MindSpore, Zhipu AI & Huawei Ascend: How China Built a Frontier AI Model Without a Single NVIDIA GPU

A Technical & Strategic Deep Dive

745 Billion Parameters. 100,000 Ascend 910B Chips. Zero NVIDIA GPUs.

On February 11, 2026, Zhipu AI released GLM-5 — a frontier-class LLM trained entirely on Huawei’s domestic silicon using the MindSpore framework. This is the full story of how it happened, what makes MindSpore special, and why it matters for every AI engineer on the planet.

1. Introduction: The AI Independence Moment

On February 11, 2026, Chinese AI lab Zhipu AI (rebranded as Z.ai) released GLM-5, a 745-billion-parameter large language model that sent shockwaves through the global AI industry. The model wasn’t just impressive for its benchmark scores — which rival those of OpenAI’s GPT-5.2 and Anthropic’s Claude Opus 4.5 — but for something far more consequential: every single parameter was trained on Huawei Ascend chips using the MindSpore framework, with zero dependency on NVIDIA hardware.

This represents a watershed moment in the global AI landscape. Despite being placed on the U.S. Entity List in January 2025 — effectively cutting off access to NVIDIA’s H100 and A100 GPUs — Zhipu demonstrated that frontier-class AI models can be built on a…