Advanced Micro Devices (AMD) is making a significant financial commitment in Taiwan, allocating over $10 billion to fortify its semiconductor supply chain. This substantial investment, projected to span until 2029, is not solely directed towards Taiwan Semiconductor Manufacturing (TSMC) but encompasses a broader spectrum of Taiwan's semiconductor industry. The funds will be strategically deployed across advanced packaging, chip substrate development—essential for modern chip packaging—and the expansion of manufacturing capabilities for complete artificial intelligence (AI) systems. This forward-looking strategy by CEO Lisa Su aims to guarantee AMD's capacity to produce sufficient hardware, particularly as the burgeoning demand for AI translates into widespread deployments of its innovative products, such as the Helios AI racks.
AMD's foresight in addressing potential bottlenecks in its AI hardware production is paramount. The company's Venice EPYC server CPU is already in production using TSMC's advanced 2-nanometer process, and AMD leverages TSMC's cutting-edge SoIC-X and CoWoS-L packaging technologies for its AI and data center chips. Recognizing the need to diversify and strengthen its manufacturing ecosystem, AMD is also collaborating with other key players. It is actively developing next-generation Elevated Fanout Bridge (EFB) chip packaging in partnership with ASE Technology and Siliconware Precision Industries, having successfully completed testing of a panel-based EFB version with Powertech Technology. Furthermore, AMD is engaging with various Taiwanese substrate suppliers and manufacturers, all working collectively to enable high-volume production of Helios AI systems. This multi-faceted approach is critical, especially considering previous warnings from TSMC's CEO, C.C. Wei, regarding the limitations posed by tight advanced-packaging capacity on customer growth, indicating that without adequate packaging and assembly capabilities, even robust AI demand could result in missed sales opportunities for AMD.
The imperative to expand manufacturing capabilities stems directly from AMD's ambitious growth projections for its data center segment. In the second quarter, data center revenue surged by an impressive 107% year-over-year, contributing approximately 58% to AMD's total revenue. Management anticipates this sector to achieve a compound annual growth rate (CAGR) exceeding 60% over the next three to five years, with data center AI specifically targeted for over 80% CAGR. To accommodate such aggressive growth, a significantly expanded supply chain is indispensable. A single Helios AI rack, for instance, integrates 72 Instinct MI455X GPUs and 18 Venice CPUs. Major clients like OpenAI, Meta Platforms, and Anthropic have already announced AMD deployments that could cumulatively require as much as 14 gigawatts of power, though these will unfold over several years. While AMD operates a fabless model, it has still committed substantial capital—purchasing $1.2 billion in property and equipment and accruing $30.3 billion in unconditional commitments for wafers, substrates, components, cloud capacity, software, and technology licenses in the first half of 2026. This demonstrates that even without owning fabrication plants, securing supply necessitates considerable capital outlay. The ultimate success of this expanded capacity will depend on its ability to drive profitable AI growth, as AMD's non-GAAP operating margin of 27% in the second quarter remains below the management's target of over 35% for the coming years. Lisa Su's multi-billion-dollar investment underscores AMD's commitment to ensuring it can meet anticipated customer demand for AI hardware and achieve its financial objectives.
AMD's substantial investment in Taiwan's semiconductor infrastructure reflects a strategic imperative to secure and scale its AI hardware production. By enhancing advanced packaging, chip substrate development, and overall manufacturing capacity, the company aims to capitalize on the soaring demand for AI, ensuring its ability to meet ambitious revenue growth targets in the data center market while navigating the complexities of a capital-intensive, even fabless, operational model.
