I'm confused, that doesn't make sense to me: > They largely come from hyperscalers who want hard drives for their AI data centers, for example to store training data on them. What type of training data? LLMs need relatively little of that. For example, DeepSeek-V3 [1], still a relatively large model: > We pre-train DeepSeek-V3 on 14.8 trillion diverse and high-quality tokens At 2 bytes per token, that's 29.6 terabyte…
There are many new data-centers they are being filled with servers. Most servers have at least 2 HDD (mirror) for the OS. I would not be surprised if on a huge scale even 2 HDD per server could cause HDD shortage.
There are likely models which are trained on 4k video and it should be stored somewhere too.
Even things like logs and metrics can consume petabytes for a large (and complex) cluster. And the less mature the software the more logs you need to debug it in production.
In the AI race investments if not unlimited at least abundant. In such conditions optimization of hardware usage is the waste of time and velocity is the only things which matters.