AMD Announces Investment of More Than $10 Billion in Taiwan, Securing Advanced Packaging Capacity First: Cooperating with ASE, Powertech, and Inventec

AMD announces over $10 billion in investments in Taiwan, collaborating with backend packaging factories including ASE, Powertech, Inventec, and Sanmina.
This money isn't chasing TSMC's front-end wafers but is competing for the final production bottleneck of AI chips: advanced packaging capacity.
(Background: AMD's investment causes OpenAI's stock to surge 24%, fully challenging Nvidia's Cuda?)
(Additional context: Nvidia's Q1 financial report is extraordinary! Revenue hits $81.6 billion, a record high, with Jensen Huang exclaiming "The era of Agentic AI has arrived")

Earlier today (21), AMD stated it will invest over $10 billion to deepen Taiwan cooperation, with partners including ASE, Powertech, Inventec, and Sanmina.

Packaging is the real bottleneck

Many believe AI chip production capacity is limited by TSMC's wafer fab, but they overlook the backend advanced packaging.

The so-called "advanced packaging," in simple terms, is: bonding multiple chips onto a substrate in high density so they can communicate like a single chip. Nvidia's H100, A100, and AMD's MI300X all rely on this process, also known as CoWoS (Chip on Wafer on Substrate).

After the wafer is produced, this step is also a critical process that determines whether the chips can be shipped.

From 2023 to 2025, the entire AI computing market has repeatedly faced capacity shortages in CoWoS, causing Nvidia to delay shipments. Taiwan's ASE and Powertech are the core companies doing this business. AMD's recent investment explicitly focuses on "increasing packaging capacity," aiming to lock down backend production lines within its own funding scope.

Su Zifeng has arrived in Taiwan

CEO Su Zifeng is currently visiting Taiwan and will participate in a media roundtable on Friday. It is understood that, besides attending Computex, he may also meet with TSMC Chairman Wei Zhejia and key supply chain partners to secure 2nm advanced process capacity.

Meanwhile, the U.S. government continues to promote "de-Taiwanization" of the chip supply chain, and Nvidia's H20 chips face export restrictions to China. At this geopolitical juncture, AMD's public declaration to deepen Taiwan investment is a clear move on the strategic chessboard: Taiwan's supply chain role will not change due to U.S. pressure.

By 2026, U.S. tech giants' AI spending has already exceeded $700 billion. Ultimately, this money flows into computing power, which flows into chips, then into packaging, and finally into Taiwan.

The hardware competition in AI has never been just about chip design; whoever locks every step of the supply chain into their capital map first will have the qualification to claim long-term dominance.