Active facility MW
300 MW
60% of target facility load
Calculator 06Power and capacity bottlenecks
Data Center Power Bottleneck Calculator
Convert announced facility MW into realistic active IT MW by testing the full power chain: credible grid supply, onsite generation, substations, cooling readiness, rack fit-out, silicon availability, and time-to-power. The output is what can actually turn into AI capacity.
Active IT MW
261 MW
39.1 MW facility overhead at 1.15 effective PUE
GPUs online
126K
83.7K GPU-equivalent gap
Binding bottleneck
Credible power supply
200 MW facility MW not yet usable
Online timing
56 mo.
Grid interconnection is the longest pre-handoff workstream
Effective power cost
$86.9/MWh
$14.9/MWh capacity adder
Constraint stack
Which layer limits usable AI capacity?
Credible power supply
Binding constraint
300 MW
Firm transformer capacity
Available capacity
450 MW
Cooling-ready capacity
Available capacity
345 MW
Rack fit-out capacity
Available capacity
322 MW
Available silicon capacity
Available capacity
382 MW
Target facility load
Available capacity
500 MW
Power bridge
Queued grid request500 MW
Credible queued MW300 MW
Already energized0.0 MW
Onsite generation0.0 MW
Firm transformers450 MW
Time to power
Partner delivery, grid work, and GPU turn-up
Grid interconnection
54 mo.
Critical path
HV transformers
18 mo.
Parallel workstream
Shell and MEP build
24 mo.
Parallel workstream
Cooling readiness
20 mo.
Parallel workstream
Schedule bridge
Data center hand-off
54 mo.
GPU install / test
2 mo.
Service slip
32 mo.
Workstreams run in parallel until the slowest one hands off the site. GPU installation then adds the turn-up and testing phase.
Delay cost
Monthly fixed carry$54.3M
Slip carrying cost$1.74B
Target online24 mo.
Power cost stack
Energy, capacity charges, facility carry, and capex
| Metric | Value | Unit | Read-through |
|---|---|---|---|
| Annual energy | $174M | $/yr | 2.42M MWh at 92% load factor |
| Annual capacity charge | $36.0M | $/yr | $329/MW-day applied to active facility MW |
| Annual facility / colo cost | $391M | $/yr | $125.00 / kW-mo on active IT load |
| Grid upgrade capex | $200M | one-time | $400.0K / queued MW |
| Total project capex | $4.20B | one-time | $8.0M / target facility MW plus grid upgrades |
| Power cost per GPU | $1.7K | $/GPU-yr | Energy plus capacity charges only |
Sensitivity analysis
Grid credibility x onsite generation
| Queue cred. \ onsite MW | 0.0 MW base | 0.0 MW base | 0.0 MW base | 50.0 MW | 100 MW |
|---|---|---|---|---|---|
| 40% | 174 MW | 174 MW | 174 MW | 200 MW | 226 MW |
| 50% | 217 MW | 217 MW | 217 MW | 239 MW | 261 MW |
| 60% base | 261 MW | 261 MW | 261 MW | 278 MW | 280 MW |
| 70% | 280 MW | 280 MW | 280 MW | 280 MW | 280 MW |
| 80% | 280 MW | 280 MW | 280 MW | 280 MW | 280 MW |
This grid holds transformers, cooling, rack fit-out, silicon, and schedule fixed while testing the most important siting question: how much of the queued grid request is real, and how much onsite power can bridge the gap?
Preset comparison
Same framework across grid, retrofit, BTM, and neocloud cases
| Preset | Market | Target MW | Active IT MW | Online | Bottleneck |
|---|---|---|---|---|---|
| PJM greenfield | PJM | 500 MW | 261 MW | 56 mo. | Credible power supply |
| ERCOT attested | ERCOT | 500 MW | 285 MW | 32 mo. | Credible power supply |
| Energized retrofit | Remote US grid | 200 MW | 75.0 MW | 12 mo. | Available silicon capacity |
| BTM overflow | US onsite generation | 300 MW | 183 MW | 17 mo. | Available silicon capacity |
| CRWV ramp | Multi-region neocloud | 3.10 GW | 1.56 GW | 20 mo. | Firm transformer capacity |
| GB200 liquid | AI-ready hyperscale campus | 150 MW | 110 MW | 20 mo. | Rack fit-out capacity |