The artificial intelligence hardware trade, which for years has been a straightforward bet on GPU manufacturers, is undergoing a foundational shift. As the initial frenzy of building out AI training models matures, the market is waking up to a new, more persistent bottleneck: data storage.
The data access demands of enterprise-level AI inference are forcing a massive capital expenditure rotation from compute to memory, catching many off guard and igniting a supercycle in the NAND flash market. Heavy AI workloads are fundamentally re-architecting the data center, and legacy storage infrastructure is proving inadequate. This has triggered an imbalance between supply and demand, with global NAND fabrication capacity unable to keep pace. For investors, the narrative has pivoted. The primary beneficiaries are no longer just the companies that train the models, but the ones that feed them.
