It seems like with the rapid scaling of data centers, HBM memory bandwidth and inference demand -- we're creating an unprecedented scenario where compute will centralize itself at data centers as the cost of essential parts for personal compute keep rising.
That would assume the AI bubble never pops, which considering the circular funding arrangements, and the level of capital being invested for relatively modest returns seems unlikely.
