Detailing and benchmarking hardware architectures for current market players and emerging AI chips for server CPUs, custom AI ASICs, and other AI chips, including AI accelerators using heterogeneous matrix-based systems and spatial AI accelerators. China's domestic AI chips took 41% of the accelerator server market in 2025. New data shows Huawei alone shipped roughly 812,000 AI chip units last. These include GPUs, custom AI application-specific integrated circuits (ASICs) used by hyperscalers and cloud service providers (CSPs), AI-capable central processing units (CPUs), and other AI ASICs developed by both AI chip-focused startups and large vendors. Graphics Processing Units (GPUs) The. In October 2023, the US government imposed new export controls that effectively limited the computing performance of advanced chips exported to China. While intended to hinder China's technological progress, these restrictions may have had an unintended consequence – fueling even greater demand for. DeepSeek's latest models highlight China's growing AI ecosystem built on domestic hardware (AI generated). Summary DeepSeek unveils its V4 series, focusing on large-scale mixture-of-experts (MoE) architectures to improve efficiency and scalability. The models are optimized for Huawei Ascend AI. Huawei has officially launched the Ascend 950PR, a 1. 8 times the FP4 performance of Nvidia's H20 — marking the most aggressive challenge yet to American semiconductor dominance from a Chinese chipmaker operating under heavy US sanctions. With 750,000. The DGX H100, launched in 2022, is NVIDIA's latest DGX system iteration and a foundation for DGX SuperPOD deployments. A DGX server is equipped with eight H100 GPUs, featuring 640 billion transistors.