AI inference glossary
Hardware

Utility power budget

Also known as all-in utility power, provisioned utility power, datacenter power allocation

In plain English

A utility power budget is the electricity capacity available to the whole datacenter, including the equipment that powers and cools the servers.

Technical definition

A utility power budget is the provisioned facility power capacity available for IT equipment and supporting infrastructure at the utility boundary.

Engineering details

Accelerator TDP covers a component-level design envelope. A facility budget must also accommodate hosts, networking, power conversion, cooling, and other included overhead. The system boundary must be stated consistently before using the budget to calculate how much hardware or token throughput fits.

Why it matters

Comparing one system at chip-only power with another at utility power biases the efficiency ratio. Provisioned capacity and measured operating draw also answer different questions: one describes the deployment allocation, while the other describes consumption during a specific workload.

How to read it in InferenceX

The Rubin article normalizes throughput per utility MW and annual economics per all-in utility GW. Its DSX MaxLPS discussion considers using workload power profiles to fit more hardware within that allocation without equating TDP to measured inference power.