SK Hynix Weighs a Japan Memory Fab as It Circles Closer to Kioxia
SK Hynix is studying potential sites for a memory chip plant in Japan, and the search ties directly to a company few outsiders expected it to court: Kioxia, the Japanese NAND flash maker it already partly owns. Chairman Chey Tae-won flagged the plan in recent days, telling Bloomberg on Monday that the company is looking at locations across Japan and confirming to Japan's Asahi Shimbun that joint production with Kioxia is "one option."
Chey's wording was careful. He listed supply chain coordination and joint research as equally plausible areas of cooperation, and pushed back on the idea that a factory is settled. "We are looking all over Japan," he told Bloomberg. "Anywhere with good power and good water." A formal decision, the company says, has not been made.
The timing is telling. SK Hynix broke ground last week on an advanced memory packaging plant in West Lafayette, Indiana, its first US production site. Now it is scouting a second overseas footprint, and Japan has emerged as one of the strongest candidates. Big AI buyers like Meta and Amazon are absorbing every gigabyte of memory the industry can produce, and SK Hynix executives have warned that supply will tighten further rather than ease.
Chey was blunter this week, describing a looming shortfall in AI memory as the industry's central problem. In an interview with CNBC he said the race to secure high-bandwidth memory is creating a structural imbalance that could hit next year. That anxiety explains why a company that spent the last two years pouring money into Korean fabs and an American plant is now weighing yet another site.
A Site Offer Already on the Table
Japan's prefectures are competing to host the plant before SK Hynix has even named a location. Miyagi Prefecture, on the island of Honshu north of Tokyo, has proposed a parcel of about 300,000 square meters in the Second Sendai Northern Core Industrial Park in Ohira Village. Miyagi Governor Yoshihiro Murai described the site at a press conference on August 26 as more than 30 hectares with access to industrial water and electricity, and close enough to central Sendai to draw workers. "There are few places in the world as good as this," he said.
Miyagi already hosts a Tokyo Electron plasma etching development and production base, along with facilities run by chip-related companies such as LAPIS and Sony. The industrial ecosystem matters. A memory fab needs a steady supply of specialized equipment and materials, and Japan has both in unusual depth. The Kanto, Hokkaido, and Kyushu regions have also been floated as possible homes for the plant.
For SK Group, a Japanese production base fits a broader political project Chey has championed for months. He has called repeatedly for a Korea-Japan economic bloc worth around $6 trillion, proposed linking power grids, gas pipelines, and even an undersea tunnel between the countries, and suggested combining Korea's memory strength with Japan's edge in materials, components, and equipment. A chip plant would be the most concrete expression of that idea to date.
Why Kioxia Is the Complicated Part
The partnership question is where the story gets delicate. SK Hynix competes head-to-head with Kioxia in NAND flash, the storage memory that sits alongside HBM in every AI server. But it is also one of the Japanese company's biggest shareholders, holding roughly a 14 percent stake through a Bain Capital-led special purpose company that was formed when a consortium bought Toshiba's memory business in 2018. In August, that SPC became Kioxia's largest shareholder.
SK Hynix's securities convert into shares carrying most of the SPC's voting rights, giving it influence that could expand if a deeper deal is struck. An existing agreement, however, stops SK Hynix from holding 15 percent or more of Kioxia's voting rights without its consent until 2028. Chey has made the alternative clear: if a manufacturing partnership fails to materialize, SK could simply exit its investment.
Deeper cooperation between the two would reshape the global NAND market. Counterpoint Research estimated that Samsung led NAND shipments in the second quarter of 2026 with a 25 percent share, followed by SK Hynix at 22 percent and Kioxia at 14 percent. Together, the latter two control more than Samsung, but they still compete as separate suppliers. A closer alliance would hand them real collective scale.
There is also the technology angle. Both companies are pushing high-bandwidth flash, or HBF, an emerging format that stacks NAND chips in a structure similar to HBM to boost capacity and speed. SK Hynix is working with US-based SanDisk on global HBF standards, and Kioxia's long-running production partnership with SanDisk in Japan could give SK Hynix an added strategic position in developing the format. Lee Jong-hwan, a professor of system semiconductor engineering at Sangmyung University, said SK Hynix appears ready to use the profits and confidence from its HBM leadership to pursue the top spot in the broader memory market. He sees Kioxia as a partner that could strengthen its NAND position and help it challenge Samsung.
The Memory Crunch Behind Every Decision
Put the Japan search next to the numbers and it starts to make sense. SK Hynix has ridden high-bandwidth memory to record earnings, becoming the most valuable part of the SK Group conglomerate. Its products feed Nvidia's accelerators and the data centers that Meta, Amazon, and Microsoft keep expanding. Demand has been so strong that the company repeatedly raised guidance through 2026 and watched its Nasdaq-listed shares climb after a listing that raised billions of dollars.
The shortage Chey describes is not a gentle tightening. AI systems consume memory in two ways: the high-bandwidth memory stacked next to each accelerator, and the NAND storage that holds training sets, checkpoints, and model weights. Both have strained supply. When inference demand grows faster than training, the need for high-capacity storage climbs with it, widening the NAND gap.
That is why the industry is working on high-bandwidth flash. HBF stacks NAND dies the way HBM stacks DRAM, boosting capacity and speed for AI workloads that hit storage limits. SK Hynix is collaborating with SanDisk on standards for the format, and the tie between SanDisk and Kioxia in Japan creates a natural meeting point. If the three companies coordinate, a single regional ecosystem around Sendai could develop HBF and the advanced packaging that goes with it.
Kioxia's own position makes the timing delicate. The Japanese company has struggled to keep pace with Samsung's scale in NAND and faces pressure from Chinese suppliers moving up the stack. A partnership with SK Hynix would give it a powerful ally, but it would also surrender some independence. For SK Hynix, the appeal is straightforward: NAND at scale, in a country with cheap power and deep materials expertise, without building an entire supply chain from scratch.
What the Japanese Government Adds
Tokyo has its own reasons to welcome SK Hynix. Japan has spent the past few years rebuilding its semiconductor base after ceding ground to Taiwan and South Korea. It has offered generous subsidies to draw TSMC to a new fab in Kumamoto, and the strategy has started to pay off as the Taiwanese giant expands there. Memory is the gap in that effort. Japan grows the materials and builds the equipment that others use, but it produces little of its own advanced memory today beyond Kioxia's NAND lines.
A SK Hynix plant would change that. It would bring high-bandwidth memory production to Japanese soil for the first time, anchor a supply chain that stretches from TEL's etching tools to Namics' materials, and give Tokyo a stake in the fastest-growing segment of the chip market. Japanese local governments have been courting the company aggressively, and the national government has signaled it is open to subsidizing the project the way it backed TSMC.
The politics of a Korea-Japan chip partnership would be notable in their own right. The two countries have spent decades as rivals in memory, exporting competing NAND and DRAM to the same customers. A joint venture would mark a genuine break from that pattern. Chey has framed it as a step toward the $6 trillion economic bloc he wants to build, and Korean-Japanese business groups have talked up closer ties for months. The plant, if it happens, would be the proof those talks were real.
None of this is certain. Chey has emphasized that discussions remain open and no final decision has been made. The company has said the same in its own statements, insisting nothing has been decided about a joint venture, specific products, or a particular location. But the direction of travel is clear, and every indicator from the memory market is pushing toward more capacity rather than less. Whether the answer lands in Japan, in partnership with Kioxia, or somewhere entirely unexpected will test just how far the industry's biggest players are willing to cooperate.
The Obstacles Before Any Shovel Hits the Ground
A Japan plant faces hurdles that go well beyond finding a site. The company has a huge domestic buildout already underway: roughly 600 trillion won for the Yongin semiconductor cluster, about 100 trillion won at Cheongju, and some 400 trillion won for a new base in southwestern Korea. Adding a Japanese facility on top of that is a question of capital allocation at the group level, not just engineering.
There is also regulatory scrutiny at home. Many advanced semiconductor technologies are designated as national core technologies under Korea's Industrial Technology Protection Act, which requires government approval or notification before they are transferred to overseas production sites. A plant building advanced products such as HBM for AI would face the most sensitive review, since it involves the newest processes.
The United States is in the mix as well. Washington has pushed SK Hynix to add local investment, and the company has left open the possibility of a front-end memory production facility in America beyond its Indiana packaging plant. Any one of these tracks could consume the same billions in capital.
For now, SK Hynix is positioning itself to grow wherever demand is strongest. Chey's remarks this week signal that Japan is a live candidate, Kioxia is being courted rather than written off, and the memory shortage he keeps warning about is fueling the urgency. Whether that produces a Japanese fab, a Kioxia venture, or something else entirely should become clearer in the coming months.
Chey summed up the stakes in his Asahi interview: "If SK Hynix is to be included in Kioxia's future strategy, we are always ready to become a partner." The question now is whether Kioxia's strategy includes SK Hynix at all, and how far either side is willing to go to find out. Investors and competitors will be watching both the rollout of semiconductors coverage and the broader SK Group announcements, which Chey has said will come once the talks are complete.
Sources: The Korea Herald, KED Global, The Elec, and Bloomberg via Yahoo Finance.
