The artificial intelligence buildout has captured the public imagination through its most visible component: the GPU. Yet in recent months, a quieter realization has spread through investors and industry analysts alike. AI is not just a story about graphics processors. It is a story about an entire infrastructure stack, and the less glamorous components of that stack — memory chips, solid-state storage, and even old-fashioned spinning hard drives — are suddenly enjoying a moment that would have been unthinkable just a few years ago.
A Reawakening for Storage and Memory
The recent earnings reports from Sandisk and Western Digital illustrate the magnitude of this shift. Western Digital posted revenue growth of 45 percent year over year, surpassing $3.3 billion. Sandisk, meanwhile, reached quarterly revenue highs above $8.25 billion, with profit comfortably exceeding expectations. These are not the kinds of numbers one typically associates with mature, cyclical hardware businesses. They reflect the way capital is now flowing into AI-related infrastructure, sweeping along components that were once considered commodities.
What makes the storage angle particularly interesting is the corporate history behind these two firms. Western Digital and Sandisk first merged and have since separated, an arc that highlights how the industry has been searching for the right structural configuration to capture this new wave of demand. The split allows each company to lean into a different part of the AI stack: NAND flash memory on one side, large-capacity hard drives on the other. Both turn out to be essential.
Reading the Market's Hesitation
Despite the strong numbers, the market's reaction has been somewhat muted, and the reasons are instructive. Sandisk has managed to push higher, while Western Digital appears to be running into a ceiling. Part of the explanation lies in expectations. Seagate, Western Digital's closest peer, reported earlier in the week with a particularly strong quarter, raising the bar that Western Digital then had to clear. Western Digital's gross margin guide was not quite as strong as some had hoped after Seagate's blowout, which created a sense of comparative disappointment even though the absolute numbers were excellent.
A similar dynamic shaped the response to Sandisk. The company had been more aggressive in raising prices than peers, and Samsung had posted phenomenal pricing numbers in its first quarter. When Sandisk's guide implied somewhat lighter price increases than expected, investors flinched — even though the company has built a foundation of longer-term contracts that may help smooth its earnings trajectory and elongate the upcycle. The deeper question hovering over the entire space is whether these results represent a peak. The answer hinges on whether this cycle behaves like the violent, short-lived booms of the past.
The Supply Constraint at the Heart of the Story
The most compelling reason to believe this cycle has further to run is the simple arithmetic of supply. Historically, what ends memory upcycles is the arrival of new production capacity. Too many bits chase too few buyers, prices collapse, and the industry enters a downturn. That mechanism is not currently in operation. New memory fabrications are not expected to come online until the second half of 2027. In the meantime, the industry has limited tools to add more bits to the market.
The hard drive side of the equation faces an even tighter bottleneck. Manufacturing the recording heads that go into hard drives requires specialized fabs, and no one has announced plans to build a new head fab. Without new head capacity, the industry simply cannot scale output to match the surging demand from data centers building for AI workloads.
This is why the runway for the current upcycle could extend further than many investors instinctively expect. Demand is outpacing supply, and the supply side cannot respond on a short timeline. A turn in the cycle is unlikely in 2026, probably not in 2027, and may not arrive until 2028 — pushing the eventual oversupply event closer to the end of the decade.
Risks: The Black Swans and the Gray Ones
Of course, no projection of this kind is risk-free. The most acute risks come from demand-side shocks rather than supply expansion. When the conflict involving Iran broke out, memory stocks briefly wobbled on fears that higher oil prices and slower economic growth could compress AI spending. That scenario did not materialize, but it illustrates the kind of black swan event that could derail the thesis: a macroeconomic disruption serious enough to cause demand destruction.
There is also a more subtle risk worth flagging. The industry is currently betting on what amounts to a static model of AI infrastructure — assuming that the ratios of compute to memory to storage observed today will persist. But as inference workloads grow alongside training, and as the field develops a clearer empirical understanding of what AI deployment actually requires, those ratios could shift. A future in which AI demands less infrastructure per unit of capability than is currently assumed would be a kind of "black swan" event of an unusual shade — not a crisis, but a quiet reassessment of just how much memory and high-speed storage are truly needed. The Western Digital side of the trade may be more insulated from this risk, given the seemingly insatiable appetite for cheap, dense, capacity-oriented storage.
A Differentiated Outlook
The components that power AI now extend well beyond the GPUs that dominate headlines. Bulk storage is the unglamorous but essential foundation underneath model training data, inference logs, and the explosion of generated content that AI systems both consume and produce. Flash memory and DRAM, meanwhile, sit closer to the compute and benefit from any expansion in real-time processing.
Within this landscape, valuations are reflecting outperformance that has already occurred but may not yet capture the duration of the cycle. A twelve-month price target of $1,200 on Sandisk, for instance, with an outperform rating, is grounded less in heroic assumptions about further price increases and more in the simple recognition that prices will continue rising as long as supply is constrained. The debate is over how much higher pricing climbs, not whether it climbs at all.
Conclusion
The story of the storage and memory industry right now is, in a sense, a microcosm of the broader AI investment thesis. Demand is real, broad, and durable. Supply takes years to add. The most likely path forward is a sustained period of strong pricing and strong margins, punctuated by occasional spasms of investor anxiety whenever a peer reports slightly differently or a geopolitical event injects volatility. Barring a genuine demand shock or a structural rethinking of AI infrastructure requirements, the upcycle that began with the AI boom looks set to extend through 2027 and possibly beyond. For an industry historically defined by sharp peaks and brutal troughs, that is a remarkably long runway — and one that is being underwritten not by hype, but by the unglamorous physics of building new fabs.