IBM CEO: Open-weight AI now 90% as good, 10% the cost
Original: IBM CEO says open weight models are now good and cheap on-prem
Why This Matters
Enterprise AI spend is shifting; IBM is betting on on-prem open-weight infrastructure
IBM CEO Arvind Krishna told Q3 earnings investors that open-weight AI models now deliver ~90% of frontier model performance at 10-25% of token cost, making on-prem enterprise deployment increasingly attractive.
On IBM's Q3 2026 earnings call, CEO Arvind Krishna made a direct case for open-weight models in enterprise AI. In 2024, he noted, the gap between closed frontier models and open alternatives was stark — top closed models scoring around 70 on broad benchmarks, open models landing at 20-30. By late 2025, that gap had narrowed to roughly 70 versus 50. Now, he says, open-weight models are "90 percent as effective" at "anywhere from 10 to 25 percent of the token cost."
Krishna framed it bluntly: "Would you like something that's 10 percent of the cost if it's 90 percent as good as the best, or do you want to pay that 10 times higher cost?" For most enterprise workloads — summarizing emails, reviewing contracts, handling customer calls — he argues the cheaper option is sufficient.
IBM is seeing surging interest in on-prem AI infrastructure as a result. Krishna says useful setups can fit in two or three racks, but enterprises face a real skills gap: finding kernel engineers, managing GPU network proximity, and handling inference caching. IBM's answer is software that abstracts that complexity — details remain thin for now.