Fujitsu will begin exporting AI semiconductors built on technology from its Fugaku supercomputer to the United States and Asia as early as next year, Nikkei reported on September 13. It is the first time the Japanese company has sold its advanced processors on external markets at scale, and it puts a domestic Japanese chip design directly into competition with American and Taiwanese data center silicon.
The chip at the center of the push is Monaka, a 2-nanometer Arm processor Fujitsu originally developed for the FugakuNEXT supercomputer, the successor to Fugaku, which held the top spot on the TOP500 supercomputer ranking for consecutive years and still sits ninth as of June 2026. Unlike the older A64FX chip that powered Fugaku and was reserved for internal use, Monaka is designed as an open commercial product, built as a 3D chiplet with 144 cores per socket and integrated AI acceleration through a fused neural processing unit.
According to Nikkei, more than 40 global companies are in talks to adopt the platform. Twenty-four firms in the financial and telecommunications sectors have already tested early systems. The next phase targets major cloud providers in the United States, where Fujitsu hopes to secure large volume contracts. The company laid the groundwork for this rollout in 2024 through a development partnership with Supermicro, which builds servers around the chip for enterprise customers.
Production runs through TSMC
Fujitsu does not fabricate the chips itself. Monaka is produced by TSMC on its 2nm process, which means the export strategy depends on the same Taiwanese foundry that supplies Apple, Nvidia and AMD. Fujitsu has also announced a separate project to develop a 1.4nm AI processor fabricated domestically through Rapidus, the Japanese government-backed foundry venture, though that chip is earlier in development and targets a later generation.
The split approach reflects Japan’s broader semiconductor policy. Export controls have made access to advanced chips a national security question, and Tokyo wants domestic capability for strategic workloads while Japanese companies still sell into global markets. A domestic-only path would limit Fujitsu to the home market; the TSMC route lets it compete for data center business abroad while the Rapidus partnership matures. Japanese officials have framed the Rapidus project as a matter of economic security, and Fujitsu’s 1.4nm design is one of its flagship commitments.
The sales target
Fujitsu is aiming for 250 billion yen, roughly $1.63 billion, in cumulative processor sales by fiscal year 2030. Reaching that figure requires shipping around 60,000 dedicated AI server units into foreign data centers. For context, Nvidia’s data center revenue runs at tens of billions of dollars per quarter, so Fujitsu is not chasing the same market. Its pitch is energy efficiency and sovereignty: the company claims up to twice the performance per watt of competitor CPUs and markets Monaka for customers who want AI compute outside the Nvidia ecosystem, including government agencies and regulated industries.
The sovereign computing angle has grown more relevant as export controls tighten. Fujitsu’s own marketing material for the chip emphasizes control over data, confidential computing features and an open AI and HPC ecosystem, positioning it for buyers in Asia and the Middle East who cannot or will not buy American accelerators for every workload. The chip also carries Arm’s latest vector and matrix extensions, SVE2 and SME2, which matter for the inference workloads that dominate enterprise AI deployment.
Why it matters outside Japan
Monaka enters a market where CPU competition has intensified as AI workloads spread beyond GPUs. Amazon’s Graviton, Google’s Axion and Microsoft’s Cobalt all push Arm-based server chips into cloud fleets, and Nvidia now sells its own Arm CPU, Grace, alongside its GPUs. Fujitsu’s difference is lineage: its supercomputer credentials are real, and the SVE vector extensions it pioneered with A64FX carried into the Arm ecosystem that all these chips now share. Fugaku itself was the first system to top all four major supercomputer rankings at once, a feat no machine had managed before it.
Whether enterprise buyers outside Japan will pick a Fujitsu CPU over established alternatives is the open question. The company has a strong position in Japanese government and financial systems and years of experience selling the PRIMEHPC line internationally, but at modest volumes. Selling 60,000 AI servers abroad by 2030 would be a meaningful result for a company that has never competed at this scale internationally, yet it is a modest number against the volumes American hyperscalers purchase in a single quarter.
Analysts cited in the Japanese coverage framed the export decision as a test of whether Japan’s semiconductor revival can extend beyond domestic procurement. The government and companies are, in the words of one Korean summary of the report, waging an all-out battle to secure future growth engines, and chips are the most visible front. Fujitsu’s decision to sell abroad rather than keep Monaka internal signals confidence that the product can win commercial customers, not just state-funded ones.
Timing also matters for the wider region. Malaysia is reportedly close to selecting Huawei’s Ascend 910C accelerators for a sovereign AI project, which would make it the first government to officially pick Chinese silicon over American alternatives. A Japanese option that avoids both Washington’s export rules and China’s sanctions exposure gives buyers in that position one more choice, and Fujitsu is clearly counting on it.
The first shipments are expected next year, with production shipments beginning in 2027 according to Fujitsu’s own roadmap. If the export push works, it gives Japan a direct supplier role in data center silicon for the first time in decades. If it stalls, the Rapidus domestic project becomes the more important bet.