Lisa Su Visits China and Taiwan

Lisa Su spent the past week visiting China and Taiwan, highlighting the role that AMD will play in the next stage of AI development.

During the trip, Su pointed to the rise of autonomous AI systems, often referred to as agentic AI, as a key shift in the industry. While AMD may not design the most advanced AI chips, Su told reporters at CommonWealth Magazine’s 45th Anniversary Summit in Taipei that she expects agentic AI to create new demands that the company is well-positioned to meet. 

Up until around the end of 2025, much of the focus in AI infrastructure was on graphics processing chips, or GPUs, which can run multiple tasks at once and play a central role in AI training and chatbot AI. Nvidia, led by Su’s distant cousin Jensen Huang, dominates the AI GPU market.

Now, with the emergence of AI agents like OpenClaw, AI systems can take action, which increases the importance of central processing units, or CPUs. These are general-purpose chips that can be used for a wide range of tasks. Earlier in the month, AMD published an explainer on how agentic AI changes the chip equation, noting that while chatbot AI requires at least four GPUs for every CPU, agentic AI is moving toward a one-to-one ratio.

“Over the next five years we see the CPU market growing at over 35% each year, and this is an area where we’re very strong,” Su said, adding that AMD technology “is very, very well represented” across all the top users in North America and China. “Our strategy is not just one CPU, we’re building a whole family of CPUs … I think that’s how you win.”

AMD announced last Thursday that its latest CPU processor, called Venice, has entered into production at TSMC. Intel’s counterpart to Venice, AMD’s main direct competitor, is rumored to be delayed until mid-2027, according to Tom’s Hardware.

But AMD, like other chip companies, is facing a bottleneck in advanced packaging, the last step of the chipmaking process when chips are assembled and sealed. TSMC accounts for about 90% of advanced packaging for AI chips, and Nvidia has secured around 60% of its capacity, according to previous reporting by Domino Theory. At an electronics expo in Taipei last month, a Taiwanese packaging executive told Domino Theory the tightness in this market has created an opening for companies other than TSMC to provide advanced packaging solutions.

Indeed, AMD also announced last Thursday that it is investing more than $10 billion “to scale advanced packaging capabilities” in Taiwan. Since the lead time on these investments is long, AMD is “co-investing” with local packaging companies like ASE, SPIL and PTI to ramp up capacity for the next several years, Su explained.  

Su beat Huang, Nvidia’s CEO, to Taiwan by a few days. Huang, who recently joined President Donald Trump’s delegation to Beijing, arrived in Taipei on Saturday. He will give the keynote address at the Nvidia GTC conference on June 1.