Micron goes all in on AI with $9.3B Japan chip plant

Micron Technology broke ground on a new plant to manufacture memory chips in western Japan. This $9.3-billion facility represents an enormous commitment by Micron to grow its ability to deliver semiconductors for AI. The new facility will enable Micron to provide large amounts of high-bandwidth memory (HBM), a key part of training and operating AI models. Production of these HBM chips will not start until approximately summer of 2028.

Micron is building a new factory in Hiroshima to make HBM chips for AI accelerators being developed by NVIDIA’s customers. The facility will be partly financed through contributions of up to ¥500B (~$3.1B) from Japan’s Ministry of Economy, Trade and Industry (METI), as part of a larger Japanese government initiative that aims to promote more domestic semiconductor manufacturing and strengthen supply chains supporting AI.

Micron joins the HBM race

The primary challenge with AI computing is the amount of memory required for it. With each new large language model and artificial intelligence image generator, plus all of the independent AIs, the amount of data being transmitted at high rates between GPUs and memory is massive. One of the solutions to this need has been the introduction of HBM. HBM can be achieved by stacking DRAM dies vertically, increasing bandwidth significantly while also increasing efficiency as compared to traditional memory.

According to a report, the recent increase in demand for artificial intelligence has caused the need for HBM to expand much faster than currently available production capabilities. This prompted the three leading manufacturers of HBM (Micron Technology Inc., SK Hynix Corp. and Samsung Electronics Co., Ltd.) to aggressively expand their production capabilities and accelerate their next-generation product roadmaps.

Micron’s financial results from the third quarter of its 2026 fiscal year demonstrate the extent of the opportunity present. According to its investor relations filing, it had record revenues of $41.46 billion for Q3 of fiscal year 2026 versus revenues of $9.30 billion during the same period a year earlier. Their operating margins for the Cloud Memory Group and the Core Data Center Group were 78% and 83%, respectively. CEO Sanjay Mehrotra stated that this performance demonstrates “the strategic value of memory in the AI era.”

Additionally, the expansion of HBM production is occurring during a time of fierce competition for leadership within this segment of the memory market. Counterpoint Research reported that SK Hynix had approximately 57% of the global HBM market at the end of 2025 while Samsung and Micron each held approximately 22% and 21% of the market, respectively.

As a result of the fierce marketplace competition, not only are companies competing for volume but also on technology roadmaps. Recently, Samsung began to ship samples of the next generation of their HBM4E chips to key customers, and in addition, they also provided roadmaps for future HBM5 products. SK Hynix has also shipped samples of 12-layer, energy-efficient HBM4E memory with greater than 20% energy savings compared to previous generations. Both companies are also preparing HBM products for the next generation of AI systems currently being developed and awaiting end-of-life for the HBM3E-based systems.

Micron doubles down on AI

Capacity expansion is accelerating as well. SK Group Chairman Chey Tae-won stated that SK Hynix plans to double its wafers produced per year to meet the demands for AI memory. In addition, South Korea recently announced an $800 trillion won ($518 billion) public-private project with Samsung and SK Hynix that will establish four new semiconductor manufacturing facilities, expand the packaging capacity of HBM, and ultimately double the amount of DRAM produced in South Korea.

Therefore, Micron’s expansion in Hiroshima represents more than just an increase in capacity. It is a strategic decision to remain competitive as its customers enter into long-term supply contracts with HBM producers years before their production is expected to become operational.

Micron’s project in Japan is aligned with its overall AI strategy. Two days prior to the groundbreaking, Micron signed a strategic agreement with Anthropic that included designing memory architecture, a multi-year supply contract and a large investment in Anthropic’s Series H financing.

In a joint announcement, Anthropic co-founder and Chief Computing Officer Tom Brown said that memory and storage are “central to how efficiently we can train and serve Claude” and that this agreement supports Anthropic’s long-term plan for scaling its computing capabilities.

Japan secures a domestic AI supply link

For Japan, the facility addresses a strategic vulnerability. Although the country remains an important semiconductor manufacturing hub, it has had limited domestic production of the most advanced AI memory chips. Nikkei Asia previously reported that the Hiroshima expansion would provide Japan with local manufacturing capacity for one of the most strategically important components in modern AI infrastructure.

The funding from Japan’s METI highlights an ongoing global competition between countries to attract advanced semiconductor manufacturing. Japan has also supported projects from Rapidus and Taiwan Semiconductor Manufacturing Company (TSMC) as it attempts to develop a well-rounded semiconductor ecosystem that encompasses many facets (such as logic chips, packaging, and high-performance memory) that are all made in-country.

Micron also continues to expand its manufacturing footprint globally. Establishing operations in the United States, Singapore, Taiwan, and Japan to diversify production and better serve AI customers worldwide.