The grid that
talks to compute.
The work matters. So does when it happens. Move an illustrative workload and see what the grid gains—and what it still needs.
ILLUSTRATIVE MODEL / ALL VALUES ARE AUTHORED ASSUMPTIONS
Assumes this work can wait two hours and finish within window 6.
105 MWPeak combined demand
120 MWhAI work energy, unchanged
5 MWhDemand above the assumed supply limit
The fourth window exceeds the assumed supply limit. Move some work and inspect both the original peak and the later rebound.
Inspect every number and assumption
| Window | Other demand | AI load | Total |
|---|---|---|---|
| 1 | 40 | 20 | 60 |
| 2 | 55 | 20 | 75 |
| 3 | 75 | 20 | 95 |
| 4 | 85 | 20 | 105 |
| 5 | 65 | 20 | 85 |
| 6 | 40 | 20 | 60 |
Fixed 100 MW supply; six one-hour windows; no network, cooling, ramping, overhead, service-level, emissions or price model. Deferral assumes a feasible deadline. Not an Aurora or Emerald AI performance measurement.
Where the real work begins.
Emerald AI and NVIDIA describe power-flexible AI infrastructure. This model explains one timing mechanism, using Bright-authored assumptions. It does not reproduce Conductor or measure Aurora.
Read the source claim, capacity estimate and limits ↗