Fujitsu plans to develop an AI inference NPU using Rapidus 1.4nm process technology, targeting servers and data centers.

The project carries a total budget of ¥58 billion, with about two-thirds subsidized by Japan’s NEDO to strengthen domestic semiconductor supply chains.

The new NPU will be co-packaged with Fujitsu’s in-house Monaka CPU, a 144-core Armv9 processor built on TSMC 2nm. Monaka uses a 3D chiplet layout and supports PCIe 6.0 and CXL 3.0. Fujitsu aims to deliver the 1.4nm NPU in 2029.

Fujitsu is not developing its own GPU. Instead, it continues working closely with NVIDIA and plans to integrate CPU and GPU on the same substrate by 2030. It is also collaborating with AMD on AI chips.

For Rapidus, Fujitsu becomes a key domestic customer. Rapidus has already partnered with IBM and Tenstorrent, and landed an image processor order from Canon. It is now in talks with over 60 potential customers. Rapidus plans to mass produce 2nm in 2027 and introduce 1.4nm in 2029.

ICgoodFind:Fujitsu’s 1.4nm NPU bet with Rapidus highlights Japan’s push for semiconductor self-reliance while opening new ground in high-end AI silicon.