If you've tried to buy a hard drive lately—or if you're an enterprise trying to expand your storage infrastructure—I have bad news. Hard drives are completely sold out for 2026. And the culprit, once again, is AI.We've spent the past two years watching AI companies snap up every available GPU, creating shortages that affected everyone from gamers to researchers. Now they've moved down the stack to storage. The data requirements for training large language models and storing inference results are so massive that AI companies are buying up the world's entire hard drive production capacity.This isn't about SSDs or consumer storage. This is about the high-capacity enterprise drives that data centers depend on—the 20TB, 24TB drives that store the world's data. And right now, if you want one, you're out of luck unless you ordered months ago.Here's what's happening. Training a single frontier AI model can generate petabytes of checkpoint data. Every iteration, every training run, companies want to save it all in case they need to roll back or analyze what went wrong. Then there's the inference side. Every ChatGPT conversation, every AI-generated image, every code completion—all of it gets logged, analyzed, and stored for improving future models.The downstream effects are starting to hit. Small cloud providers can't expand capacity. Research institutions can't upgrade their infrastructure. Even large enterprises are facing delays on storage buildouts. When AI companies buy up the world's storage capacity, everyone else waits.The storage manufacturers are ramping up production, but it takes time to build new factories and scale supply chains. Meanwhile, AI companies are pre-ordering entire production runs, sometimes a year in advance. It's the GPU shortage all over again, except this time it's affecting an even more fundamental piece of computing infrastructure.This is AI's hidden infrastructure cost playing out in real time. We talk about the electricity consumption, the water usage for cooling, the GPU shortage. But nobody warned us about the storage crunch. Because who could have predicted that the bottleneck for AI progress wouldn't be compute or algorithms, but hard drives?The question now is what happens next. Do we see hard drive prices spike like we did with GPUs? Do storage manufacturers build new fabs fast enough to meet demand? Or do AI companies start deleting their old checkpoints because even they can't store everything?One thing's certain: the era of cheap, abundant storage is over. At least until 2027.
|
