April 2, 2026 /SemiMedia/ — Fujitsu is working with Rapidus to develop a 1.4nm AI chip for server use, as Japan steps up efforts to build its own semiconductor supply chain.

The chip is a neural processing unit designed for AI inference. Total development cost is estimated at about 58 billion yen, with government support expected to cover a large share.

Unlike GPUs used for training, NPUs are better suited for running AI models. Demand for inference is growing as more AI services move into data centers.

Fujitsu plans to combine the NPU with its Arm-based Monaka CPU in one package. The current version of the CPU uses a 2nm process and targets high-performance computing.

The company will continue to rely on partners such as Nvidia and AMD for GPU capability, with a goal to bring CPU and GPU closer together in future systems.

For Rapidus, the project adds to its push into advanced nodes. The company aims for 2nm production later this decade and is targeting 1.4nm around 2029.