Datacenter Build Cost

Full lifecycle cost analysis: CapEx, OpEx, depreciation, growth, and build-vs-colo comparison.

View full document →

Should you build a small dedicated datacenter or rent space in someone else’s? This walkthrough models the full lifecycle cost using CalcMark, showcasing exchange rates, growth functions, depreciation, and napkin math along the way.

The complete CalcMark file is available at testdata/examples/datacenter-cost.cm.

Front Matter #

Every CalcMark document can start with YAML front matter. Here we define currency exchange rates and a global variable for facility size that’s available throughout the document.

---
exchange:
  USD_EUR: 0.92
  EUR_USD: 1.09
  USD_GBP: 0.79
  GBP_USD: 1.27
globals:
  sqft: 1000
---

CalcMark features: exchange front matter for currency conversion rates (FROM_TO: rate format); globals front matter for document-wide variables.

Facility Sizing & Baseline CapEx #

Construction costs for dedicated datacenters typically range from $625 to $1,135 per gross square foot. We use a midpoint estimate for a 1,000 sqft facility:

cost_sqft = $875
capex = @globals.sqft * cost_sqft

@globals.sqft references the sqft variable from the front matter globals block. The @globals. prefix makes it clear the value comes from frontmatter, not from a local definition. Multiplying a number by a currency preserves the currency unit: 1000 * $875 = $875,000.

CalcMark features: @globals.name directive references; currency arithmetic.

CapEx Breakdown by Component #

The budget is heavily weighted toward electrical and mechanical systems, not the building shell. The X% of value syntax calculates a percentage directly:

elec = 42.5% of capex
hvac = 17.5% of capex
fitout = 22.5% of capex
shell = 17.5% of capex

Sanity check -- components should sum to total:

components = elec + hvac + fitout + shell

The percentages (42.5 + 17.5 + 22.5 + 17.5 = 100%) mirror industry benchmarks: electrical systems dominate at 40-45%, followed by building fit-out, HVAC/cooling, and the land/shell.

CalcMark features: X% of value natural language syntax; currency addition.

Tier Comparison #

Uptime Institute tiers define redundancy levels. Tier II (partial redundancy) and Tier III (concurrently maintainable) differ dramatically in cost:

tier_ii = $2300 * @globals.sqft
tier_iii = $7700 * @globals.sqft
tier_mult = 7700 / 2300

Moving to Tier III more than triples the cost per square foot. For AI-optimized facilities requiring specialized liquid cooling, costs can reach $20M+ per MW.

CalcMark features: Currency multiplication; plain division for ratios.

Power Infrastructure #

Datacenter construction is often quoted per megawatt (MW) of commissioned IT load. A small 1,000 sqft facility might support 200 kW of IT load:

it_load = 200 kilowatts

A standard 1 MW facility costs $7M-$12M. Our load as a fraction of 1 MW:

load_frac = 200 / 1000
mw_cost = $9.5M
power = load_frac * mw_cost

For reference, 1000 kilowatts converts cleanly:

1000 kilowatts in megawatts

CalcMark knows SI power units and can convert between them using the in keyword. The $9.5M syntax uses a multiplier suffix – CalcMark supports k (thousand), M (million), and B (billion).

CalcMark features: Quantities with units (kilowatts); in unit conversion; multiplier suffixes ($9.5M); markdown prose between calculations.

International Equipment Pricing #

Specialized cooling equipment from a European vendor is quoted in euros. The exchange rates defined in the front matter make conversion seamless:

eu_cooling = €45000
cooling = eu_cooling in USD

The in USD expression looks up the EUR_USD rate from the front matter (1.09) and multiplies: €45,000 * 1.09 = $49,050.

CalcMark features: € currency literal; in USD exchange rate conversion; front matter exchange rates.

Operating Expenses #

Ongoing costs for a small datacenter typically fall between $50,000 and $100,000 annually:

opex = $75000
maint = 40% of opex

Electricity is often quoted per kilowatt-hour. At $0.10/hour for our
200 kW load, the annual electricity cost accumulates quickly:

annual_elec = $0.10/hour over 1 year

What does that rate look like per day?

daily_elec = $0.10/hour per day

The $0.10/hour is a rate literal. The over 1 year syntax accumulates that rate across the time period (0.10 * 8,760 hours = $876). The per day syntax converts the rate to a different time unit without accumulating.

CalcMark features: Rate literals ($0.10/hour); rate over duration for accumulation; rate per unit for rate conversion; X% of value for percentages.

OpEx Growth Projection #

Operating costs don’t stay flat. The compound function models exponential growth – here, 4% annual inflation over 5 and 10 years:

opex_5yr = compound $75000 by 4% over 5 years
opex_10yr = compound $75000 by 4% over 10 years

This is the natural language form of compound(principal, rate, periods). After 10 years at 4% inflation, $75K in annual opex grows to over $111K.

CalcMark features: compound function with natural language syntax (compound X by Y% over Z years).

Equipment Depreciation #

The depreciate function models declining-balance depreciation. Servers lose value quickly at 20% per year:

servers = depreciate $200000 by 20% over 5 years

cooling_depr = depreciate $45000 by 15% over 10 years to $5000

The to $5000 clause sets a salvage floor – the asset won’t depreciate below that value regardless of the rate and time. After 5 years, $200K in servers is worth only ~$65.5K. The cooling unit drops from $45K toward the $5K floor over 10 years.

CalcMark features: depreciate function with natural language syntax; salvage floor with to keyword.

Build vs. Colocation #

Colocation means renting space in a third-party facility. A typical 10-cabinet setup in the UK runs about £2,200/month:

colo_mo = £2200
colo_usd = colo_mo in USD

Using the rate-over-time syntax to accumulate monthly costs:

colo_3yr = $2794/month over 3 years

Compare with building your own over the same 3 years:

build_tco = capex + (opex * 3)
savings = colo_3yr - build_tco
build_cheaper = build_tco < colo_3yr

The $2794/month over 3 years accumulates monthly colocation fees into a total. Because the rate uses a currency ($), the result is a proper currency value that can be subtracted from other currencies. Building costs ~$1.1M over 3 years vs ~$102K for colocation. For a small facility, colo wins decisively.

CalcMark features: £ currency literal; in USD exchange rate conversion; rate over duration for cost accumulation; comparison operators (<) producing booleans.

Modular Alternative #

Modular (prefab) datacenters offer a pay-as-you-grow model. The grow function models linear growth – adding one $150K module per year:

grow $150000 by $150000 over 4

modular_savings = 25% of capex

Unlike compound (exponential), grow adds a fixed increment each period. Four modules at $150K each totals $750K, and modular builds typically save 20-30% vs traditional construction.

CalcMark features: grow function with natural language syntax (grow X by Y over Z); linear vs exponential growth.

Executive Summary #

The as napkin modifier rounds any value to 2 significant figures with a human-readable suffix (K, M, B):

capex as napkin
opex as napkin
build_tco as napkin
colo_3yr as napkin

Quick-reference numbers for stakeholder conversations: ~$880K to build, ~$75K/year to run, ~$1.1M total over 3 years vs ~$100K for colocation.

CalcMark features: as napkin for human-readable rounding.

Features Demonstrated #

This example showcases the following CalcMark features:

• Front matter – exchange rates and globals variables
• Currency literals – $, €, £ symbols
• Exchange rate conversion – value in USD, value in EUR
• Multiplier suffixes – $9.5M, $2300
• Percentage calculations – 42.5% of capex
• Unit conversion – 1000 kilowatts in megawatts
• Rate literals – $0.10/hour
• Rate accumulation – $0.10/hour over 1 year, $2794/month over 3 years
• Rate conversion – $0.10/hour per day
• compound() – exponential growth for inflation projection
• grow() – linear growth for modular capacity
• depreciate() – declining-balance depreciation with salvage floor
• as napkin – human-readable rounding for executive summaries
• Markdown prose – headings, paragraphs, and inline comments between calculations
• Template interpolation – {{variable}} to embed computed values in prose

Try It #

cm testdata/examples/datacenter-cost.cm