NVIDIA and Qualcomm rumored to consider Samsung for 2nm, as TSMC 2nm costs are 'too high'

NVIDIA and Qualcomm are reportedly considering switching over production of their next-gen 2nm chips to Samsung Foundry, as the costs of TSMC making new 2nm chips is too high.

2

Samsung Foundry needs all the semiconductor customers it can get, and if it can poach two of the biggest in NVIDIA and Qualcomm for its next-gen 2nm (SF2) process node, then that's a huge win for the South Korean semiconductor industry against TSMC.

TSMC is currently in the lead with 2nm node production with a yield rate of approximately 60% and is conducting process testing with customers including Apple, NVIDIA, and Qualcomm, which are all major customers of TSMC. However, in new reports from Koreran media, the high price of TSMC's new 2nm chips and limited production capacity, Apple has moved away from using 2nm for its new chips until 2026, while NVIDIA and Qualcomm are now considering moving to Samsung Foundry.

• Read more: Apple to delay using TSMC's new 2nm node, until capacity expands 8x in 2026
• Read more: TSMC's new 2nm chip yields are 'better than expected' in trial runs in Taiwan
• Read more: TSMC's new 2nm Kaohsiung fab is 6 months ahead of schedule: Apple and AMD are first customers

TSMC's high price on 2nm chips and low quantity, which won't be increased massively until 2026, have NVIDIA and Qualcomm not wanting to rely on TSMC too much, hence the rumors of shifting to Samsung Foundry. These companies hope to diversify their supply chain, and have Samsung Foundry fab its new chips on 2nm.

But, 2nm with Samsung Foundry doesn't = 2nm with TSMC.

The performance of Samsung Foundry's advanced processes aren't the best, where it made 4nm chips for Qualcomm that were overheating and experiencing performance issues. So this might be the last time that Samsung Foundry can come out and swing for high-end 2nm chips for NVIDIA and Qualcomm, and hopefully not miss... or else TSMC will be flooded (too much so) with orders in 2026, 2027, and beyond for advanced nodes.