Geopolitical tensions drive AI chip independence in US and China, reports TrendForce

TrendForce research indicates that rising demand for AI servers is prompting major US cloud service providers (CSPs) to develop in-house ASICs, with new versions released every one to two years. In China, US export controls introduced in April 2025 are expected to decrease the share of imported chips, such as those from NVIDIA and AMD, from 63% in 2024 to 42% in 2025. Domestic Chinese chipmakers, led by Huawei, are projected to increase their market share to 40%, supported by government policies promoting local AI processors.
CSPs in both regions are focusing on ASIC development to reduce dependence on NVIDIA and AMD, aiming to control costs, enhance performance, and improve supply chain flexibility to manage growing AI workloads. 
Among US CSPs, Google’s TPU v6 Trillium improves energy efficiency and performance for large-scale AI models. Google has shifted from relying solely on Broadcom to a dual-sourcing strategy with MediaTek to enhance design flexibility and reduce supply chain risks. AWS is developing Trainium v2 with Marvell for generative AI and large language model (LLM) training, and collaborating with Alchip on Trainium v3, with TrendForce forecasting significant ASIC shipment growth for AWS in 2025. Meta’s MTIA v2, co-developed with Broadcom, focuses on energy efficiency and low-latency for customized inference workloads. Microsoft, while ...