# The grid that talks to compute.

Move a flexible workload in time and inspect demand, deferred work and the rebound. A transparent illustrative model.

Canonical: https://brightaifuture.com/experiences/compute-meets-grid
Format: experience
Source publication: Not established
Bright publication: 2026-09-09
Substantive update: None recorded
Evidence and review: Bright-authored assumptions and arithmetic, separate from company claims.

## The model

Six one-hour windows. The illustrative AI workload uses 20 MW in each window (120 MWh total). Move up to 20 MWh from window four to window six. Grid supply is held at 100 MW. All values are authored assumptions, not Aurora measurements.

## What changes

Moving work can relieve one peak but creates demand later. A sustained high-demand scenario shows why flexibility alone cannot solve a supply shortage.

## What is excluded

No network topology, ramp constraints, cooling dynamics, compute overhead, emissions, prices or service reliability are modeled. The model assumes all shifted work can meet its deadline in window six.

## Limitations

An illustrative energy-balance model, not a production controller, Aurora operating trace, grid-capacity estimate or reliability validation.

## Provenance and history

{}

## Original sources

- [NVIDIA: DSX grid-flexibility blueprint](https://blogs.nvidia.com/blog/omniverse-dsx-blueprint/)

## Continue exploring

- [Emerald AI / DSX evidence record](https://brightaifuture.com/discoveries/emerald-aurora)
- [Energy world](https://brightaifuture.com/worlds/energy)