Expired50% confidenceFactExact time
The GLM-5 series was trained end-to-end on Huawei Ascend 910B chips, using approximately 100,000 units, paired with Huawei's self-developed MindSpore framework, with zero dependence on NVIDIA throughout.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026(expired)
Sources
GLM-5.2 Deep Dive: Million-Token Context, MIT Open Source & Full-Stack Domestic Chip Training
bilibiliAI技术投降派6/14/2026
Related Claims
UnverifiedA research team successfully completed post-training of the DeepSeek v4 Pro model using Huawei Ascend 910C chips.72% similarUnverifiedZhipu's model is reportedly trained on Huawei Ascend 910D chips, with the 910D targeting a benchmark against NVIDIA's H100.71% similarUnverifiedGLM-5 and GLM-5.1 have completed inference adaptation for 7 domestic chips: Huawei Ascend, Moore Threads, Cambricon, Kunlunxin, Metax, Enflame, and Hygon.70% similarUnverifiedZhipu AI stated that GLM-5.2 was trained on full-stack domestic chips.69% similarUnverifiedHuawei has built a full-stack self-developed ecosystem comprising five components: Ascend chips, MindSpore framework, Pangu model, HarmonyOS, and Kirin chips.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/46082API
curl https://kongchang.com/api/v1/knowledge/claims/46082MCP
get_claim(id=46082)