Japan’s Push for AI Chip Sovereignty
Japan is placing a $16.3 billion bet on reclaiming its position in the global semiconductor industry. The government recently approved an additional 631.5 billion yen ($4 billion) in subsidies for Rapidus, a domestic chip foundry startup, as part of an aggressive strategy to produce the world’s most advanced artificial intelligence processors entirely within Japanese borders. At the center of this initiative sits Fujitsu, the century-old technology conglomerate that is developing a cutting-edge 1.4-nanometer AI inference chip to be manufactured by Rapidus at a new facility in Hokkaido by the decade’s end.
The project represents Tokyo’s most ambitious attempt yet to reduce dependence on Taiwan Semiconductor Manufacturing Co. (TSMC) and other foreign suppliers amid escalating geopolitical tensions and supply chain vulnerabilities. Industry Minister Ryohei Akazawa emphasized the strategic imperative during a recent Rapidus event in Hokkaido.
No matter how excellent technology a firm may have, its business will fail unless it obtains customers who want to use the technology.
Akazawa explained that government support aims to help Rapidus create demand and secure customers beyond the company’s own marketing efforts, reflecting a recognition that manufacturing capabilities require a robust ecosystem of domestic chip designers willing to take risks on unproven foundry technology.
The 1.4-Nanometer Technical Challenge
Fujitsu’s planned neural processing unit (NPU) targets a process node that does not yet exist in mass production anywhere in the world. The 1.4-nanometer scale represents the next frontier after 2-nanometer technology, featuring gate-all-around (GAA) transistor architectures that promise significant gains in energy efficiency and computational density. While TSMC and Samsung are racing toward 2-nanometer production, Fujitsu and Rapidus aim to leapfrog to 1.4-nanometers for AI inference tasks specifically.
The distinction between training and inference chips proves crucial here. Graphics processing units (GPUs) from Nvidia dominate AI training, the process of teaching large language models to recognize patterns. However, inference, the actual deployment of trained models to make predictions or decisions, requires different architectures. Neural processing units handle these calculations more efficiently than general-purpose GPUs, particularly for edge computing and data center applications where power consumption and latency matter. NPUs are optimized for the matrix multiplication operations common in neural networks while minimizing the memory bandwidth and power overhead that GPUs require for their flexibility.
Fujitsu intends to package these NPUs alongside its Arm-based Monaka-X central processing units in a single integrated device. The Monaka-X represents an evolution of the company’s current 144-core Armv9 data center chip, fabricated using TSMC’s 2-nanometer process. By 2029, Fujitsu hopes to transition the Monaka-X to Rapidus’s 1.4-nanometer production line, combining both CPU and NPU functions for what it calls Fugaku NEXT, the successor to Japan’s current flagship supercomputer. This integration reduces data movement between separate chips, cutting energy consumption and increasing processing speed for AI workloads. Fujitsu also aims to embed encryption technology directly into these processors to protect sensitive data during processing, addressing growing concerns about data sovereignty and security in AI applications.
Rapidus and the Domestic Foundry Gamble
Founded in 2022 with backing from eight major Japanese corporations including Toyota, Sony, and NEC, Rapidus carries the burden of reviving Japan’s semiconductor manufacturing capabilities after decades of decline. The company is currently constructing its first fabrication facility, IIM-1, in Chitose, Hokkaido, targeting mass production of 2-nanometer chips in the second half of fiscal 2027. A second factory, planned for construction beginning in 2027, would handle the 1.4-nanometer production for Fujitsu’s AI chips.
Technical milestones suggest cautious optimism. In July 2025, Rapidus announced that prototyping for 2-nanometer GAA chips had begun at the IIM-1 facility. More recently, the company launched a back-end prototype line and opened the Rapidus Chiplet Solutions (RCS) facility for advanced packaging, critical for integrating multiple processor types into single packages. The Analysis Center adjacent to the foundry now provides physical, environmental, and chemical analysis capabilities necessary to refine manufacturing yields.
Chief Executive Atsuyoshi Koike revealed in February that Rapidus is actively discussing potential partnerships with numerous prospective customers targeting AI, robotics, and edge computing applications.
We are in active discussions with more than 60 prospective customers for chips targeting AI, robotics, and edge computing.
While IBM and chip-design startup Tenstorrent have emerged as leading candidates for the initial 2-nanometer capacity, Fujitsu represents the anchor domestic customer that Tokyo hopes will validate the foundry’s capabilities.
Funding the Future
The financial architecture supporting this initiative spans both public and private sectors. Japan’s New Energy and Industrial Technology Development Organization (NEDO), operating under the Ministry of Economy, Trade and Industry, expects to cover approximately two-thirds of the estimated 58 billion yen ($363 million) initial development cost for Fujitsu’s NPU. This subsidy program specifically targets the creation of energy-efficient AI semiconductors that can be manufactured domestically.
The broader government commitment extends far beyond Fujitsu’s specific project. By March 2027, total public and private investment in Rapidus will reach 2.6 trillion yen ($16.3 billion), including the recent additional subsidy of 631.5 billion yen. Japan plans to invest more than 10 trillion yen ($62.8 billion) across chips and AI by fiscal 2030, designating semiconductors as specified critical products for national security. Prime Minister Sanae Takaichi has underlined the need to step up investment in areas critical for national security, particularly as Japan seeks to guard against external shocks and risks such as recent flare-ups in regional tensions.
An external committee inspected Rapidus’s foundry in Hokkaido and signed off on its technological progress before the latest funding approval, suggesting that independent experts see viable pathways to the company’s ambitious production targets.
Fujitsu’s Parallel Partnerships
While the Rapidus collaboration emphasizes technological independence, Fujitsu is not betting exclusively on domestic capabilities. The company maintains a complex web of international partnerships that illustrate the practical challenges of complete semiconductor sovereignty. In October 2025, Fujitsu announced a collaboration with Nvidia to co-develop AI chips for Japanese data centers by 2030, connecting Fujitsu’s CPUs with Nvidia GPUs on shared substrates.
Additionally, Fujitsu is working with AMD on separate AI chip initiatives. The company’s current Monaka CPU relies on TSMC’s 2-nanometer process, demonstrating that even as Japan builds domestic alternatives, cutting-edge technology still flows through Taiwan and other established manufacturing hubs. This dual-track approach allows Fujitsu to maintain competitive positioning while nurturing domestic capabilities that may take years to mature.
The Fugaku NEXT supercomputer project exemplifies this strategy. Scheduled to go online around 2030 with a target of 100-fold performance improvement over the current Fugaku system, the next-generation machine will likely combine various processor technologies. While the Monaka-X CPUs may come from Rapidus’s 1.4-nanometer line, the system will almost certainly incorporate GPUs and other accelerators from international suppliers to achieve its computational goals.
Building a Broader Ecosystem
Fujitsu is not Rapidus’s only customer, though it represents the most significant in terms of technological ambition. Canon has committed to ordering image-processing semiconductors for digital cameras and surveillance devices using Rapidus’s 2-nanometer process, with trial production already underway. IBM Japan is also receiving NEDO support to develop energy-efficient AI chips potentially manufactured by Rapidus, with up to 17.5 billion yen allocated specifically for IBM’s projects compared to 58.5 billion yen for Fujitsu.
This emerging ecosystem extends beyond simple manufacturing contracts. Rapidus has partnered with IBM on the underlying 1.4-nanometer process technology development, leveraging Big Blue’s research capabilities to accelerate the timeline. Synopsys, a leading electronic design automation company, is participating in Canon’s image sensor development, suggesting that Rapidus is building the infrastructure to support complex, multi-party semiconductor projects.
The concentration of activity in Hokkaido carries regional economic implications. The northern island offers advantages including cooler climates that reduce cooling costs for fabrication facilities, though energy costs in resource-poor Japan continue to rise amid global instability. Rapidus is betting that the combination of government support, technical partnerships, and geographic advantages can overcome the structural challenges of manufacturing in Japan.
Competing Against Industry Giants
The timeline Fujitsu and Rapidus face is daunting. TSMC began mass production of 2-nanometer chips in 2025 and maintains relationships with AI powerhouses including Nvidia and Apple. Intel has already commenced mass production at its 18A node (roughly equivalent to 1.8-nanometers), unveiling Panther Lake processors built on the advanced process. Rapidus targets 2-nanometer production only in late 2027, followed by 1.4-nanometers around 2029.
This gap matters because semiconductor manufacturing exhibits strong network effects. TSMC’s early lead in advanced packaging, high-volume production experience, and established design ecosystems create barriers that new entrants struggle to overcome. Yield rates, the percentage of functional chips produced from each silicon wafer, often determine economic viability. Rapidus’s new Analysis Center and diagnostic capabilities suggest awareness of these challenges, but achieving competitive yields at 1.4-nanometers remains unproven.
Energy costs present additional complications. Semiconductor fabrication requires enormous electricity and water inputs. Japan’s resource constraints and rising material costs amid Middle East instability could disadvantage Rapidus compared to competitors operating in regions with cheaper energy or established supply chains. The government’s willingness to subsidize both development and production costs reflects recognition that market forces alone cannot establish domestic manufacturing capacity.
The Bottom Line
- Fujitsu is developing a 1.4-nanometer neural processing unit for AI inference, targeting production by Rapidus in Hokkaido by decade’s end.
- Japan has committed $16.3 billion in total funding to Rapidus by March 2027, including a recent $4 billion additional subsidy specifically intended to support work for Fujitsu.
- The NPU will integrate with Fujitsu’s Monaka-X CPUs for data center and supercomputing applications, including the Fugaku NEXT system planned for 2030.
- Rapidus aims to begin 2-nanometer mass production in late 2027, with 1.4-nanometer manufacturing starting around 2029 at a second fabrication facility.
- NEDO, Japan’s technology development organization, is funding roughly two-thirds of the $363 million initial development cost for the AI chip project.
- Fujitsu maintains parallel partnerships with Nvidia and AMD, balancing domestic independence efforts with immediate access to cutting-edge international technology.
- Canon and IBM Japan represent additional customers for Rapidus, indicating early steps toward a domestic semiconductor ecosystem.
