Source: Hackaday
Generative AI's insatiable demand for training data and inference compute is creating a storage capacity crunch that's making magnetic tape—a 70-year-old technology—economically rational again for cold storage and archival workloads. Tape costs pennies per terabyte annually versus SSDs' exponential price premiums, and LLM training pipelines generate petabytes of data that don't need fast access, only reliable retention. Fujifilm and competitors like Quantum are filling the gap between what cloud providers can economically store on spinning disk and what enterprises can afford to keep online, making tape the hidden infrastructure layer that underpins the AI boom's financial viability.