Aller au contenu
Thursday, August 27, 2026

B-EMPIRE

Culture without borders. / La culture sans frontières.

Kioxia-Sandisk: $31 Billion for the Memory That Feeds AI

Kioxia and Sandisk plan more than $31 billion in investments in Japan through 2032. Behind flash memory factories, the entire industrial chain of artificial intelligence is being redrawn.


Cheventong Vil
Cheventong Vil
August 27, 2026  ·  5 min de lecture
Kioxia-Sandisk : 31 milliards pour la mémoire qui nourrit l'IA
B-EMPIRE Magazine

In the great battle over artificial intelligence, processors attract the spotlight. But the machines that learn, generate, classify and archive also need another raw material: memory. Kioxia and Sandisk announced on Thursday, August 27, 2026, that they plan to invest more than $31 billion in Japan through 2032, contingent on government support. This is not merely an industrial program. It is a sharp reminder that AI is not an abstract cloud; it is a geography of factories, clean rooms, electricity, capital and sovereignty.

The plan concerns the Yokkaichi and Kitakami sites, two names that become more important as global demand for flash memory rises. According to Reuters, a new production facility is to be built at Kitakami in northern Japan, with an announced investment of 1.8 trillion yen, or about $11.3 billion. Kioxia also says site preparation for Fab3 has begun, with operations targeted for fiscal 2029.

Memory, the Blind Spot of the AI Boom

For two years, the dominant AI story has focused on GPUs, accelerators and the valuations of computing giants. That is understandable: those components embody the visible power of the moment. Yet AI infrastructure is not just computation. It must store massive datasets, move colossal volumes of information, preserve outputs, feed models and support increasingly frequent uses.

NAND flash memory is therefore less spectacular than the star chips, but it is indispensable. Without fast, dense and more efficient storage, data centers become more expensive, slower and more fragile. Kioxia and Sandisk operate precisely on that terrain: their 3D memory technologies are meant to support the rise of generative AI, on-device AI and so-called physical AI, the AI of robots, vehicles, sensors and automated factories.

A Twenty-Five-Year Partnership Scales Up

The announcement does not come from nowhere. Kioxia and Sandisk note that they have already invested more than $50 billion in Japan over the past twenty-five years. Their joint venture, extended in January through December 2034 at the Yokkaichi site, is one of the historical pillars of flash memory. The new plan adds a strategic layer: it turns an old industrial partnership into an organized response to the demand cycle opened by AI.

The timing is revealing. In early July, the two companies announced the start of production of their tenth-generation 3D flash memory at Kitakami Fab2. A few weeks later, they placed Fab3 in the next trajectory. The industry is no longer only talking about optimizing an existing factory; it is preparing extra capacity for a decade in which data will be treated as a resource of power.

Japan Wants to Become a Stronghold Again

For Tokyo, the operation reaches beyond Kioxia. Japan long symbolized global electronics power before losing ground to South Korea, Taiwan, China and the United States. The return of industrial policy, pushed by trade tensions and dependence on supply chains, gives the archipelago a new chance to reposition itself.

The fact that the investment is conditioned on public support is essential. Semiconductors are no longer an ordinary industry: they are part of economic security. Subsidies, taxation, land, energy and diplomacy have become tools in a competition where every country wants to secure a share of capacity. Kioxia and Sandisk also present their plan as aligned with the Japanese government’s economic goals and as a symbol of cooperation between Japan and the United States.

A Global Race, Not an Isolated Announcement

This program arrives in a broader sequence. Memory manufacturers are raising their ambitions because data centers are demanding more numerous, higher-performing and more energy-efficient components. Reuters notes that the growth of AI has contributed to Kioxia’s turnaround, after the company had long been seen as a symbol of Japan’s struggles in chips. The company is now manufacturing tenth-generation BiCS Flash memory at Kitakami, developed with Sandisk.

The message sent to the market is clear: the AI boom is not playing out only among processor designers. It is also playing out among those who can guarantee memory, storage and supply stability. In this context, a factory is no longer just a production unit. It becomes a strategic insurance policy against shortages, export restrictions, demand shocks and geopolitical rivalries.

The True Technological Luxury: Producing for the Long Term

There is a business lesson in this announcement. The technological luxury of 2026 is not only having the most advanced chip or the most talked-about software. It is being able to produce for a long time, regularly, at scale, with enough capital to absorb cycles. Flash memory is a brutal market: prices can fall, inventories can swell, investments are immense and returns take years.

That is precisely why a commitment through 2032 matters. It tells customers that capacity will be there. It tells investors that Kioxia and Sandisk want to defend their place against Korean, Chinese and American competitors. It tells governments that advanced manufacturing remains a lever of power. And it tells AI players that the next constraint will not be computation alone, but the full hardware ecosystem that makes it usable.

What B-EMPIRE Takes Away

Kioxia and Sandisk are not only financing production lines. They are financing the hardware backstage of AI. Where markets like visible names, this announcement reminds us that technology empires are also built in less glamorous layers: memory, cooling, energy, chemistry, machine tools, engineers and industrial territories.

The $31 billion figure measures this shift. AI began as a race for models; it is becoming a race for infrastructure. The winners will not only be those who write the best algorithms. They will also be those who control the material, factories and financial patience required to turn artificial intelligence into a durable industry.

Sources

Vous êtes hors ligne. Voici les derniers articles disponibles.