Used in: NBIS analysis, AI infrastructure, Week 20
What It Is
GPU utilization rate is the percentage of a data center’s graphics processing units that are actively running workloads at any given time. For an AI infrastructure company like Nebius, it is the primary measure of whether the capacity being built is actually being used by paying customers.
Why It Matters
A data center full of GPUs that are idle is not generating revenue. High utilization rates (above 85 percent) indicate that demand is real and the infrastructure is productive. Low or declining utilization rates indicate either that the company built more capacity than customers need or that customers are not staying on the platform. During an AI infrastructure shortage, utilization is naturally high — but if it drops as more supply comes online, that tells you whether customers were choosing the platform or just using what was available.
Where to Find It
In operational metrics disclosures within earnings releases and investor presentations. Not all AI infrastructure companies disclose this number. If a company stops disclosing utilization rates, or never discloses them, that is worth noting. Disclosure of specific utilization metrics signals confidence; vague directional language signals the opposite.
Real Example
Nebius Group, supply vs. demand. The core question for NBIS is whether 684 percent revenue growth reflects customers choosing Nebius or customers using Nebius because GPU supply was short everywhere. High, sustained utilization above 85 percent with long contract durations would support the first interpretation. Declining utilization as more hyperscaler capacity comes online would support the second.