Nov 26 (News On Japan) - Nvidia maintains an estimated 80% share of the global AI chip market, but a surge of new competitors is beginning to challenge the company’s dominance, reshaping expectations across the semiconductor industry.
Google recently opened access to its most advanced AI processor, Ironwood, marking a significant expansion of its cloud strategy. After months of internal testing, the chip is now available as part of Google’s Tensor Processing Unit lineup, which CEO Sundar Pichai has identified as a key driver of cloud-related growth. Analysts note, however, that Google’s tightly controlled ecosystem may limit broader market impact unless access is expanded further.
Amazon and Microsoft are also advancing their in-house semiconductor efforts. Amazon continues to roll out its Trainium chips, while Microsoft is developing the Maya processor, designed specifically for high-precision mathematical operations central to AI training and scientific computing. Both initiatives remain largely confined to each company’s cloud infrastructure.
One of the more unusual developments is emerging from China, where researchers claim progress in analog computing—a technology whose foundations date back to ancient Greece. Analog processors have historically struggled with scalability and precision, leading digital chips to dominate. Chinese research institutes now say new designs could enable analog AI chips up to 1,000 times faster than Nvidia’s current offerings. While the claims remain unverified, experts say analog components could complement digital processors in tasks requiring extreme energy efficiency and speed.
Yet some of the boldest ambitions are coming from Japan, which is attempting a full-scale reentry into advanced chip manufacturing after decades of decline. In the late 1980s, Japanese firms produced more than half of the world’s semiconductors. Today, that share has fallen below 10%, the result of strategic missteps tied to the integrated device manufacturer model, while competitors such as TSMC scaled rapidly through specialization. By the 2010s, Japan’s most advanced domestic production had slipped to 40-nanometer technology, far behind Taiwan and South Korea.
Japan’s renewed push began during the recent global supply chain crisis, when chip shortages halted production of automobiles, electronics, and even bank cards. The government responded with a dual strategy: attracting foreign leaders such as TSMC to build fabs domestically, while fostering a new national champion capable of producing cutting-edge nodes.
Central to this plan is Rapidus, a public-private venture established three years ago. Backed by eight major Japanese corporations—including Toyota, Sony, NTT, and SoftBank—the company aims to mass-produce 2-nanometer chips by 2027. The government has pledged more than US$6.5 billion in subsidies, with total investments expected to reach around $35 billion.
Rapidus began construction of its semiconductor plant in Hokkaido in September 2023 and launched a pilot production line in April 2025. The facility includes more than 200 pieces of advanced fabrication equipment, including EUV lithography systems costing more than $300 million each. The company is collaborating closely with IBM, drawing on nanosheet transistor research at IBM’s Albany facility, where more than 100 Rapidus engineers are currently based.
Unlike conventional fabs that process wafers in large batches, Rapidus plans to adopt a single-wafer manufacturing method, allowing each wafer to move independently through production tools. While the approach may reduce throughput, it enables fast turnaround and customization—advantages for AI startups, research laboratories, and specialized industries requiring rapid iteration.
The company also intends to integrate AI-driven monitoring systems with dense sensor networks, feeding real-time data into both design and process optimization. The goal is to accelerate the path to high yields, though the method remains untested at industrial scale.
The challenges ahead are substantial. Funding gaps remain large, with government subsidies covering only a fraction of projected needs, requiring significant private investment. Japan’s semiconductor talent pool has thinned since the 1990s, creating fierce competition for experienced engineers. Technologically, Rapidus faces a narrow window to scale from pilot production to mass output, a transition that has challenged even established manufacturers. By 2027, when Rapidus aims to reach 2-nanometer production, both TSMC and Samsung are expected to have already spent years at that node and may be moving toward 1.4-nanometer technology.
Despite these hurdles, Japan’s reentry into advanced semiconductor manufacturing is viewed by many industry analysts as one of the most ambitious national initiatives in the sector. As global competition intensifies and nations seek supply-chain resilience, Japan’s push has the potential to add a significant new axis to the rapidly evolving AI hardware landscape.
Source: Fahd Mirza
