NVIDIA and TSMC have just achieved a major milestone that significantly bolsters the prospects of on-shore chip fabrication, insulating the US against geopolitical shocks. Even so, there is still quite a lot left to be done, including on the advanced packaging front.
TSMC has just fabricated some of the first NVIDIA GB300 GPUs in its Arizona facility, taking a significant step towards fulfilling the demands of the US to onshore chip manufacturing
According to the Commercial Times, TSMC has just fabricated some of the first NVIDIA GB300 AI GPUs on its 4nm node in Arizona's Fab 21.
Even so, the GPUs manufactured in Arizona still have to be sent back to Taiwan for CoWoS advanced packaging before being assembled by system manufacturers.
Therefore, TSMC need only localize CoWoS advanced packaging now to realize its goal of end-to-end AI chip manufacturing in the US, as repeatedly demanded by the US government.
Do note that Foxconn has already built a GB300 AI server manufacturing base in Houston, Texas, while Wistron has also set up a server assembly line in Dallas, Texas.
TSMC plans to spend $265 billion in the US to onshore production, which include two advanced packaging facilities in Arizona, thereby forming a complete manufacturing cluster, replete with an all-new wafer fab.
Once TSMC's CoWoS- and SoIC-related packaging plants come online, it will be able to handle the entirety of the workflows for NVIDIA's Blackwell and Rubin chips within the US. TSMC also plans to onshore its N2 (2nm) and A16 (1.6nm) by 2028.
Follow Wccftech on Google to get more of our news coverage in your feeds.





