According to supply chain sources, AMD has terminated its cooperation with Samsung Foundry on the 4nm process and has instead transferred its EPYC server CPU orders to TSMC’s factory in Arizona, USA. At the same time, AMD has confirmed that its next-generation EPYC “Venice” processor will adopt TSMC’s 2nm process, becoming the first high-performance computing (HPC) product based on this technology.
AMD had previously partnered with Samsung, planning to use Samsung’s SF4X 4nm process to produce “Prometheus” processors based on the Zen 5c architecture, covering product lines such as EPYC server CPUs, Ryzen APUs, and Radeon GPUs. This collaboration was seen as AMD’s attempt to diversify its supply chain and reduce its reliance on a single foundry. However, current indications are that Samsung is still lagging behind, with a significant gap in advanced process yields compared to TSMC, leading to AMD’s final decision to abandon this plan. Industry analysts point out that although Samsung’s 4nm process yield has improved from 50% at the beginning of the year to 75%, approaching TSMC’s 80%, it is still insufficient to meet AMD’s stringent requirements for high-performance server chips. Moreover, Samsung’s 3nm process yield is only 60%, currently mainly serving its own products such as the Galaxy Watch7, making it difficult to attract external high-end customers.
In contrast, TSMC’s 4nm process has demonstrated mature technological capabilities and stable production performance. TSMC’s Fab 21 in Arizona began mass production of 4nm chips at the end of 2024, with initial customers including AMD, Apple, and Qualcomm. AMD’s EPYC server CPUs, such as the fifth-generation EPYC 9005 series, have partially adopted TSMC’s 4nm process, featuring up to 128 Zen 5 cores or 192 Zen 5c cores, with significant improvements in performance and energy efficiency. TSMC’s supply chain flexibility and global layout also provide AMD with more reliable production guarantees. The second factory in Arizona is expected to start production in 2028, using 2nm and 3nm processes.
In the 2nm field, TSMC’s N2 process has become the focus of the industry. AMD’s EPYC “Venice” processor, based on the Zen 6 architecture, has completed tape-out and entered the verification stage, with commercialization planned for 2026. This is TSMC’s first HPC product using the 2nm process, marking AMD’s technological leadership in the high-performance computing market. TSMC’s N2 process uses nanosheet transistors (GAAFET), which, compared to the 3nm process (N3E), offers a 10% to 15% performance increase at the same power consumption, or a 25% to 30% power reduction at the same performance. Its defect density is already comparable to 3nm and 5nm, showing the rapid maturity of the process. TSMC expects to start mass production of 2nm chips in the second half of 2025, with an initial monthly capacity of about 50,000 wafers, which is expected to double by 2027.
In terms of market demand, TSMC’s 2nm process has attracted many technology giants. Apple plans to use it for the A-series chips in the iPhone 18 series, and Nvidia may adopt this process in its Vera Rubin architecture. AMD, as one of the first 2nm customers, with its leading deployment of EPYC “Venice,” is expected to further expand its share in the server market. TSMC’s 2nm capacity planning is also quite proactive. In addition to the Taiwan factories, the third factory in Arizona is expected to start producing 2nm chips in 2028, ensuring long-term supply capabilities.
Currently, Samsung is actively promoting the research and development of its 2nm process, with mass production planned for 2025, and is using discount strategies to attract customers such as Nvidia and Qualcomm. In addition, Samsung’s progress in HBM4 high-bandwidth memory has also won the attention of Nvidia, Google, and Broadcom. Samsung’s market share in advanced processes still lags far behind TSMC, which accounted for about 62% of the global foundry market in 2024, with the capacity utilization rates of its 3nm and 5nm processes close to 100%.
TSMC’s 2nm process is undoubtedly a bellwether for the current semiconductor industry, and its technological advantages and market demand are reshaping the foundry landscape. In the future, with the full mass production of the 2nm process and more customers joining, TSMC is expected to continue to consolidate its industry leadership. The subsequent development will depend on whether Intel’s 18A process can succeed, which, if it does, will undoubtedly change the market landscape.