IBM CEO says open weight models are now irresistibly good and cheap on-prem
Summary
IBM's CEO Arvind Krishna argues open-weight AI models now provide strong value for on-prem deployments, claiming 25% to 10% of frontier costs while delivering up to 90% of frontier performance. The move signals growing enterprise interest in on-prem AI infrastructure, though running and optimizing such systems requires specialized hardware and kernel/network expertise. The piece also references IBM's z17 mainframes and related accelerator hardware as part of the on-prem strategy.