Dell and Intel: enterprises burn token budgets, shift more inference on-prem
Speaking with theCUBE at Dell's “AI Leadership Symposium,” Dell's Helen O'Sullivan and Intel's Tim Wood said growing agentic workloads are causing some customers to burn through token budgets allotted for months in a matter of weeks. That cost pressure, they said, is pushing enterprises to rethink where inference runs, keeping frontier models in the cloud while using open-weight models and on-premises infrastructure to absorb part of the load. They also described Intel's agentic AI architecture as spanning orchestration, inference and edge computing, and advised enterprises to tap existing CPU headroom in their data centers and test open-weight models in sandbox environments before production.
If token costs keep outstripping budgets, enterprise AI deployments may be redistributed between cloud and on-premises, drawing more attention to CPU capacity and open-weight models.