In the late 1980s Japan accounted for about half of the global semiconductor market. Large electronics groups such as NEC, Hitachi and Toshiba were among the biggest chipmakers in the world; in 1986, Japanese companies held the top three places and six of the top ten positions in industry rankings. Much of that strength came from memory, especially DRAM, which Japanese firms produced at high quality and in great volume. [1][2]

Japan's rise caused friction with the United States, whose memory makers were losing share. The 1986 US-Japan Semiconductor Agreement was the most visible response. Among other things it included confidential provisions under which Japan was to raise the share of foreign-made chips in its domestic market to 20 percent, an intervention that later commentators have linked to the weakening of Japan's competitive position. [1]

Japan also contributed foundational inventions. Kioxia, formerly Toshiba's memory business, notes that Toshiba invented NAND flash memory in 1987, the technology behind today's SSDs. [4]

Through the 1990s and 2000s, South Korean memory makers and Taiwanese contract foundries took the lead in volume manufacturing. Japan's strength had been concentrated in DRAM, which left it exposed as demand shifted toward logic and application-specific chips. CSIS estimates that Japan's share of world chip production fell to about 9 percent by 2022. [3][1]

Consolidation followed. The DRAM businesses of NEC, Hitachi and Mitsubishi were combined into Elpida Memory, which was later acquired by Micron. Toshiba's memory unit was sold in 2018 to a consortium that included SK hynix and became Kioxia. By 2024 no Japanese company appeared in Gartner's list of the ten largest semiconductor vendors. [1][5]

Japan's leading-edge logic capability also stalled. CSIS noted that domestic logic production remained at around 40nm, roughly a decade behind TSMC and Samsung, before the current revival began. [3]

While Japan lost ground in chipmaking itself, its suppliers of tools and materials remained central to the whole industry. Brookings, citing industry data, puts Japanese companies at about 88 percent of the market for coater-developers, the machines that apply and develop photoresist, led by Tokyo Electron and Screen Holdings. Japanese firms also hold about 53 percent of silicon wafer supply through Shin-Etsu Chemical and SUMCO, and about half of photoresist supply through companies such as Shin-Etsu, JSR and Tokyo Ohka Kogyo. [2]

CSIS lists even higher shares in some specialised segments tied to extreme ultraviolet (EUV) lithography, including nearly all EUV in-line coater-developers and a large majority of the market for advanced photoresists and wafer-processing equipment. These positions mean that almost every advanced chip, including the GPUs used for AI, passes through Japanese equipment or uses Japanese materials at some stage. [3]

Equipment and materials are a quieter form of leverage than owning a leading-edge fab, but they are hard to replicate. In our view they are the main reason Japan could credibly attempt a manufacturing comeback: the supply base, engineering know-how and customer relationships never left the country.

Japan's government began to treat chips as an economic security issue in the early 2020s. Brookings estimates that Japan committed about 3.9 trillion yen to semiconductor subsidies between 2022 and 2025, equal to around 0.71 percent of GDP, a higher share of national output than Germany, the United States or France. [2]

In November 2024, Prime Minister Shigeru Ishiba announced a framework to provide more than 10 trillion yen in public support for semiconductors and AI by fiscal 2030, which the government said should help draw more than 50 trillion yen of combined public and private investment over ten years. A supplementary budget later that month set aside a further 1.5 trillion yen for chips and AI. [6][7]

The spending is directed at three kinds of projects: bringing foreign manufacturers to Japan, supporting domestic memory makers, and creating a new leading-edge logic company. Each is described below. [2][3]

Japan Advanced Semiconductor Manufacturing, or JASM, is a TSMC-majority company with Sony Semiconductor Solutions, DENSO and Toyota as minority shareholders. Its first fab in Kumamoto Prefecture on the island of Kyushu held an opening ceremony in February 2024, and according to the Taipei Times it entered mass production in late 2024, mainly making 12nm to 28nm chips for automotive and industrial customers. CSIS reports that the Japanese government subsidised the first fab with about 476 billion yen. [8][9][3]

The second Kumamoto fab has changed shape. When it was announced in 2024 it was meant to add 6nm and 7nm production. In February 2026, TSMC chairman C.C. Wei told Prime Minister Sanae Takaichi that the company would instead produce 3nm chips there, a process suited to AI processors. Taiwan's government approved the upgrade in 2026, and the Taipei Times reports the fab is scheduled to come online in 2028 with a capacity of about 15,000 wafers per month. [8][10][9]

Rapidus was founded in August 2022 with backing from major Japanese companies to manufacture leading-edge logic chips. Its first plant, IIM-1 in Chitose, Hokkaido, broke ground in September 2023. Rapidus exposed its first wafers with EUV equipment in April 2025, and in July 2025 announced that it had begun prototyping 2nm gate-all-around transistors that showed the expected electrical characteristics. The process was developed with IBM, and the company has said it plans to begin mass production in 2027. [11][12][13]

The project depends heavily on public money. TrendForce reported that the Ministry of Economy, Trade and Industry approved a further 631.5 billion yen in April 2026, bringing government support to about 2.4 trillion yen. In February 2026 Rapidus raised 267.6 billion yen in equity, including 100 billion yen from the state-backed Information-Technology Promotion Agency and 167.6 billion yen from 32 private companies such as Toyota, Sony, NTT, SoftBank and several banks, and the government added another 150 billion yen in equity in June 2026. [14][13][15]

The main open question is demand. Tom's Hardware reported in 2026 that Rapidus had held talks with more than 60 potential customers but had not yet announced a committed volume customer for its 2nm capacity. Competing with TSMC, which already produces at comparable nodes at very large scale, will require both technical results and customers willing to diversify. [12]

Kioxia listed on the Tokyo Stock Exchange in December 2024. It operates major NAND fabs in Yokkaichi and Kitakami, largely in a joint venture with Sandisk that it says accounts for about 29 percent of global NAND production capacity. In September 2025 Kioxia and Sandisk began operating Fab2 at Kitakami, which makes eighth-generation 218-layer BiCS flash and was partly subsidised by the Japanese government. TrendForce estimated Kioxia's share of NAND revenue at 13.6 percent in the second quarter of 2026. [4][16][17]

Japan is also backing foreign memory makers. Micron, which acquired Elpida, plans to invest about 1.5 trillion yen in a new plant there to produce high-bandwidth memory using EUV lithography, with the Ministry of Economy, Trade and Industry offering up to 500 billion yen in subsidies. TrendForce reported that construction was due to begin in May 2026, with HBM output targeted around 2028. [18][1]

For buyers of AI compute, Japan's revival matters in two ways. Its equipment and materials makers already sit underneath every GPU in the market, and new capacity such as TSMC's 3nm Kumamoto fab and Micron's Hiroshima HBM plant could, from around 2028, add supply of exactly the logic and memory that AI accelerators need. Rapidus is a longer-term and less certain bet. None of these projects will change GPU rental prices this year, but they are part of why the supply picture for 2028 and beyond looks more geographically diverse than today's. You can use Kovara to see which Japanese and Asia-Pacific regions offer GPU capacity today, compare prices, or ask Kova how supply-chain changes might affect the instances you rely on.