CONNECT WITH US
Sign out

Why Taiwanese tech giants pour record funds into US AI server hubs

, Taipei
0

Credit: Digitimes

Taiwanese EMS accelerates US and global investments amid AI server demand and geopolitical dynamics

Major EMS providers are initiating an unprecedented wave of global investments, injecting over NT$20 billion (US$666 million) on a single day. Compal Electronics, Quanta Computer, Inventec, and Wistron Group each revealed large-scale commitments, focusing chiefly on expanding production capacity for AI servers in North America and Southeast Asia.

This surge in capital deployment contrasts with a decade of relatively cautious investment and is driven primarily by geopolitical dynamics and the accelerating demand for AI server infrastructure. Both factors are reshaping manufacturing strategies and prompting Taiwanese firms to prioritize facilities that align with US market needs.

Geopolitical influences and AI infrastructure

The current investment momentum is largely a response to the Trump administration's "Made in America" policy and efforts to curb reliance on China, creating incentives for relocating production closer to or within the US. Compal, Quanta, and Wistron are targeting US locations directly, while Inventec's expansion in Thailand caters to US-based assembly demands, reflecting a strategic geographic spread.

Compal Electronics is allocating US$300 million to establish new manufacturing operations across the US and Mexico. The investment is split between a US$225 million increase in capital for Compal Americas (US) Inc. and a US$75 million injection to form a new US subsidiary. The company plans to build server production facilities focusing on L10 to L11 server manufacturing, with completion anticipated by mid-2026.

In parallel, Quanta Computer is dedicating US$170 million toward expanding its Tennessee and Mexican factories, with the Mexican plant shifting focus from automotive electronic assembly to producing higher-tier servers beyond motherboard manufacturing. Trial runs aim to start by the end of this year, with full production scaling in early 2026.

In Southeast Asia, Inventec is investing THB5.1 billion (around US$157.36 million) to construct a new factory in Thailand and increase capital for its local subsidiary. This move addresses the growing demand for servers and notebooks intended for US customers. Wistron Group, following investments exceeding US$1.2 billion since last quarter across locations in Texas, California, and Mexico, announced a further US$62.5 million allocation to enhance its Dallas Eagle site, primarily supporting after-sales service operations.

G2 tariff tensions and strategic adjustments

Tariff volatility has been a significant concern, but these investments aim to mitigate such risks by relocating production away from China and closer to the US market. Compal's chairman Ray Chen stated that shifting manufacturing lines has reduced exposure to tariffs. Wistron's chairman Simon Lin echoed this sentiment, emphasizing that during the current AI-driven market expansion, tariff impacts are minimal and considered a minor disruption.

The investment strategies align strongly with the rapidly growing demand for AI server production. Companies are also committing resources to broaden research and development capabilities to cater to a wider array of chip platforms. For example, Quanta has historically concentrated on Nvidia Corp.'s products but now plans to increase its focus on ASIC servers, projecting its share of production to grow by 2026. Expanding R&D and production flexibility form core elements of these capital projects.

Financing these large-scale initiatives necessitates raising additional funds or reallocating assets. Quanta plans to issue unsecured convertible bonds overseas with a maximum value of US$1 billion, while Wistron intends to optimize capital by divesting a stake in Tube Incorporated for no less than US$11 million. Wiwynn Technology Corp., a related player, also announced a US$500 million capital increase for its US subsidiary to underpin operations.

Article translated by Jingyue Hsiao and edited by Joseph Chen