The most capital-hungry corner of infrastructure, supercharged by AI. The scarce input is not land but grid power. The decision: secure powered land and build to a pre-let, or buy a stabilised, fully-leased campus. The prize is the development spread, the gap between the yield you build at and the cap rate you exit on.
Tip, hover (or tap) any underlined term for a definition.
AI has turned data centres into the defining infrastructure trade of the decade, Blackstone and CPP bought AirTrunk for A$24bn, and platforms like Vantage have raised tens of billions. But it is not a single asset class: a hyperscale campus on a 15-year lease to one cloud tenant is closer to a bond, while a colocation facility serving many customers is an operating business. Both are gated by the same constraint, can you secure the power?
The binding constraint is a secured grid connection. Powered land, a site with committed power, is the scarce, valuable input; a platform’s real moat is its pipeline of power and interconnection.
A pre-let to a creditworthy hyperscale tenant turns a speculative build into a contracted one, it underpins the financing and the lease-up, and is what lenders and the IC want to see.
You build at a yield-on-cost of ~9–12% and a stabilised campus exits at a cap rate of ~5–6%. That spread is the developer’s margin, and the reason developing out-returns buying, if you can execute.
The risk is the mirror image: build cost and timeline (equipment lead times, power-delivery slippage), lease-up on any un-let capacity, technology and obsolescence (power density, cooling, PUE), tenant concentration, and the cap rate you actually exit on. The case below puts the develop-vs-buy trade-off on numbers.
Both routes land on the same campus. Develop builds to a pre-let, funds construction with debt-to-cost plus equity, and leases up the balance; buy stabilised acquires it fully let at a market multiple. The development spread is the difference.
—
How it works. Stabilised EBITDA = capacity (kW) × utilisation × lease rate × 12 × (1 − opex). Develop spends the build cost over the construction/lease-up period, funded by construction debt to a loan-to-cost limit plus equity; EBITDA ramps as capacity comes online and fills (a pre-let speeds the ramp). Buy stabilised acquires the fully-let campus at the entry multiple. Both sweep surplus cash to debt then distribute, and exit at the chosen multiple. The development spread is the build yield-on-cost minus the exit cap rate. Illustrative ($), not a forecast or investment advice.
A data-centre platform is assembled, not bought in one piece. The work front-loads onto securing power and a pre-let:
Option or buy sites with committed grid capacity (or a credible path to it), the diligence is utility connection agreements, timeline and cost, planning, water and fibre/interconnection. The power queue is the asset.
Negotiate the pre-let with the hyperscale or enterprise tenant, term, rate, escalators, power commitment, fit-out responsibilities and credit. The lease (and the tenant covenant) underpins the entire financing and valuation.
You can also buy a platform outright, a developer with land, power pipeline, a team and a customer book, which is what the marquee deals (e.g. AirTrunk) are: a competitive auction for scarcity and a pipeline, valued on committed and pipeline MW and stabilised EV/EBITDA. Diligence then centres on the power pipeline’s deliverability, the lease book and tenant concentration, the development team, and the technical spec against where the market (and AI density) is heading.
Like fibre, the debt evolves with the asset. A speculative or pre-let build is funded by a construction / development facility sized to a loan-to-cost limit, drawn against milestones, with the rest equity, the J-curve here is short and the pre-let de-risks it. Once a campus is built and leased, the contracted rent supports far more leverage: a long-dated term facility, a private placement, green bonds, and, for stabilised, leased portfolios, data-centre asset-backed securitisation (ABS), where the leases themselves are the collateral. The equity strategy is to recycle: build or buy, stabilise, refinance or sell down to a core buyer, and redeploy into the next site, the platform compounds on a pipeline of power.
Illustrative examples, not endorsements.
M&A advice, plus specialist data-centre technical and commercial DD.
Construction facilities, then long-term and securitised debt.
Co-invest, the grid, and the contractors who build it.
Data-centre development is, in this series, the highest-return and among the highest-risk cases, the mirror of the greenfield-fibre build, with bigger cheques and a hotter market. The point-estimate IRR flatters developing because it books the development spread as certain; in reality you are underwriting power on time, a tenant that signs and pays, a build to budget, technology that does not date, and a cap rate that holds. Buying stabilised inverts that, contracted income, low execution risk, a return capped near the cap rate. The disciplined position is to develop only what you have pre-let and powered, keep construction leverage modest, underwrite a conservative exit, and treat the AI-demand tailwind as an accelerant, not a substitute for any of the above. Get those right and it is the standout return in infrastructure today; get the power or the tenant wrong and a half-built, unpowered shell is worth a fraction of its cost.
Market context (the AirTrunk and Vantage transactions and current data-centre economics) is from public disclosure and is illustrative. The model is simplified, it omits phasing detail, working capital, the full tax and lease structure and securitisation mechanics, and is for illustration, not a forecast or investment advice. Named firms are illustrative examples of active categories, not recommendations.