Home
Industries
Data centres

Industry brief · Power & digital

Data centreshyperscale, colocation, AI training campuses

Hyperscale self-build and leased capacity, wholesale/retail colocation, and gigawatt-scale AI training campuses (Stargate Abilene, Meta Hyperion, xAI Colossus, Reliance Jamnagar, Google Visakhapatnam). Data centres used ~415 TWh (1.5% of global electricity) in 2024, heading to ~945 TWh by 2030; big-tech capex reached record levels (Amazon USD 132bn and Alphabet USD 91bn in 2025; Microsoft USD 116bn in FY2026). Capital risk is now governed by power availability (grid queues, on-site generation), electrical long-lead equipment (transformers, switchgear, gensets, turbines), liquid cooling for 100-600 kW racks, GPU obsolescence and AI-demand/financing cycles.

Start a Decision Sprint See the parameters

Pack v1.0.0 · updated 5 Oct 2026 · 20 cited sources · 14 parameters · 10 swing factors.

At a glance

12Capex intensity · USD million per MW critical IT (facility only); AI all-in 35-60range 5–60 · estimate

20Construction time · months (shell to RFS)range 12–36 · estimate

10%Mean cost overrunrange 0%–35% · estimate

25%Mean schedule overrunrange 0%–100% · sourced

Pack defaults with their low–high range. Overruns are sector means as a share of the original budget or schedule. All parameters and sources

Sub-industries

Capex intensity ranges by configuration. The unit changes with the asset, so compare within a row, not across rows.

Sub-industry	Capex intensity	Notes
Hyperscale self-build campuses100 MW-1 GW+ owner-built campuses by AWS, Microsoft, Google, Meta, Oracle; shell + powered core + MEP; IT equipment financed separately.	9–15USD million per MW (critical IT, ex-IT equipment)	estimate: US/EU shell+MEP ~USD 10-14m/MW for air/liquid hybrid designs; India ~USD 5-8m/MW.
Wholesale and retail colocationEquinix, Digital Realty, NTT, STACK, Vantage, AdaniConneX, CtrlS, Yotta, Sify, Nxtra; leased per kW/month with power pass-through.	5–13USD million per MW	estimate; India colocation ~Rs 45-70 crore/MW (~USD 5-7m/MW); US hyperscale-leased powered shells lower, retail colo higher.
AI training / GPU 'AI factory' campuses incl. neocloudsLiquid-cooled 120-600 kW racks (GB200/GB300 NVL72, Rubin), 300 MW-5 GW campuses (Stargate, Hyperion, Colossus), CoreWeave/Nebius/Crusoe/IREN neoclouds.	35–60USD million per MW (all-in incl. IT)	estimate: facility ~USD 10-15m/MW plus GPUs/networking ~USD 25-45m/MW. Stargate targets up to USD 500bn by 2029.
Edge and enterprise data centres<10 MW sites for inference, telco edge, enterprise on-prem and sovereign clouds.	8–20USD million per MW	estimate.
On-site / bridging power for data centresBehind-the-meter gas turbines/recips, fuel cells (Bloom), BESS and nuclear co-location/PPAs to bypass grid delays.	1,100–3,500USD per kW	Gas peaking USD 1,100-1,650/kW (Lazard 2026); aeroderivatives USD 2,275/kW (EIA AEO2026); fuel cells USD 8,287/kW (EIA).

Value chain

Inputs

land with power and fibre
grid capacity / large-load interconnection
power transformers, MV/LV switchgear, busway
UPS, batteries, diesel/gas generators
chillers, CDUs, cold plates, cooling towers
structural steel, precast concrete
GPUs/accelerators, servers, network switches, optics
water (or dry/adiabatic cooling)
PPAs and RECs/24x7 clean energy

Process

site selection and power procurement
zoning, environmental and water permits
design (Tier III/IV, liquid cooling, 800 V DC readiness)
shell and core construction
electrical and mechanical fit-out
utility substation and interconnection
integrated systems testing (L1-L5)
IT deployment and cluster bring-up
operations, energy management, refresh

Outputs

leased critical IT capacity (MW, kW/month)
cloud and AI compute services
AI model training and inference
interconnection/cross-connect services
waste heat (district heating in Nordics)

Parameters

14 parameters: 4 sourced, 10 informed estimates. The bar shows the low–high range with the default marked; estimates are labelled and never presented as observed data.

Parameter	Default & range	Evidence	Source
Capex intensity(facility only); AI all-in 35-60	12 USD million per MW critical ITrange 5–60	estimateas of 5 Oct 2026	industry cost benchmarks; hyperscaler disclosures
Construction months(shell to RFS)	20 monthsrange 12–36	estimateas of 5 Oct 2026	xAI Colossus ~4 months exceptional; typical 18-24
Cost overrun mean	10%range 0%–35%	estimateas of 5 Oct 2026	Informed estimate; no single source
Schedule overrun mean	25%range 0%–100%	sourcedas of 5 Oct 2026	iea.org
Asset life (years)(shell/MEP; IT 4-6)	25 yearsrange 15–40	estimateas of 5 Oct 2026	Informed estimate; no single source
WACC	8%range 6%–13%	estimateas of 5 Oct 2026	REITs ~7-8%; neoclouds 10-13%
Typical IRR(unlevered, stabilised colo)	11%range 8%–18%	estimateas of 5 Oct 2026	Equinix targets ~25% cash-on-cash for xScale/retail builds
Utilization(leased/occupied IT capacity)	85%range 60%–98%	estimateas of 5 Oct 2026	Informed estimate; no single source
Opex % of revenue	55%range 40%–70%	estimateas of 5 Oct 2026	power dominant
EBITDA margin(colocation REITs)	47%range 35%–60%	estimateas of 5 Oct 2026	Equinix adj. EBITDA ~47-50%, Digital Realty ~55%
PUE	1.4 ratiorange 1.1–1.8	sourcedas of 5 Oct 2026	uptimeinstitute.com
Power share of global electricity(2024)	1.5%range 1.5%–3%	sourcedas of 5 Oct 2026	iea.org
Global DC TWh 2030	945 TWhrange 700–1,700	sourcedas of 5 Oct 2026	iea.org
Electrical share of facility capex	42%range 35%–50%	estimateas of 5 Oct 2026	Informed estimate; no single source

What moves value

Sorted by expected cost (probability × cost impact). Impacts are typical values for the sector; each one is driven by the shared factors listed, which is how one shock reaches several projects at once.

Swing factorProbabilityCost impactSchedule

Electrical long-lead equipment delays (transformers, switchgear, gensets, turbines)Supply chain Shared factors: Electrical equipment 0.7 · Copper 0.2 · Data-centre demand 0.1Watch: transformer/switchgear PPI; Vertiv/Eaton/Schneider book-to-bill; genset delivery quotes

45%

9 mo

Technology obsolescence (rack density, cooling, 800 VDC)Technical Shared factors: Electrical equipment 0.5 · Data-centre demand 0.5Watch: NVIDIA roadmap (Rubin/Kyber 600 kW racks); design change orders mid-construction

35%

10%

3 mo

Electrician and MEP labour shortageLabour Shared factors: Labour 0.8 · Data-centre demand 0.2Watch: construction wage growth; concurrent megaprojects in metro; per-diem/travel premiums

40%

4 mo

Grid power not available on schedule (large-load interconnection)Interface Shared factors: Grid queue 0.6 · Data-centre demand 0.2 · Power price 0.2Watch: utility large-load queue (GW); substation in-service dates; new large-load tariffs/collateral rules

45%

18 mo

On-site generation execution (gas turbines, emissions permits)Regulatory Shared factors: Electrical equipment 0.4 · Gas 0.3 · Policy 0.3Watch: air permits for BTM turbines; turbine slot availability; gas lateral capacity

30%

6 mo

Interest-rate and refinancing riskFinancing Shared factors: Interest rates 1Watch: US 10Y (5.24% Oct-2026); DC ABS/CMBS spreads; private credit appetite

30%

0 mo

Local opposition, moratoria, water and noise permitsESG Shared factors: Policy 0.9 · Labour 0.1Watch: county zoning moratoria (VA, GA, AZ); Dublin/Amsterdam grid moratoria; water-use disputes

30%

6 mo

Policy: export controls, data localisation, energy-efficiency rulesRegulatory Shared factors: Policy 1Watch: US AI chip export rules; India DPDP Act rules; EU EED reporting/PUE mandates

25%

3 mo

AI demand / tenant concentration and financing cycleMarket Shared factors: Data-centre demand 0.7 · Interest rates 0.3Watch: hyperscaler capex guidance revisions; neocloud credit spreads; GPU rental prices (H100/B200 per hour)

30%

6 mo

Electricity price and supply-cost inflationCommodity Shared factors: Power price 0.6 · Gas 0.4Watch: PJM capacity prices; industrial power PPI; TTF/Henry Hub

35%

0 mo

Market signals

What moves the case

Key variables with their typical range and what they hit in the model.

Variable	Typical range	Affects
Big-tech capex (AMZN+MSFT+GOOGL+META+ORCL)2025: AMZN 131.8, GOOGL 91.4, META 69.7; MSFT FY26 115.9; ORCL FY26 55.7 (SEC XBRL). Demand signal for every supplier in the chain.	200–700USD bn/yr	Revenue, Capex
Grid capacity availability / time-to-powerIEA: ~20% of planned data-centre projects at risk of delay from grid constraints; 3-7 yr waits in N. Virginia, Dublin, Amsterdam, Frankfurt.	12–84months to energisation	Schedule, Revenue
Power transformer PPI+86% since 2020; electrical systems ~40-45% of facility capex.	250–500index	Capex, Schedule
Switchgear PPI+13% y/y Aug-2026.	200–450index	Capex
AC/refrigeration equipment PPIMechanical/cooling ~15-20% of facility capex; liquid cooling (CDUs) adds USD 0.5-1.5m/MW - estimate.	200–350index	Capex
Industrial electricity price (PPI)Power is ~40-60% of colo opex; 100 MW at PUE 1.3 ~ 1.1 TWh/yr: +USD 10/MWh ~ +USD 11m/yr.	250–400index	Opex, Margin
Power usage effectivenessHyperscale 1.1-1.2; industry average ~1.5-1.6 (Uptime).	1.1–1.8ratio	Opex
Rack power densityGB200 NVL72 ~120-140 kW; Rubin Ultra Kyber ~600 kW (2027) - forces liquid cooling and 800 VDC distribution.	8–600kW/rack	Capex

Show 4 more variables

Variable	Typical range	Affects
Wholesale colocation rentRecord-low vacancy pushed North American wholesale rents up ~10-20%/yr in 2023-25 - estimate.	100–250USD/kW-month	Revenue
Semiconductor device PPIIT is 60-75% of AI campus all-in capex; GPU generation cadence ~1 yr drives obsolescence.	25–35index	Capex
US 10-year TreasuryNeocloud/colo debt (ABS, GPU-backed loans) highly rate-sensitive; 10Y at 5.24% (Oct-2026).	3–6percent	Financing
US construction hourly earningsElectricians are the binding trade; labour ~25-30% of facility cost.	35–50USD/h	Capex

Price feeds (15)

Series	Source / id	Unit	Frequency
PPI: Power and specialty transformers	PCU335311335311FRED	index	monthly
PPI: Switchgear and switchboard apparatus	PCU335313335313FRED	index	monthly
PPI: AC, refrigeration and forced-air heating equipment	PCU333415333415FRED	index	monthly
PPI: Other engine equipment (generator sets)	PCU333618333618FRED	index	monthly
PPI: Motor and generator manufacturing	PCU335312335312FRED	index	monthly
PPI: Electronic computer manufacturing	PCU334111334111FRED	index	monthly
PPI: Semiconductor and related devices	PCU334413334413FRED	index	monthly
PPI: Data processing, hosting and related services	PCU518210518210FRED	index	monthly

Show 7 more feeds

Series	Source / id	Unit	Frequency
PPI: Industrial electric power	WPU0543FRED	index	monthly
Average price: electricity per kWh	APU000072610FRED	USD/kWh	monthly
Global price of Copper	PCOPPUSDMFRED	USD/t	monthly
PPI: Construction materials	WPUSI012011FRED	index	monthly
Average hourly earnings, construction	CES2000000003FRED	USD/h	monthly
US 10-year Treasury yield	DGS10FRED	percent	daily
Pacer Data & Infrastructure Real Estate ETF	SRVRYahoo	USD	daily

Listed tickers (30)

Market context only: these prices tend to react first when a driver moves. Named for context; no affiliation.

Operators · 12

ANANTRAJ.NS Anant Raj (data centre parks) NSE · IN
TATACOMM.NS Tata Communications NSE · IN
RELIANCE.NS Reliance Industries (Jamnagar AI DC / Jio) NSE · IN
E2E.NS E2E Networks (GPU cloud) NSE · IN
SIFY Sify Technologies NASDAQ · IN
EQIX Equinix NASDAQ · US
DLR Digital Realty NYSE · US
IRM Iron Mountain (data centres) NYSE · US
CRWV CoreWeave NASDAQ · US
NBIS Nebius Group NASDAQ · EU
IREN IREN NASDAQ · US
APLD Applied Digital NASDAQ · US

Suppliers · 12

NETWEB.NS Netweb Technologies (AI servers) NSE · IN
TECHNOE.NS Techno Electric (DC development, EPC) NSE · IN
BLUESTARCO.NS Blue Star (DC cooling) NSE · IN
CUMMINSIND.NS Cummins India (gensets) NSE · IN
NVDA NVIDIA NASDAQ · US
VRT Vertiv NYSE · US
ETN Eaton NYSE · US
ANET Arista Networks NYSE · US
BE Bloom Energy (on-site fuel cells) NYSE · US
SU.PA Schneider Electric Euronext Paris · EU
LR.PA Legrand Euronext Paris · EU
ABBN.SW ABB SIX · EU

Consumers · 5

AMZN Amazon (AWS) NASDAQ · US
MSFT Microsoft NASDAQ · US
GOOGL Alphabet NASDAQ · US
META Meta Platforms NASDAQ · US
ORCL Oracle (OCI, Stargate) NYSE · US

Proxys · 1

SRVR Pacer Data & Infrastructure Real Estate ETF NYSE Arca · US

Trends and research findings

Trends

Data-centre electricity doubles by 20302024-2035
Data centres used ~415 TWh in 2024 (1.5% of global electricity; US 45%, China 25%, Europe 15%) rising to ~945 TWh by 2030 and ~1,200 TWh by 2035.
iea.org
Record hyperscaler capex2025-2027
Amazon USD 131.8bn (2025), Microsoft USD 115.9bn (FY Jun-2026), Alphabet USD 91.4bn (2025), Meta USD 69.7bn (2025), Oracle USD 55.7bn (FY May-2026) in purchases of property and equipment.
data.sec.gov
Gigawatt AI campuses and Stargate2025-2029
Stargate (OpenAI/SoftBank/Oracle/MGX) plans up to USD 500bn by 2029 starting with ~10 data centres in Abilene, TX.
en.wikipedia.org
Grid is the bottleneck2025-2030
~20% of planned data-centre projects could be delayed by grid constraints; 50% of US capacity under development sits in existing clusters, raising local bottleneck risk.
iea.org
Firm clean power procurement2025-2035
Hyperscalers sign nuclear restart/uprate and SMR deals (Crane-Microsoft 20-yr PPA) and add behind-the-meter gas/fuel cells.
en.wikipedia.org
India as an AI infrastructure hub2025-2030
Hyperscaler and conglomerate commitments (Google Vizag hub, Reliance Jamnagar, AdaniConneX 1 GW, Microsoft/AWS expansions) plus IndiaAI Mission GPUs; India capacity expected to roughly triple by 2030 - estimate.
estimate Informed estimate; no single source

Research findings

Data-centre investment
Global investment in data centres nearly doubled since 2022 to ~USD 0.5tn in 2024.
iea.org
Hyperscaler capex step-change (SEC 10-K)
Alphabet capex rose from USD 52.5bn (2024) to USD 91.4bn (2025); Meta from USD 37.3bn to USD 69.7bn; Amazon from USD 83.0bn to USD 131.8bn.
sec.gov
Microsoft FY2026
Purchases of property and equipment USD 115.9bn in FY ending Jun-2026 vs USD 64.6bn FY2025 and USD 44.5bn FY2024.
sec.gov
Oracle FY2026
Capex USD 55.7bn in FY ending May-2026 vs USD 21.2bn FY2025 as OCI builds Stargate capacity.
sec.gov
Supply mix for data centres
Renewables grow by >450 TWh and natural gas by 175 TWh to meet data-centre demand to 2035 (IEA base case).
iea.org
Electrical equipment inflation
Transformer PPI +86%, switchgear +89% and HVAC equipment +59% since Jan-2020 (FRED, Aug-2026).
fred.stlouisfed.org
NVIDIA revenue as demand proxy
NVIDIA revenue USD 215.9bn in FY ending Jan-2026 vs USD 130.5bn prior year.
sec.gov

KPIs, regions and who builds

KPIs the pack tracks

MW critical IT under construction / RFS
Capex per MW (facility and all-in)
Months from land to energisation
PUE and WUE
Utilisation / pre-leasing (%)
Rent (USD/kW-month)
Power contracted (GW) and PPA coverage
EBITDA margin (%)
Rack density (kW/rack)
Construction cost escalation vs budget
Interconnection/cross-connect growth
Tenant concentration (% top-3)

Regional notes

India	~1.5 GW operating with rapid growth driven by data-localisation (DPDP Act), cloud and AI (IndiaAI Mission); clusters in Mumbai/Navi Mumbai, Chennai, Hyderabad, Noida, Vizag. Advantages: RE open access/green tariffs, lower capex (~USD 5-7m/MW); constraints: state power reliability, transformer supply, water, DC policy incentives vary by state. Key players: AdaniConneX, Yotta, CtrlS, Nxtra, STT, NTT, Sify, Reliance; Google/Microsoft/AWS multi-billion commitments.
United States	N. Virginia, Texas, Ohio, Georgia, Arizona, Louisiana lead; utility large-load queues are hundreds of GW; PJM capacity prices at cap; FERC co-location rulings; BTM gas and nuclear PPAs proliferate; CHIPS-era tariffs and Section 232 raise equipment costs.
Europe	FLAP-D markets power-constrained (Dublin moratorium, Amsterdam, Frankfurt); growth shifts to Nordics, Iberia, Italy; EU Energy Efficiency Directive mandates reporting; InvestAI 'AI gigafactories'; heat reuse rules (Germany EnEfG).
Middle East	UAE (G42/Khazna, 5 GW UAE Stargate) and Saudi (Humain) build with cheap gas/solar and US chip-export arrangements.

Who builds and operates

Named for context; no affiliation. These are public market participants drawn from the research behind the pack. They are not Capibud clients or partners.

Operators · 13

Amazon Web Services Seattle · USAmazon capex USD 131.8bn in 2025 (USD 83.0bn in 2024); Project Rainier (Anthropic) campus in Indiana; large India commitments (Maharashtra, Telangana).
Microsoft Redmond · USCapex USD 115.9bn FY2026 (USD 64.6bn FY2025) ex-finance leases; Fairwater AI campuses (Wisconsin, Atlanta); Crane nuclear PPA.
Alphabet / Google Mountain View · USCapex USD 91.4bn in 2025 (USD 52.5bn 2024); ~USD 15bn AI hub at Visakhapatnam, India with AdaniConneX/Airtel (announced Oct-2025).
Meta Menlo Park · USCapex USD 69.7bn in 2025; Hyperion (Louisiana, multi-GW) and Prometheus (Ohio) AI campuses; gas-fired supply via Entergy.
Oracle / OpenAI / SoftBank (Stargate) Austin / San Francisco / Tokyo · USOracle capex USD 55.7bn FY2026; Stargate up to USD 500bn by 2029 with Abilene, TX flagship (~1.2 GW).
Equinix Redwood City · US~270 IBX sites; capex USD 4.3bn 2025; xScale JV for hyperscale.
Digital Realty Austin · US~300+ data centres; development spend USD 3.2bn 2025; JV with Reliance (Digital Connexion) in India.
CoreWeave / Nebius / Crusoe / Lambda US / Amsterdam · USGPU neoclouds with multi-GW contracted power; highly leveraged GPU-backed financing.
AdaniConneX (Adani-EdgeConneX) Ahmedabad · INTargets 1 GW in India; partner for Google Vizag AI hub.
Yotta / CtrlS / Nxtra (Airtel) / Sify / STT GDC India / NTT India Mumbai / Hyderabad / Delhi / Chennai · INIndia colocation leaders; IndiaAI Mission GPU providers (Yotta, E2E).
Reliance Jio / Reliance Intelligence Mumbai · INJamnagar GW-scale AI data centre powered by own renewables; Meta/Google partnerships.
NTT Data / Vantage / STACK / CyrusOne / Data4 Tokyo / Denver / Denver / Dallas / Paris · EUPan-EU campuses (Frankfurt, Paris, Milan, Madrid) facing FLAP-D power constraints.
G42 / Khazna / Humain Abu Dhabi / Riyadh · MEUAE Stargate (5 GW campus) and Saudi Humain AI build-out.

Suppliers · 12

NVIDIA OEM · USFY2026 revenue USD 215.9bn (USD 130.5bn FY2025); GB200/GB300/Rubin systems define facility design (liquid cooling, 800 VDC).
Vertiv OEM · USUPS, thermal management, CDUs, prefabricated power modules.
Schneider Electric / Eaton / ABB / Legrand / Siemens OEM · EUMV/LV switchgear, UPS, busway; multi-year DC backlogs.
Hitachi Energy / Siemens Energy / HD Hyundai Electric / Hyosung OEM · EULarge power transformers and substations for 300 MW+ campuses.
Caterpillar / Cummins / Rolls-Royce mtu / Kohler OEM · USBackup gensets (2-3 MW units) with 1-2 year lead times; also prime-power recips.
GE Vernova / Siemens Energy / Mitsubishi / Solar Turbines / Bloom OEM · USOn-site gas turbines and fuel cells for bridging/prime power.
Trane / Carrier / Johnson Controls / Daikin / Blue Star OEM · USChillers, CRAH, liquid-cooling heat rejection.
Arista / Cisco / Broadcom / Corning / Coherent OEM · USNetworking, optics and structured fibre.
Dell / Supermicro / HPE / Foxconn / Wiwynn / Netweb OEM · USAI rack-scale server integration.
Turner / DPR / Holder / Fluor / L&T / Tata Projects / Sterling & Wilson EPC · USGeneral contractors and MEP EPC; Indian EPCs for domestic DCs.
Quanta / MYR / Rosendin / IES EPC · USElectrical contractors and substation builders.
Constellation / Talen / Vistra / NextEra / Brookfield Services · USPower suppliers via PPAs, co-location and bring-your-own-generation.

Customers · 6

Frontier AI labs (OpenAI, Anthropic, xAI, Google DeepMind, Meta) AI trainingGW-scale training clusters; long-term take-or-pay compute contracts.
Cloud customers / enterprises cloud & inferenceShift of enterprise workloads plus AI inference growth.
Financial services and exchanges colocation / low-latencyRetail colo and interconnection hubs (Mumbai BKC, NJ, London, Frankfurt).
Governments / sovereign AI (IndiaAI Mission, EU AI gigafactories) sovereign computeIndiaAI subsidised GPU pools (~34,000+ GPUs); EU InvestAI AI gigafactories.
Content, streaming and telecom operators edge/CDNEdge caching, 5G core.
Utilities (as counterparties) large-load tariffsMinimum-take, collateral, and curtailable-load (flexibility) tariffs.

Project template

Work breakdown

Share of total duration per step; steps overlap where predecessors allow, as the bars show. O / ML / P are optimistic, most-likely and pessimistic multipliers on each step's duration; the hatched tail is the pessimistic case.

ID	Step & timing	Share	O / ML / P
A100	Site selection, land, fibre and power due diligenceDevelop · factors: Grid queue, Data-centre demand	12%	0.9 / 1 / 1.4
A110	Utility large-load agreement / ESA and substation scopeDevelop · factors: Grid queue, Policy	35%	0.85 / 1 / 2
A120	Zoning, environmental, air (gensets) and water permitsDevelop · factors: Policy, Labour	15%	0.9 / 1 / 1.8
A130	Anchor tenant / pre-lease and financingDevelop · factors: Data-centre demand, Interest rates	10%	0.9 / 1 / 1.5
B100	Basis of design (Tier, density, liquid cooling, power topology) and detailed designEngineer · EPC · factors: Labour, Data-centre demand	12%	0.9 / 1 / 1.4
C100	Long-lead electrical: HV/MV transformers, switchgear, UPS, gensetsProcure · OEM · factors: Electrical equipment, Copper, Data-centre demand	40%	0.9 / 1 / 1.6
C110	Mechanical: chillers, CDUs, cooling towers / dry coolersProcure · OEM · factors: Electrical equipment, Steel, Data-centre demand	30%	0.9 / 1 / 1.5
D100	Site works, foundations and shell (steel/precast)Construct · EPC · factors: Construction materials, Steel, Labour	25%	0.9 / 1 / 1.4
D110	Electrical and mechanical fit-out (MEP)Construct · EPC · factors: Labour, Copper, Electrical equipment	30%	0.9 / 1 / 1.5
D120	Utility substation, transmission tap and on-site generationConstruct · EPC · factors: Grid queue, Electrical equipment, Gas	30%	0.9 / 1 / 1.8
E100	Integrated systems testing (L1-L5) and ready-for-serviceCommission · EPC · factors: Labour, Electrical equipment	6%	0.9 / 1 / 1.5
E110	IT deployment, cluster bring-up and tenant acceptanceCommission · OEM · factors: Data-centre demand, Electrical equipment	8%	0.9 / 1 / 1.6

Cost breakdown

Share of capex by line, with the commodity exposure of each line (the share of that line that moves with the commodity).

Electrical systems (substation, transformers, switchgear, UPS, gensets, busway)38%
Equipment
- Electrical equipment 60%
- Copper 30%
- Steel 10%
Mechanical / cooling (chillers, CDUs, piping, heat rejection)18%
Equipment
- Electrical equipment 40%
- Steel 30%
- Copper 30%
Shell, structure and site works16%
Construction
- Construction materials 60%
- Steel 40%
Construction labour and GC fees12%
Labour
- Labour 100%
Land, power and fibre development5%
Develop
- Grid queue 60%
- Data-centre demand 40%
Financing, insurance and soft costs5%
Indirects
- Interest rates 100%
Design and engineering4%
Engineering
- Labour 100%
Commissioning (IST, load banks)2%
Commissioning
- Power price 50%

Financial template

Revenue model

Colocation: kW/month rents (10-15 yr hyperscale leases with escalators) plus power pass-through and interconnection fees; hyperscalers: cloud/AI service revenue on self-built capacity; neoclouds: take-or-pay GPU-hour contracts (multi-year with AI labs/hyperscalers); facility capex is depreciated over 15-40 yrs and IT over ~5-6 yrs.

Margin driver

Wholesale colocation rent

Ramp-up

2 years to steady state

How Capibud uses this pack

A pack is a starting position, not a forecast. When a data centres decision opens in Capibud, pack v1.0.0 is copied in and pinned to that decision. The 14 parameters, 10 swing factors and templates the committee saw stay attached to the decision record, even after the pack is next refreshed.

Parameter ranges and WBS durations seed the P50, P80 and P90 for capex, schedule and value. Your own estimates, vendor quotes and actuals replace pack defaults line by line, and the record shows which numbers are still defaults.

Each swing factor is wired to shared factors; here the heaviest are electrical equipment, data-centre demand, grid queue and policy. A move in one factor hits every exposed project in the same scenario, which is how the capital committee sees concentration across a book rather than project by project.

The 15 price feeds and 30 listed tickers feed the brain, which watches for drivers moving outside their pack range and drafts the change for a person to review.

Shared factors behind this pack's swing factors

Electrical equipment
Data-centre demand
Grid queue
Policy
Labour
Power price
Interest rates

Relative weight: probability × impact × driver weight, summed over swing factors.

Capex tied to a shared factor

Electrical equipment30%
Copper16.8%
Labour16%
Steel15.6%
Construction materials9.6%
Interest rates5%
Grid queue3%

Share of total capex, from the cost breakdown and its commodity exposures.

Sources

Every link opens the original source in a new tab. Values marked as estimates are informed estimates by the Capibud research team.

iea.orghttps://www.iea.org/reports/energy-and-ai/executive-summary
iea.orghttps://www.iea.org/news/ai-is-set-to-drive-surging-electricity-demand-from-data-centres-while-offering-the-potential-to-transform-how-the-energy-sector-works
data.sec.govhttps://data.sec.gov/api/xbrl/companyconcept/CIK0000789019/us-gaap/PaymentsToAcquirePropertyPlantAndEquipment.json
data.sec.govhttps://data.sec.gov/api/xbrl/companyconcept/CIK0001018724/us-gaap/PaymentsToAcquireProductiveAssets.json
data.sec.govhttps://data.sec.gov/api/xbrl/companyconcept/CIK0001652044/us-gaap/PaymentsToAcquirePropertyPlantAndEquipment.json
data.sec.govhttps://data.sec.gov/api/xbrl/companyconcept/CIK0001326801/us-gaap/PaymentsToAcquirePropertyPlantAndEquipment.json
data.sec.govhttps://data.sec.gov/api/xbrl/companyconcept/CIK0001341439/us-gaap/PaymentsToAcquirePropertyPlantAndEquipment.json
sec.govhttps://www.sec.gov/Archives/edgar/data/789019/000119312526323660/
sec.govhttps://www.sec.gov/Archives/edgar/data/1018724/000101872426000004/
sec.govhttps://www.sec.gov/Archives/edgar/data/1652044/000165204426000018/
sec.govhttps://www.sec.gov/Archives/edgar/data/1326801/000162828026003942/
sec.govhttps://www.sec.gov/Archives/edgar/data/1341439/000119312526277521/
sec.govhttps://www.sec.gov/Archives/edgar/data/1101239/000110123926000032/
sec.govhttps://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/
en.wikipedia.orghttps://en.wikipedia.org/wiki/Stargate_LLC
lazard.comhttps://www.lazard.com/media/kcfconhf/lazards-lcoeplus_vf.pdf
eia.govhttps://www.eia.gov/outlooks/aeo/assumptions/pdf/EMM_Assumptions.pdf
uptimeinstitute.comhttps://uptimeinstitute.com/
fred.stlouisfed.orghttps://fred.stlouisfed.org/series/PCU335313335313

Also cited in this brief

en.wikipedia.orghttps://en.wikipedia.org/wiki/Three_Mile_Island_Nuclear_Generating_Station

Bring us one real decision in data centres.

We start it from the Data centres pack v1.0.0: 14 parameters, 10 swing factors and the 12-step project template, pinned to your decision.