Kioxia's latest earnings show that AI is not just lifting chipmakers that build processors. It is also turning storage and memory into one of the most valuable parts of the hardware stack.
Kioxia's shares were hit with a flood of buy orders after the Japanese flash memory maker posted a stronger-than-expected profit surge, a reminder that the AI buildout is now reaching deep into the semiconductor supply chain. Reuters reported that the company forecast operating profit of 1.3 trillion yen, or about $8.2 billion, for the April to June quarter, far above market expectations, after annual operating profit jumped 92.7% to 870.4 billion yen in the year ended March.
That matters because Kioxia is not a GPU story. It makes NAND flash memory, the kind of storage that sits underneath the AI systems hyperscalers are assembling at speed, and the company is now benefiting from a market that treats memory as a strategic input rather than a commodity. Bloomberg and Reuters both pointed to a powerful demand backdrop from AI infrastructure, with Kioxia also saying it is preparing to list American depositary shares in the US, a sign that management sees the current cycle as strong enough to support a broader investor base.
The easiest way to understand Kioxia's move is to look at how AI data centers actually work. The market spent years focusing on accelerators, especially Nvidia's GPUs, because they were the obvious bottleneck in model training. That is still true, but the next layer down is becoming just as important. Large AI systems need huge amounts of fast storage to move data in and out of compute clusters, feed inference systems, and keep growing libraries of training and retrieval data within reach.
That is why memory vendors are suddenly seeing pricing power. Reuters noted that Kioxia's results were driven by AI boom demand, while Bloomberg described the company as one of the world's best-performing major stocks this year. Other coverage in Japan has been even more explicit about the market shift, with reports that investors now see NAND as the next major beneficiary of the AI trade after processors. The logic is simple. If model builders keep adding data centers, they need more storage, and that demand runs straight into a supply base that cannot expand overnight.
Kioxia's position is especially interesting because NAND flash is not a speculative corner of the market. It is a core component of the infrastructure stack, used in solid-state drives and enterprise storage systems that sit inside the AI data center itself. As Bloomberg reported, Kioxia's outlook reflects a broader memory shortage, not just a one-off earnings beat. In other words, this is less about one good quarter and more about a repricing of where value sits in the AI economy.
Investors are widening the lens
For startups and investors, the message is that the AI hardware opportunity is broader than many of the early trade narratives suggested. The first wave rewarded companies tied directly to model training. The current wave is rewarding the firms that make the infrastructure usable at scale. That includes memory makers, storage vendors, networking companies, and the surrounding tools that help data move efficiently through systems.
There is also a clear capital-markets signal here. Kioxia's plan to pursue a US ADS listing suggests that it wants to tap deeper international demand while its valuation momentum is still strong. That would have been hard to imagine when memory was being treated as a cyclical business with little pricing power. Now, with AI customers building capacity and competing for supply, the market is assigning a more durable growth story to companies that can actually ship the parts the buildout requires.
The larger lesson for the startup ecosystem is not just that AI demand is real. It is that the winners may include companies far from the model layer. Storage has become one of the clearest examples. Kioxia's earnings show that the AI boom is not only creating software winners and GPU winners, it is also creating a new class of infrastructure beneficiaries that sit quietly in the background and still capture a meaningful share of the value.
That is the part investors should not miss. When a memory supplier starts posting record numbers and the market rushes to buy the stock, it usually means the cycle has moved past hype and into physical constraints. AI is now consuming real hardware, at scale, and the companies that feed that demand are starting to look indispensable.
Also read: AI license plate readers are becoming a trust crisis for local governments • Publicis buys LiveRamp for $2.2bn, betting the ad holding company can own the identity layer that powers AI marketing • LinkedIn prompt injection shows how brittle AI recruiting has become
