OpenAI says fake accounts from China tried to turn Americans against data centers

[reACTIONary]

38. FYI ....

.... the necessity is not primarily data storage, its computing capacity.

To perform AI training / execution requires a huge parallel processing array of very high end graphics processing units (GPUs). It turns out that the sort of mathematics required for AI and graphic rendering are very much the same - parallel processing of large floating point matrices.

You need a lot of them, and they gobble up a lot of electricity and generate a lot of heat.

Crypto used to rely on on GPUs, but used the parallel processing capability in a different way. Crypto migrated from GPUs to specialized ASICs (application specific integrated circuits). However, these are just as energy hungry and also have to be organized into dense computing arrays.

GPUs are good for both AI and graphics, but AI specific ASICs will probably be developed. AI ASICs have the potential to decrease the energy consumption of AI focused data centers.