Cost estimating and cost control

Cost is the language in which owners, lenders, contractors and boards discuss a project. An estimate that is too low leads to a project approved on false premises and later overruns; one that is too high may kill a viable project or invite padded bids. Once the project is under way, cost control tracks spending against budget, forecasts the final cost, and manages contingency and changes so that problems surface early enough to act. This lesson covers estimate classes and accuracy, the structure of a data center estimate, benchmarks and their limits, escalation and cash flow, value engineering and lifecycle cost, cost control systems, and earned value management.

Estimate classes

Estimates are prepared repeatedly as the project develops, each more detailed and accurate than the last. The AACE International classification, widely used in North America, defines five classes according to the maturity of the project definition:

Two lessons follow. First, an early estimate is a range, not a number, and decisions should be made on the range: a Class 5 estimate of 400 million dollars might mean anything from 250 to 700 million. Second, accuracy ranges are asymmetric, with more upside than downside, because early estimates tend to miss scope rather than include too much. Funding decisions, such as final investment approval, are usually made on Class 3 estimates.

Structure of an estimate

A complete project estimate includes far more than the construction contract:

• Direct costs: labor, materials, equipment and subcontracts for each part of the work, organized by WBS or a standard classification such as MasterFormat or UniFormat.
• Indirect costs: the contractor's general conditions (site management, temporary facilities, cranes, security, insurance, bonds) and its overhead and profit (fee).
• Design and professional fees: architects, engineers, commissioning agents, owner's engineers, surveys, testing and legal.
• Owner costs: land, utility connection and substation contributions, permits and development charges, owner-furnished equipment, owner's staff, financing during construction, and start-up costs.
• Escalation: the expected increase in prices between the estimate date and the time of purchase.
• Contingency: an allowance for risks and unknowns within the defined scope (Lesson 5).

Utility contributions can be large: a project requiring a new substation or transmission upgrade may pay tens of millions of dollars, sometimes more than the building shell. Omitting owner costs is one of the commonest reasons early estimates fall short.

Benchmarks and parametric estimates

Early estimates rely on parametric relationships: cost per megawatt of critical IT load, cost per square meter of white space, or cost per rack. In the mid-2020s, North American data center construction cost (shell, core and MEP, excluding land and IT equipment) has often been cited in the range of about 8 to 15 million dollars per MW, with wide variation. Within that, electrical systems typically account for the largest share (often 40 % or more), mechanical systems another substantial share (often 15 % to 25 %), with the building, site and other trades making up the rest. Benchmarks must be adjusted for:

• Location: labor rates and productivity, climate (which affects cooling), seismic design and taxes.
• Resilience: Tier III and Tier IV designs with more redundancy cost more per MW.
• Density and cooling: high-density AI designs with liquid cooling shift costs from air handling to piping, coolant distribution and more electrical capacity per square meter.
• Scale and repetition: large campuses and repeat designs benefit from economies of scale and learning.
• Time: escalation since the benchmark was recorded.

Benchmarks are a starting point for screening and for checking detailed estimates, not a substitute for them.

Escalation and cash flow

Escalation is applied from the estimate date to the midpoint of each cost item's expenditure, using forecasts of construction cost indices or specific indices for major commodities such as copper and steel. In periods of rapid price change, escalation can be several percent per year and is a major estimate component. Contracts can manage escalation risk with price adjustment clauses tied to published indices, early purchase of materials, or fixed-price supply agreements.

The cash flow forecast spreads the estimate over time, usually following the schedule's S-curve (Lesson 1), to plan financing and to set the planned value baseline for earned value. Owners also forecast cash flows for deposits on long-lead equipment, which may require substantial payments years before installation.

Value engineering and lifecycle cost

Value engineering (VE) is a structured review of the design to identify alternatives that deliver the required functions at lower total cost, ideally carried out during schematic design, when changes are cheap. Done well, it examines functions and lifecycle costs; done badly, late in the project, it becomes cost cutting that reduces quality or reliability and generates redesign. For data centers, operating costs, especially energy, are large: a 40 MW facility at a PUE of 1.3 uses about 450,000 MWh a year (40 × 1.3 × 8,760), costing tens of millions of dollars annually. A design change that raises capital cost but reduces PUE or maintenance may have a lower lifecycle cost, assessed using net present value over the facility's life (DCE-460). VE proposals should therefore be judged on lifecycle cost, reliability and schedule, not only on first cost.

Cost control

Cost control compares what was planned with what is committed, spent and forecast. A cost report for each WBS element typically shows:

• Original budget and approved changes, giving the current budget.
• Commitments: contracts and purchase orders awarded.
• Actual costs: invoices received and paid.
• Pending changes and potential claims, not yet approved.
• Estimate to complete and estimate at completion (forecast final cost).
• Variance between current budget and forecast.

The most important number is the forecast at completion, updated every month. Projects get into trouble when forecasts are not updated honestly, for example when known pending changes are left out because they have not yet been approved. Contingency drawdown should be tracked against the risk register: if contingency is being consumed faster than risks are being retired, the forecast is optimistic.

Earned value management

Earned value management (EVM) integrates scope, schedule and cost to measure performance objectively. Three quantities are tracked:

• Planned value (PV): the budgeted cost of the work scheduled to be done by now.
• Earned value (EV): the budgeted cost of the work actually done by now.
• Actual cost (AC): what that work actually cost.

From these, cost and schedule performance are measured:

A CPI below 1 means work is costing more than budgeted; an SPI below 1 means less work has been done than planned. The estimate at completion (EAC) projects the final cost if current efficiency continues, and the to-complete performance index (TCPI) shows the efficiency needed on the remaining work to finish within budget.

# Earned value at month 12 of a 24-month, 440 million dollar project
BAC = 440.0       # budget at completion, million dollars
PV = 120.0        # planned value: budgeted cost of work scheduled to date
EV = 104.0        # earned value: budgeted cost of work actually performed
AC = 116.0        # actual cost of the work performed
CV, SV = EV - AC, EV - PV
CPI, SPI = EV / AC, EV / PV
EAC = BAC / CPI                     # if current cost efficiency continues
EAC_mixed = AC + (BAC - EV) / (CPI * SPI)
TCPI = (BAC - EV) / (BAC - AC)      # efficiency needed on remaining work to finish on budget
print(f"cost variance {CV:+.1f} M, schedule variance {SV:+.1f} M")
print(f"CPI {CPI:.3f}, SPI {SPI:.3f}")
print(f"EAC at current CPI {EAC:.1f} M (overrun {EAC - BAC:.1f} M)")
print(f"EAC with CPI x SPI {EAC_mixed:.1f} M")
print(f"TCPI to finish on budget {TCPI:.3f}")
print(f"approximate schedule slip {24 / SPI - 24:.1f} months if SPI persists")

cost variance -12.0 M, schedule variance -16.0 M
CPI 0.897, SPI 0.867
EAC at current CPI 490.8 M (overrun 50.8 M)
EAC with CPI x SPI 548.4 M
TCPI to finish on budget 1.037
approximate schedule slip 3.7 months if SPI persists

Halfway through, the project has earned 104 million dollars of work but spent 116 million (CPI about 0.90) and is behind the plan (SPI about 0.87). If cost efficiency continues, the final cost will be about 491 million, a 51 million overrun; if schedule pressure adds further costs, as the combined formula assumes, it could approach 550 million. Finishing on budget would require the remaining work to be performed at 1.04, better than anything achieved so far, which is unlikely. Research on large projects shows that CPI rarely improves much after a project is 20 % complete, so early EVM signals deserve action: re-planning, scope review, a revised budget and contingency, rather than hope. SPI in cost terms becomes unreliable late in a project, because EV converges to the budget as work completes; schedule analysis (Lesson 6) remains essential.

Cost control on programs and repeat projects

Owners delivering many similar facilities gain a powerful cost control tool: their own historical data. Recording final costs for each project in a consistent structure (by system, per MW and per square meter, with notes on scope, location and market conditions) allows each new estimate to be checked against actual outcomes rather than published benchmarks, and shows which systems vary most between projects. Programs also allow unit-rate agreements with contractors and suppliers, where repeated items such as power skids or cooling modules are priced per unit with indexed escalation, reducing estimating effort and negotiation time. The risk is complacency: a standard estimate copied from the last project can miss site-specific costs such as rock, utility contributions or local labor shortages, so every estimate should still be reviewed against the specific site and schedule.