What this indicator measures
This indicator approximates the amount of GPU compute time consumed by AI-related workloads.
Calculation method
The estimate translates AI activity signals into equivalent GPU-hours using hardware and workload assumptions.
Simplified formula
GPU-hours ≈ AI workload volume × estimated GPU time per workload unitData sources
Sources may include public AI usage indicators, benchmark assumptions, GPU specifications, model training reports and infrastructure research.
Published calculation inputs
The values below are the explicit inputs currently used by this indicator. They are assumptions or public proxies, not official totals.
- Baseline
- 100M GPU-hours/d
- Model assumption
- Conversion factor
- 1.5 kWh/GPU-hour
- Model assumption
Direct references
Last reviewed: September 27, 2026
Known limitations
Actual GPU time varies widely depending on model architecture, hardware generation, batching, optimization and utilization rate.
Refresh and interpolation
The counter is periodically synchronized server-side and interpolated live between updates.