Lisa Su landed in Taipei this week with a list of people to see and a number to raise. On Monday the Advanced Micro Devices chief executive met with executives at Foxconn, Quanta, Asus, and Acer, then traveled to Hsinchu to sit down with Taiwan Semiconductor Manufacturing Co. The message at every stop was the same: AMD’s $10 billion bet on Taiwan’s supply chain is on track, and it is no longer big enough.
Su confirmed that the company will increase the $10 billion Taiwan investment it announced in May, citing demand for AI computing that continues to outstrip supply. When the original commitment was made, she told partners that customers were rapidly scaling AI infrastructure to meet growing compute demand. On this visit she went further, telling the people who build AMD’s chips that memory in particular is in short supply and that the company will significantly raise supply in 2027.
The visit was as much about locking in capacity as about spending money. AMD is extending its planning horizon with Taiwanese partners from one or two years to as long as five, according to local reports, a sign the company expects AI demand to keep growing well past the current build-out. The company has begun production of its MI450-series AI platform on TSMC’s 2-nanometer process, with higher-volume shipments expected in 2027, and the next generation, the MI455, is scheduled to enter production in the first half of that year.
Taiwan is where the AI boom meets its own supply chain. The island hosts TSMC, the world’s dominant contract chipmaker, along with the packaging, substrate, and server manufacturers that turn designs into data-center racks. AMD’s $10 billion commitment is aimed squarely at that ecosystem, with partnerships in advanced packaging through ASE Technology and Siliconware Precision, and in substrates and assembly through Powertech, Sanmina, and Inventec.
Su’s visit carried the weight of a turnaround that has taken years. When she took over AMD in 2014, the company was losing money and its share of the server market had dwindled to almost nothing. A decade of rebuilding around high-performance computing has made it the clearest rival to Nvidia and Intel in the chips that power data centers, and the AI boom has turned that position into revenue. Taiwan has been the foundation of the entire effort.
The money is a fraction of what the industry as a whole is committing. Nvidia, which still dominates the market for AI processors, has tied up a growing share of the world’s high-bandwidth memory and TSMC’s most advanced capacity. AMD’s push is an attempt to carve out a credible second source at a moment when customers, wary of relying on one supplier, are actively seeking alternatives.
The memory crunch Su described is the industry’s newest constraint. High-bandwidth memory, the fast storage that sits next to AI processors, is sold out well in advance, and its cost has climbed as suppliers have struggled to keep pace. AMD’s promise to lift supply sharply in 2027 depends on the memory makers as much as on TSMC, which is why Su framed the trip as a call for cooperation across the whole chain.
Su’s public case has been that AMD can win on compute at the high end while offering buyers a way to hedge. The Venice EPYC server processor, the first high-performance computing product built on TSMC’s 2-nanometer node, began ramping in Taiwan this year, and AMD plans to produce it at TSMC’s Arizona fab as well, diversifying its manufacturing beyond the island. The shift toward agentic AI workloads, which demand more from the processors that coordinate data movement and networking, has worked in AMD’s favor.
The request she carried went beyond silicon. Su used the visit to urge AI companies to cooperate more closely, an appeal with a practical edge: the industry’s bottleneck is no longer chip design but the packaging, memory, and assembly that turn a processor into a working rack. AMD’s ability to ship in 2027 depends on partners whose own capacity is being claimed by rivals, and coordination is the only way to keep the supply chain from becoming the limit on the boom.
The stakes are visible in the numbers. AMD’s Taiwan investment, once increased, will exceed $10 billion, a figure that signals the company’s answer to the central question in the chip business: whether there is room for a strong number two in AI computing. Su is betting there is, and she spent the week asking the people who would have to build it whether they agree.
The larger contest is with the company that set the pace. Nvidia’s lead in AI chips has narrowed AMD’s room for error, and every dollar AMD commits to Taiwan is a bet that customers want a genuine alternative. Su’s trip was the latest move in that argument, conducted in the factories and boardrooms where the capacity that will decide it is being poured into concrete and glass.
