Lisa Su Returns to Taiwan as AI Demand Strains the Chip Supply Chain

Lisa Su, CEO of AMD, returned to Taiwan on October 5 for a visit focused on the company’s supply chain, less than five months after her previous trip. The visit comes as strong demand for artificial intelligence (AI) infrastructure keeps pressure on chip manufacturing and advanced packaging capacity.

Key takeaways from Lisa Su’s visit to Taiwan, in 30 seconds

  • AMD’s CEO returned to Taiwan on October 5, just months after her last visit.
  • The company is holding talks with key suppliers as demand for AI infrastructure keeps rising.
  • AMD announced in May an investment of over $10 billion in Taiwan’s ecosystem.
  • The plan includes expanding advanced packaging and developing 2.5D EFB technology with local partners.
  • TSMC remains a central supplier for AMD, but the company is also widening its network of packaging and manufacturing partners.

Su’s new visit comes at a time when capacity availability has become an increasingly critical issue for makers of accelerators and processors destined for data centers. AMD hasn’t announced any major public event tied to the trip for now, but DIGITIMES reports that the executive plans to meet with strategic suppliers.

The situation isn’t entirely new for AMD. In May, during her previous visit to Taiwan, Su explained that processor market demand was outpacing forecasts and that the company was working with its partners to ramp up production quickly.

AMD is looking for capacity beyond chipmaking itself

The AI bottleneck doesn’t depend solely on how many chips a foundry can manufacture. Advanced packaging, substrates, high-bandwidth memory, and other components needed to build AI systems also shape the capacity that ultimately reaches data centers.

AMD has already taken steps to expand this part of its supply chain. In May, it announced investments of over $10 billion in Taiwan’s tech ecosystem over the coming years. The stated goal is to expand strategic partnerships and increase advanced packaging capacity for the next generation of AI infrastructure.

One of those efforts is EFB (Elevated Fanout Bridge), a 2.5D interconnect technology AMD is developing and qualifying together with Taiwanese companies such as ASE and SPIL. AMD is also working with Powertech Technology on a panel-based variant.

The company positions EFB as a technology alternative that can complement packaging solutions like CoWoS, developed by TSMC. Lisa Su herself said in May that AMD would keep securing CoWoS capacity while evaluating how demand and EFB’s development evolve.

That move widens the number of companies involved in building AMD’s platforms. Beyond packaging, the group of Taiwanese suppliers includes substrate makers and companies that help build complete systems.

TSMC remains a key piece for AMD

Even as AMD’s new approach widens its supplier network, Taiwan remains especially important because of its relationship with Taiwan Semiconductor Manufacturing Company (TSMC), the world’s largest contract chipmaker.

AMD confirmed in May that its next-generation EPYC processors, known as Venice, had entered production using TSMC’s 2-nanometer process. The company described the processor as the first high-performance computing product to enter production on that advanced node.

The pressure on TSMC isn’t limited to AMD, either. Demand for AI accelerators has forced the expansion of advanced packaging capacity over the past few years, as chipmakers and major data center customers compete to reserve future capacity.

AMD’s strategy is therefore to secure capacity at multiple points along the chain. The goal isn’t to replace TSMC, but to reduce dependence on any single point where packaging, substrates, or specific processes could become a growth bottleneck.

AMD’s own announced investment reflects that approach. The program includes packaging makers like ASE, SPIL, and Powertech, along with substrate suppliers such as Unimicron, Nan Ya PCB, and Kinsus. Its Helios systems also involve original design manufacturers (ODMs) like Sanmina, Wiwynn, Wistron, and Inventec.

AI growth is reshaping supply chain priorities

AMD is increasing its exposure to the AI infrastructure business as demand grows for processors and accelerators used in data centers. In its first-quarter 2026 results, the company said its data center business had become the main driver of revenue and profit growth. It also pointed to rising demand for CPUs and accelerators tied to inference and agentic AI workloads.

The company is also preparing complete systems for large-scale deployments. Its Helios platform combines EPYC processors, Instinct accelerators, networking, and ROCm software in a rack-scale architecture designed for large AI deployments. AMD said in May it expected to start multi-gigawatt rollouts during the second half of 2026.

That kind of infrastructure means the availability of a single component can affect the entire system. Having enough GPUs or CPUs doesn’t by itself guarantee they can become working servers if there isn’t enough packaging capacity, memory, substrates, boards, or power and cooling systems.

Lisa Su’s new visit to Taiwan comes precisely as AMD shifts from securing capacity for specific products to building a supply network ready for much larger volumes. For now, no specific agreements from this trip have been publicly announced, so any additional capacity changes will have to wait for announcements from AMD or its suppliers.

What is confirmed is that the company already views advanced packaging as a strategic part of its AI expansion, and that it has committed over $10 billion to Taiwan’s ecosystem to strengthen that capacity.

Frequently Asked Questions

Why did Lisa Su return to Taiwan?

AMD’s CEO traveled to Taiwan again to meet with strategic suppliers at a time of strong demand for AI infrastructure. DIGITIMES reports meetings with supply chain partners, though the details haven’t been made public.

How much will AMD invest in Taiwan?

AMD announced in May investments of over $10 billion in Taiwan’s tech ecosystem to expand partnerships and advanced packaging capacity for AI infrastructure.

What is AMD’s EFB technology?

EFB is a 2.5D interconnect technology AMD is developing with Taiwanese partners to boost bandwidth between components and improve system efficiency. The company is working on both wafer-based and panel-based versions.

What role does TSMC play in AMD’s newest chips?

TSMC remains a central partner for AMD. The EPYC Venice processor, for example, has begun production using TSMC’s 2-nanometer process.

Scroll to Top