TSMC’s 3nm project list grows: AMD MI350 series, NVIDIA Rubin AI GPU, expected in 2H 2025

TSMC’s list of 3nm customers is growing, with AI added to the 3nm project list with AMD’s upcoming Instinct MI350 AI accelerator series, and NVIDIA’s next-gen Rubin R100 AI GPU.

TSMC's 3nm project list grows: AMD MI350 series, NVIDIA Rubin AI GPU, expected in 2H 2025 84

VIEW GALLERY – 2 IMAGES

In a new report from Ctee citing investment bankers, TSMC will fab the new MI350 series and Rubin R100 series AI chips in the second half of 2025, joining 3nm customers with Apple, Intel, MediaTek, and Qualcomm… as well as AMD on its CPU side making chips on TSMC’s 3nm process node (already, and more into 2025).

The full list of TSMC’s 3nm project list includes the Apple M4, Apple A18 and A18 chips for the iPhone, MediaTek’s new Dimensity C-X1 (auto chip), Qualcomm’s upcoming Snapdragon 8 Gen4 chip, Intel’s new Lunar Lake and Arrow Lake CPUs, AMD’s new Zen 5 CPUs and upcoming Instinct MI350 series AI accelerators, and finally, NVIDIA’s next-gen Rubin R100 AI GPU (which will sport future-gen HBM4 memory).

AMD’s new Instinct MI350 series shifting to TSMC’s new 3nm process node will bring a more level playing field in the advanced process node field, with NVIDIA’s new Rubin R100 AI GPU also made on the 3nm node. Ctee cites its sources, saying that NVIDIA’s orders to TSMC in 2025 will exceed the numbers of orders in 2024, with its 3nm and 5nm production capacity stretched even further.

There’s also the NVIDIA + MediaTek co-developed Arm-based AI PC processor that will be built on 3nm at TSMC, which will extend production capacity even further, with this new AI PC chip to be revealed in Q2 2025, and mass produced in Q3 2025. Intel’s new Lunar Lake CPUs are also built on 3nm, with most of its future products gearing up for production at TSMC, too.

Industry insiders have said that with market trends the way they are, that the entry of new AI chips into the battle on TSMC’s new 3nm node will consume more wafer foundry production capacity, especially in the advanced packaging side of things.

NVIDIA’s new Blackwell B200, AMD’s Instinct MI300, and Intel’s Gaudi 3 AI chips use CoWoS advanced packaging, which is 3.3x the size of the TSMC photomask, and can only cut about 16 wafers. Moving forward, we’ll see an increase to 5.5x, with the number of wafers being cut becoming less.

A list of chips being fabbed on TSMC’s new 3nm process node:

  • Apple M4
  • Apple A18 (iPhone)
  • Apple A19 (iPhone)
  • MediaTek Dimensity C-X1 (auto)
  • Qualcomm Snapdragon 8 Gen4
  • Intel Arrow Lake CPUs
  • Intel Lunar Lake CPUs
  • AMD Zen 5 CPUs
  • AMD Instinct MI350 AI accelerator
  • NVIDIA Rubin AI GPU