention, kernels and data-processing workloads.
Building Developer Adoption: Huawei says Ascend now supports more than 90 leading open-source projects, including PyTorch, Triton, vLLM and veRL.
Limits Still Remain: Huawei faces component constraints, while DeepSeek continues using Nvidia hardware for some demanding model-training workloads.