Fujitsu's decision to have Rapidus produce its 1.4nm Neural Processing Unit (NPU) signals that Japan's 'Sovereign AI Silicon' strategy is moving from a possibility to a concrete execution phase with real customers.
This move is a convergence point for two major streams. The first is Fujitsu's own roadmap for an integrated CPU+NPU for its next-generation supercomputer, 'Fugaku NEXT'. The second is the Japanese government's aggressive policy drive, which has funneled massive public and private funds into Rapidus to accelerate its domestic production roadmap from 2nm down to 1.4nm. Essentially, national strategy and corporate need have perfectly aligned.
So, what made this possible? Several factors created the foundation for today's announcement. First, on the policy and economic security front, Japan has been working to reduce supply chain risks. By joining the U.S. and the Netherlands in restricting advanced semiconductor equipment exports to China and simultaneously pouring subsidies into Rapidus, the government created a strong incentive to 'buy Japanese'.
Second, from a technology and process standpoint, Rapidus has been making steady progress. Its collaboration with IBM on 2nm process technology and advanced packaging, coupled with the successful installation of ASML's cutting-edge EUV machines, demonstrated its growing technical maturity. This gave Fujitsu the confidence that Rapidus could handle the demanding 1.4nm node.
Finally, the demand side provided the direct push. The 'Fugaku NEXT' project, a joint initiative by RIKEN, Fujitsu, and NVIDIA, explicitly requires advanced AI acceleration. Fujitsu's own plans for its 'MONAKA-X' chip architecture centered on a 1.4nm process. With early customer wins from Canon and Tenstorrent, Rapidus built a pipeline that proved it was a viable foundry partner, reducing the risk for a large commitment from Fujitsu.
This partnership not only secures a key customer for Rapidus but also allows Japan to internalize some of the massive costs associated with using overseas foundries like TSMC, a crucial step toward semiconductor independence.
- NPU (Neural Processing Unit): A specialized processor designed to accelerate machine learning and AI tasks, much like a GPU is for graphics.
- 1.4nm Process: Refers to a highly advanced generation of semiconductor manufacturing technology. A smaller number generally means more powerful and energy-efficient chips.
- Foundry: A company that manufactures semiconductor chips for other companies that design them (known as 'fabless' companies).