Dell's O'Sullivan: token budgets blown in weeks as agentic AI grows

The conversation points to a gap between how enterprises plan and how agentic AI actually consumes resources. Token budgets that looked adequate for months are being exhausted in weeks, according to Dell's Helen O'Sullivan, as agentic workloads multiply. That mismatch is pushing on-premises infrastructure and open-source models back into the deployment conversation alongside frontier cloud models.

O'Sullivan and Intel's Tim Wood describe a hardware mix in flux. Agentic AI spans orchestration, inference and edge computing, and O'Sullivan says her team anticipated heavier CPU utilization months ago, even though many buyers still fixate on GPUs. Wood suggests enterprises may already have unused CPU capacity in their data centers capable of handling some next-generation workloads, and that sandbox testing of open-weight models can help assess what those workloads require.

Both companies are taking a consultative approach, weighing existing infrastructure and capital versus operating expense models before recommending a plan. O'Sullivan frames the decision as broader than chip selection, noting that customers often ask how fast they can get a GPU, when the real question is about the data.