Amid the AI Investment Boom, Is Japan's Semiconductor Industry Overheated—or Still Expanding?

Share prices have been falling sharply even as companies announce new factory investment. On July 17, shares in Japanese flash-memory maker Kioxia fell 16.1% in a single day as semiconductor stocks sold off globally; rising tensions involving the United States and Iran, together with U.S. interest-rate expectations, further weakened risk appetite.[1] On July 28, Kioxia fell another 18.3% as investors shifted their focus to whether massive spending on AI infrastructure could generate adequate returns and whether capacity expansion and technological progress among Chinese memory producers could intensify future competition.[2] The market is therefore reassessing a basic question: has the AI semiconductor cycle already overheated, or is the industry still in an expansion phase?

A share-price correction first reflects a change in expectations; it does not necessarily mean underlying demand has weakened at the same pace. Kioxia's July 31 results showed operating profit of about ¥1.27 trillion for the quarter ended in June, with the company projecting roughly ¥1.89 trillion for July–September and explicitly stating that data-center demand was expected to remain strong.[3] At least from the perspective of orders and earnings expectations, AI servers and data centers are still supporting demand for flash memory and SSDs.

More revealing is how companies are actually allocating production capacity. On August 27, Kioxia announced preparations to build a third manufacturing building, K3, at its Kitakami plant in Iwate Prefecture. The facility is targeted to begin operations during fiscal 2029 and is intended to expand production capacity for advanced three-dimensional BiCS FLASH memory. Kioxia cited the growth of Agentic AI, Physical AI and On-device AI among the drivers of medium- to long-term demand.[4] On the same day, Kioxia and SanDisk also announced plans—subject to government support—to invest a cumulative roughly ¥5 trillion in Japan through 2032 in production equipment, infrastructure and technology upgrades at their Yokkaichi and Kitakami operations.[5]

There is no contradiction between falling share prices and continued capital investment. Equity markets are asking whether valuations have already priced in too much growth and whether enormous AI capital spending will ultimately generate sufficient profits. Semiconductor manufacturers, by contrast, have to prepare capacity years ahead of demand. Memory fabs take a long time to plan, build, equip and ramp into volume production; waiting until demand is fully confirmed can leave capacity coming on stream too late. Large-scale investment therefore signals continued confidence in medium- and long-term demand, while also increasing the risk of excess capacity and depreciation pressure if demand undershoots expectations.

As of early September, sentiment in Japanese manufacturing was still receiving support from semiconductor demand. The Reuters Tankan showed sentiment among large manufacturers near a five-year high, with the electronics-sector index rising from +24 in August to +39. Companies frequently cited improving demand related to AI, data centers and semiconductors.[6] If that expansion continues, the benefits will not be confined to GPUs or memory. Advanced packaging materials, electronic chemicals, ceramics and insulating materials, thermal-management materials and high-reliability interconnect materials could all see stronger demand as wafer, packaging and data-center investment expands.

Whether Japan's semiconductor industry has become “overheated” therefore cannot be judged simply by how far related shares have fallen from their highs. The more useful indicators are whether data-center construction, actual memory shipments, capacity utilization and capital investment continue to grow together. For now, the picture looks more like capital markets cooling their expectations for AI valuations while the underlying industry remains in expansion. What will determine how far this cycle can run is not how much higher the AI theme can push share prices, but whether massive investment in computing capacity converts into sustained real demand. For Japan's semiconductor industry, the next question may be less “should we keep expanding?” and more “will the market still be able to absorb the capacity being built today several years from now?”