昇腾

Shi Shi Tech builds an intelligent compute grid integrating heterogeneous domestic AI chips, achieving 40% lower token cost, 30–50% higher throughput, and 99.9% availability—enabling a paradigm shift from raw compute resources to standardized, scalable token production capacity.

DeepSeek V4 marks a paradigm shift from ‘chip adapting to models’ to ‘chip-model co-design’ in China’s computing ecosystem; with CANN open-sourced, developers now solve issues autonomously, 70+ mainstream LLMs are plug-and-play on Ascend, AIGCode achieves 65% MFU, USTC’s LU solver reaches up to 200× speedup, and financial-grade AI systems are deployed in core risk control—Kunpeng/Ascend developer base exceeds 4.1 million.

A Chinese AI company has broken the bottleneck of running a 60 billion parameter model on mobile devices using ternary quantization, saving 6x memory with minimal performance loss.

Huawei drives a body-aware intelligence startup boom through internal talent flow and external policy support, forming a comprehensive industry ecosystem.

文章标题提及DeepSeek V4在昇腾平台的性能表现促使客户快速下单，但正文内容无法正常访问，无法评估深度、新颖性、实用性与可信度。