On July 24, AMD CEO Lisa Su revealed that chips manufactured for AMD at TSMC's Arizona fab cost 5% to 20% more than those made at TSMC's facilities in Taiwan. Speaking at a tech-policy forum hosted by the All-In Podcast and Hill and Valley Forum, Su confirmed AMD expects its first U.S.-produced chips to launch by the end of 2025.

While a previous TechInsights report estimated wafer costs in Arizona to be less than 10% higher than in Taiwan, Su's comments suggest that total costs—factoring in construction, logistics, and local economic conditions—are more significant. Despite high labor costs in the U.S., labor accounts for less than 2% of total wafer cost due to high fab automation. Still, equipment, construction delays, and regulatory hurdles drive up overall costs.

TSMC founder Morris Chang previously warned that chip production in the U.S. could cost up to 60% more than in Taiwan. Other reports from firms like Exyte and Macquarie have echoed concerns, citing slower fab construction timelines and ongoing challenges in establishing local chemical supply chains. Macquarie estimates Arizona wafer costs could be 30% higher than those in Taiwan, potentially eroding 1%–2% of gross margins for nodes like 4nm.

Su's remarks indicate TSMC may still be absorbing part of these elevated costs rather than fully passing them on to customers. Yet growing geopolitical risks and U.S. trade policies could reduce cost advantages associated with sourcing from Taiwan. Tariffs on Taiwan-made chips or finished goods could narrow the price gap between U.S. and Taiwan fabs, making domestic sourcing more competitive.

AMD was one of the earliest customers to place 4nm orders with TSMC's Arizona fab and has plans to expand to 2nm for its EPYC “Venice” data center CPUs. In April, AMD announced that Venice became the first high-performance computing product to tape out on TSMC Taiwan's 2nm process. The company also confirmed that its fifth-generation EPYC chips have been successfully validated on the Arizona line.

Su also addressed the booming demand for AI chips, citing projections that the AI accelerator market could exceed $500 billion within a few years. “Five years from now, you'll be able to judge whether we made the right decisions based on our accomplishments,” she said, referencing investments by OpenAI CEO Sam Altman and xAI CEO Elon Musk.

AMD's 2024 revenue hit a record $25.8 billion, with data center revenue doubling year-over-year to $12.6 billion. Sales of EPYC CPUs and Instinct GPUs fueled this surge, with data center revenue reaching $3.7 billion in Q1 alone—up 57% year-over-year. JPMorgan analysts expect AMD's AI GPU business to grow by more than 60% in 2025, driven by its next-generation MI350 accelerators.

None of this growth would be possible without TSMC's advanced process technology. As Taiwan's fabs near full capacity and global supply chains become increasingly politicized, AMD and other U.S. chipmakers are turning to TSMC's Arizona facility despite higher costs.

Meanwhile, AMD has made a historic leap in the server CPU market. According to Wccftech and PassMark, AMD's EPYC processors have captured a 50% market share, ending Intel's decades-long dominance. Just a few quarters ago, AMD held only 20% of the server CPU market—a sharp rise credited to the Zen-based EPYC lineup and consistent execution under CEO Lisa Su's leadership.

Back in 2017, AMD's server share was just 2%. The turnaround began with Su prioritizing the data center segment, launching Naples—AMD's first Zen architecture server chip. In contrast, Intel has struggled with delays and leadership instability. Despite architectural improvements to its Xeon line, Intel's lagging process technology has pushed customers toward AMD.

As AMD continues its partnership with TSMC and pushes the envelope on process nodes and AI acceleration, its influence in both HPC and data center markets looks set to expand further.