#Taiwan #Taipei #AI Chips #geopolitical tensions #TrendForce

The Race for AI Chip Independence Amid Geopolitical Tensions

TrendForce's recent findings highlight a significant shift in the AI chip market fueled by geopolitical factors. Major Cloud Service Providers (CSPs) in the US and China are racing to develop their own Application-Specific Integrated Circuits (ASICs) to reduce reliance on foreign imports and enhance market control.

As the global demand for AI servers escalates, US CSPs are expediting their in-house ASIC development. New versions of these chips are being rolled out every one to two years, indicating a robust commitment to innovation. On the other hand, Chinese CSPs are adapting to new US export restrictions that are projected to cut the share of imported chips from leading manufacturers like NVIDIA and AMD from 63% in 2024 to approximately 42% in 2025.

China's Response to Export Controls

The domestic chip market in China is undergoing a transformation, with companies such as Huawei emerging as strong contenders. Supported by government policies that favor local production, Huawei is anticipated to boost its market share to 40%. This significant growth can be attributed to rising demand for AI solutions within China, particularly in sectors such as smart cities and telecommunications.

Meanwhile, Chinese chipmaker Cambricon is also addressing the needs of the AI market by expanding its Siyuan chip family, designed for AI training and cloud inference. Initial tests with prominent local CSPs in 2024 indicate a full-scale deployment by 2025, contributing to China's further independence in the AI chip landscape.

US Giants Leading the Charge

In the US market, Google is leading with its TPU v6 Trillium, which offers substantial enhancements in energy efficiency and performance tailored for expansive AI applications. The company has moved from a singular supplier model with Broadcom to a dual-sourcing strategy by bringing MediaTek on board. This change is not only mitigating supply chain risks but also stimulating more aggressive technological advancements.

Amazon Web Services (AWS) focuses on its Trainium chip line, co-developed with Marvell, specifically aimed at generative AI and training large language models (LLMs). This commitment positions AWS well, as it anticipates the strongest growth in ASIC shipments among US companies in 2025.

Meta, now fleshing out its AI infrastructure with the MTIA (Meta Training and Inference Accelerator), is collaborating with Broadcom on the MTIA v2. This enhanced version prioritizes energy efficiency while ensuring it meets Meta's unique performance metrics needed for specific AI applications.

Microsoft, heavily relying on NVIDIA GPUs, is also making strides in ASIC development with its Maia series targeted for Azure's generative AI needs. Though still in a developmental phase, the collaboration with Marvell aims to escalate its capabilities and ensure robust supply chain solutions.

A Divided Market Ahead

With shifting geopolitical landscapes and supply chain dynamics, the AI server market seems set for divergence. As Chinese CSPs accelerate their ASIC development, they are firming up their position in the market while aiming to rival international standards. Companies like Alibaba, Baidu, and Tencent are paving the way for a formidable domestic ecosystem optimized for AI deployment.

TrendForce emphasizes that in light of these developments, the global AI server industry might evolve into two distinct ecosystems: one primarily governed by Chinese technological advancements and the other led by US capabilities. As these trends unfold, the balance of power in the AI sector will continue to evolve, significantly affecting global supply chains and market strategies.

For continual updates and market intelligence reports, visit TrendForce's website or connect with their research department for insight on future tech trends and forecasts.