IBM CEO Arvind Krishna says open-weight models are 90% as good as frontier models at 10 to 25% of the token cost, and that two or three racks are enough. I checked it: on the Artificial Analysis index the best downloadable model reaches 80% of the leader. What on-prem really costs (rent vs. buy with list prices), when cloud APIs still win, and the gpu-unlock mod that lifts a tensor speed limit on RTX 30/40/20 cards, with its risks.
IBM’s CEO Says Run AI Yourself. Open-Weight Models Are Close, but On-Prem Is a Volume Bet
