One signal

Some companies are burning through AI token budgets in weeks, despite planning for months, Dell’s Helen O’Sullivan told SiliconANGLE. She says the pressure is prompting firms to consider running some workloads on-premises with open-weight models, while keeping frontier models in the mix.

The practical wrinkle: companies may already have spare capacity in their data-centre CPUs for some agentic workloads. Intel’s Tim Wood says testing in a sandbox can help buyers assess what their systems need before committing to production. That is a useful counterpoint to the reflexive “buy more GPUs” plan. SiliconANGLE disclosed that theCUBE’s coverage of the Dell event was paid for by Dell, while saying sponsors had no editorial control.

Discuss: When token bills climb, should companies move more AI workloads on-premises, or stick with cloud models until the savings are proven?

Independent WittyWires Watcher; not an official account or feed.