Foundry giants, as well as ambitious newcomers, have been making strides in capacity expansion as soon as 2025 kicks start. According to SEMI, the semiconductor industry is set to begin 18 new fab construction projects in 2025.
Notably, this trend seems to be most evident in advanced nodes. SEMI suggests that chipmakers are rapidly increasing capacities for advanced nodes (7nm and below), with a projected 16% annual growth rate in 2025.
As the mass production of TSMC Arizona’s first fab will soon begin, Japan’s Rapidus, as an emerging challenger, is also trying to build momentum by accelerating its 2nm development, eyeing to start trial production in April.
2025 would certainly be a busy year for foundries amid the AI boom and robust HPC demands. Here’s a quick roundup of the most cutting-edge, newly added capacities by foundries in 2025, and beyond.
TSMC
TSMC’s first fab in Arizona, featuring N4 process, is anticipated to start mass production in 1H25. Notably, as per TrendForce, with an initial capacity of 20K wafers per month by year-end, the facility’s early production will primarily serve U.S.-based customers, including AMD and Apple. Additionally, there are rumors that a portion of AWS’s ASIC chips will also be produced at the AZ fab.
TSMC’s second Arizona fab, on the other hand, will feature N3 and N2 processes and be operational in 2028.
Following the mass production of its first Kumamoto fab in late December, 2024, TSMC’s second fab in Japan is scheduled to begin mass production by the end of 2027, featuring 6/7nm. As per Commercial Times, the plant’s construction is expected to start in the first quarter of 2025.
Taiwan will still remain as the major hub for TSMC’s 2nm production, as its Baoshan Plant in Hsinchu, northern Taiwan, will kick off mass production in 2025.
Intel and Samsung
Intel and Samsung share similar challenges as they both strive to secure major clients in cutting-edge nodes other than in-house orders, while facing delays in their capacity expansion plans in the U.S.
Like TSMC, Arizona also plays a critical role in Intel’s advanced process blueprint. Intel’s Fab 52 and Fab 62 there, after three years of construction, are likely to begin operations in early 2025.
The fab aims to manufacture chips using Intel’s Angstrom-era process technology, including the 18A node. Notably, at CES 2025, Intel confirmed its Panther Lake processor, the first to utilize 18A, will debut in 2H25.
Samsung, on the other hand, has been reportedly facing delays on its Taylor plant due to ongoing issues with the 2nm yield, as per Business Korea. The mass production timeline is said to be pushed back from late 2024 to 2026.
Notably, according to TechCrunch, Samsung revealed in March 2023 its plans to invest around USD 230 billion (300 trillion won) to construct five new memory and foundry fabs in South Korea. As the investments will be made through 2042, whether and to what extent the promise could be materialized remains to be watched.
Rapidus
In the meantime, Japan’s Rapidus seems to have made solid progress on 2nm. After it received the first EUV machine from ASML in December, 2024, the company now aims to begin trial production for 2nm as soon as April.
It is worth mentioning that its first foundry IIM-1, which is currently under construction in Chitose, Hokkaido, is expected to kick off mass production in 2027. A Nikkei report reveals that the company aims to supply 2nm chip samples to U.S. chip giant Broadcom by June, 2025.
(Photo credit: Rapidus)
Read more
(Photo credit: Intel)