← Analytics / Capacity planningWhen does capacity
When does capacity
go live?
Follow equipment through suppliers, integration, installation, and commissioning. Change the plan. See what holds it back.
INTERACTIVE CASE STUDY · SYNTHETIC DATATrace the full route from supply to service. Fixed program: 96 racks across two campuses: 64 AI and 32 general compute.
Commissioned by W2680 / 96 racks
0 vs baseline
IT capacity online · W264.70 MW
6.21 MW facility demand at modeled PUE
Due-date backlog · W2616 racks
16 total racks still to complete
Commitments met by W2683%
80 of 96 due by target week
Full program completionWeek 31
0 weeks vs baseline
Illustrative budget change$0
$53,137,120 equipment, 26-week labor & expansion
FOUR ECHELONS / ONE CONNECTED PLAN
From equipment to ready for service.
Week 18 snapshot
Select a node to trace its constraints.
01 Equipment supply
Compute equipment
Network equipment
Rack / power / cooling kits
02 Integration hubs
A complete compute, network, and infrastructure kit is required for every rack.
03 Site deployment
Two-week hub-to-site transport. Fixed routes: Pacific → West; Central → East.
04 Ready for service
Equipment delivery is a milestone. Only commissioned racks count as live capacity.
SELECTED NODE / WEEK 18
West campus
- Incoming supply
- From Pacific integration after test and transport
- Available capacity
- 3.24 MW IT / 2.88 MW thermal / 60 positions
- Queue / work in process
- 7 racks at installation / commissioning
- Downstream commitment
- 48 racks → West campus
- Current waiting cause
- commissioning hours · 1 racks
- Ready for service
- 23 racks / 1.16 MW IT / 1.51 MW facility demand
16 racks remain outside service at week 26.
The largest accumulated waiting cause in weeks 1–26 is West campus · Incomplete equipment kit, first recorded in week 1. At the selected week, East campus · commissioning hours is holding 7 racks. Change the relevant resource, then compare the completion curve to see whether the overall program improves.
Waiting counts describe this schedule; they are not proof that one intervention alone will recover the plan. 10 / 96 racks meet their individual due dates within the 78-week projection.
Deployment trajectory
Cumulative racks · weeks 1–26Committed demandBaseline commissionedScenario commissioned
MULTIPLE SOURCES
Test the supply response.
Allocate rack-equivalent kits between two sources per family. Qualification and delivery times still apply.
Network equipment
Primary A · 5 kits/week · 2 weeks transit
Alternate B · 3 kits/week · 4 weeks transit
Rack / power / cooling kits
Primary A · 4 kits/week · 4 weeks transit
Alternate B · 2 kits/week · 5 weeks transit
Allocation is applied to whole kits, so actual percentages can differ slightly. Disabling B assigns new scenario orders to A; the simulation recalculates from week 1. Existing opening inventory and scheduled receipts remain available.
START WITH THREE QUESTIONS
A short walkthrough.
- Can we get the equipment? Choose Supplier delay, then shift compute allocation toward B. Check how qualification and transit affect kit availability.
- Can our teams process it? Choose Commissioning shortage. Open Work Centers and restore hours. Follow the change in ready-for-service dates.
- Can the site accept it? Choose Cooling expansion delay. In Site Capacity, bring expansion forward. Extra supply cannot overcome missing cooling.
READ THE TERMS
The essentials.
- Echelon
- A distinct layer in the deployment network, with its own inventory, capacity, and lead time.
- Work center
- A team or process with finite productive hours, such as rack assembly or commissioning.
- Commissioning
- The final modeled verification step before a rack is counted as ready for service.
- MW versus MWh
- MW is power capacity. MWh is energy consumed over time; this model does not calculate energy consumption.
- PUE
- Power usage effectiveness: total facility power divided by IT power, used here as a fixed planning estimate.
- Reserved headroom
- Power and cooling withheld from ordinary deployment. It is not a certification of N+1 or 2N resilience.
Model assumptions, calculations, and reference sources
An illustrative deployment program
All organizations, rates, prices, and dates are synthetic. This case study demonstrates planning logic, not an actual customer deployment or a claim of personal data center delivery experience.
| Profile | IT kW | Cooling kW thermal | Assembly h | Test h | Install h | Commission h |
|---|---|---|---|---|---|---|
| General compute | 12 | 12 | 6 | 4 | 5 | 5 |
| High-density AI | 80 | 80 | 12 | 8 | 10 | 10 |
- Each rack consumes one profile-specific kit from each of the three equipment families. These are rack-equivalent bundles, not interchangeable individual servers or components. Each hub starts with two kits per profile and family; one additional kit per profile and family arrives in week 3 from pre-existing orders.
- All remaining program orders are released from week 1. Source shares assign integer kits using cumulative rounding within each destination and profile. A repeating AI/West, AI/East, general/West, general/East sequence dispatches up to each supplier’s weekly limit after qualification. All allocated demand is fulfilled eventually if capacity permits; unused allocation does not automatically spill to another source.
- Within each work center, oldest-due work receives hours first, with rack ID as the tie-breaker. A blocked rack can be bypassed by another eligible rack. Incomplete work carries forward. Kits become eligible for assembly the next week; each subsequent work center starts no earlier than the next week. Test release is followed by two weeks of transport and next-week installation.
- Commitments start in week 8 and increase by three racks per campus per week. An acceleration shifts due dates forward; target week changes the reporting checkpoint only. There is no workload scheduling, cross-campus failover, or automatic optimization.
- Usable power or cooling = (base capacity + expansion available that week) × (1 − reserved headroom). Installation requires positions, both capacities, network readiness, and—for AI—liquid cooling readiness. Reservations include racks still installing and commissioning, preventing double counting.
- Commissioned IT MW = sum of completed rack kW ÷ 1,000. Facility electrical MW = commissioned IT MW × site PUE. Cooling is modeled at rack IT heat load, without a detailed thermal, hydraulic, electrical-distribution, or redundancy simulation.
- Backlog counts due racks not yet commissioned. Target attainment counts commissioned racks due by the chosen target divided by all racks due then. Full completion projects the fixed program through week 78, assuming the configured capacities persist.
- Budget includes new equipment orders for the entire program, 26 weeks of staffed work-center hours, and specified power/cooling expansions regardless of readiness date. Source B costs 12% more. Labor is $95/hour, or $140/hour for commissioning; expansion is $450,000 per IT MW plus $350,000 per cooling MW. Opening stock and scheduled receipts are treated as sunk cost. Excludes base facilities, energy, freight, taxes, and contingency; it is not a project estimate.
| Family | General compute | AI |
|---|---|---|
| Compute equipment | $90,000 | $650,000 |
| Network equipment | $12,000 | $60,000 |
| Rack / power / cooling kits | $18,000 | $90,000 |
Infrastructure concepts: NVIDIA — Planning a Data Center Deployment ↗. The source explains the interaction of power, cooling, space, networking, and delivery coordination; the numerical assumptions above are independent synthetic examples.