34
posted ago by bubble_bursts ago by bubble_bursts +34 / -0

The term "Data Centre" conjures up the image of racks and racks of storage filled with data. Its normal for anyone who think of this to wonder "Why do we need to store so much data?" and the answer to the worried mind comes quickly "Oh my God, its the surveillance state (as if we didnt live in one already for decades) and they will store all my data and thats why they need thousands of data centres"

It just clicked to me while talking to a friend about data centres and I realised that its probably the most unfortunate misnomer, because data centers are not about data at all. Sure they will have some modest data storage for accessing working data set, but data centers are really about Computational Power, specifically for AI work load.

Its basically racks and racks of servers running high power GPUs, and large amounts of VRAMs that will be used to train LLMs ("AI" if you like) and to run inference against these LLMs - when billions of people want to "talk" to AI, but more importantly when billions of people use LLMs as a day to day tool to do all sorts of things.

Not sure if this helps anyone, but I thought I should put this out there after realising many might be operating under wrong assumptions.