IBM CEO: Open-weight models are good enough and cheap enough that on-premises enterprise deployment is hard to resist

IBM CEO Arvind Krishna said on the Q3 earnings call that open-weight models now achieve about 90% of the performance of frontier models, at only 10% to 25% of the token cost, and that on-premises enterprise deployment requires just two or three racks. He believes that the vast majority of tasks—such as summarizing emails, reviewing contracts, and answering customer service inquiries—can be handled with cheaper solutions, and predicted that demand for local AI will drive sales of IBM's Spyre accelerator cards.