Lesson 10 / 12 · Aggregate AI power demand
Interrupt a 50 MW AI cluster
How long can 5 MWh of stored energy bridge a 50 MW load?
Predict before you run
Write down your prediction, then change only the input below. In the interactive lesson, compare the starting case with the challenge and export a worksheet to retain your reasoning.
| Input | Starting value | Challenge value |
|---|---|---|
| Initial battery | 5000 kWh | 2500 kWh |
Reported quantity: s elapsed. The live control accepts 0–5000 kWh.
Worked answer
Scaling the electrical ratings and storage by 50 preserves the 307.8 s ride-through duration. Halving initial reserve gives depletion at 478.2675 s with pre-outage charging. This is aggregate demand, not GPU job or grid adequacy simulation.
Reproduce the configuration
| IT demand / simulated duration | 50000 kW / 1800 s |
|---|---|
| Opening battery / energy capacity | 5000 kWh / 5000 kWh |
| Charging power limit | 5000 kW |
| Charge / discharge efficiency | 0.95 / 0.9 |
| Distribution efficiency | 0.95 |
| Generator start delay | 30 s |
Scheduled events
- 300 s: utility — asset down
- 300 s: generator — asset down
- 900 s: utility — asset up
- 900 s: generator — asset up
The browser lesson applies its configuration to the ai_cluster_generator_failure preset. Open it above, run the starting case, switch to the challenge, and use “Verify against Python” to compare complete results. Use the worksheet export to keep the prediction, completed inputs and answer together.
Sources and verification
- Exact lesson definitions and input transformations
- Native Python default/challenge checks
- 1 MW equations and independent energy-ledger reconstruction
- Electrical continuity contract
Original material by Mohammad Rezwan Khan, engine 1.0.0. Synthetic teaching cases; no facility calibration or independent external reproduction has been established. Annual PUE planning and outage continuity are different calculations. Full scope and evidence.