Own or rent:
the split inside the AI datacenter buildout.
The largest capex cycle in corporate history has developed a fault line. As CNBC's Kate Rooney reported this week, the hyperscalers are diverging on how they acquire AI capacity: Amazon is leaning into building and owning its own datacenters, while Microsoft is leaning on renting capacity from third parties.
The split is worth more attention than it gets. It is not a disagreement about demand — both companies are spending at record levels. It is a disagreement about who should own the buildings — and for the independent datacenter market, one side of that split is the demand curve itself.
Whyte Consolidated Research · 2026-08-13· 8 min read · On who owns the buildings
Same demand. Different balance sheets.
Every hyperscaler faces the same problem: AI demand is compounding faster than datacenters can be built. Microsoft, Alphabet, Amazon and Meta have collectively guided to roughly $725 billion of capital expenditure in 2026, and the vast majority lands in datacenters and what fills them. Where the strategies split is on the question underneath the spending: should we own the building, or rent it?
The build-and-own path treats the campus as a strategic asset. Owning the land, the power position, the shell and the cooling design means controlling cost, security and the upgrade path across hardware generations — and never negotiating with a landlord for the space your models run in.
The rent path treats capacity as the product. Leasing from third-party developers and specialized operators means capacity arrives on someone else's construction timeline — a way to add compute faster than your own pipeline allows, moderate balance-sheet intensity, and keep flexibility if demand shifts. The price is the developer's margin, and competition for space in a market with 1.4% vacancy where 74.3% of new supply is preleased before delivery.
| Aspect | Build & own | Rent |
|---|---|---|
| Speed to capacity | gated by your own land, power and construction pipeline | gated by someone else's — capacity arrives on a developer's timeline |
| Control | full stack — site, power, cooling, silicon, security, upgrade path | negotiated — the landlord owns the shell and the power position |
| Balance sheet | capex-heavy; the asset and its depreciation are yours | lease obligations; capacity without owning the building |
| Demand uncertainty | a campus is forever — overbuild risk is yours | leases roll off — flexibility is the product being bought |
| Long-run cost | cheapest per megawatt at sustained full utilization | carries the developer's margin — the price of speed and optionality |
| Who ends up owning the asset | the hyperscaler | the developer or investor holding the lease — the landlord |
Two strategies, one scarcity.
Read the split closely and something important emerges: it is not really a disagreement about datacenters. Both paths need the same physical inputs, in the same power-constrained markets, from the same supply chains. Whether the capex flows through a hyperscaler's construction arm or a developer's balance sheet, it converges on the same scarce inputs:
- →Entitled land in power-advantaged corridors — the same short list of markets for both strategies.
- →Secured megawatts and interconnection position — queues measured in years regardless of who is building.
- →Transformers, switchgear and cooling on multi-year lead times — one equipment market, two buyer types.
- →Record-low vacancy and preleased pipelines — the rent path bids against the build path for the same space.
That is why neither strategy is “winning.” Both are expanding at once, because demand exceeds what either path can deliver alone. Even committed self-builders lease when interconnection queues run long; even lease-heavy operators keep building flagship campuses. The constraint is not strategy — it is deliverable, powered capacity.
Every rent decision creates a landlord.
For the independent datacenter market, the most important line in the split is the quiet one: a hyperscaler that rents is a hyperscaler signing leases — multi-year commitments from some of the most creditworthy counterparties in the world, on facilities somebody else owns.
Those are precisely the anchor contracts that make third-party datacenters financeable. As we laid out in our analysis of NVIDIA's $500 billion financing initiative, the institutional capital now forming around AI infrastructure underwrites one thing above all: a strong offtake from a creditworthy customer. A lease-leaning hyperscaler supplies exactly that — at scale, on repeat. The rent strategy is, mechanically, the demand side of the landlord business.
And because the rent path bids against the build path for the same entitled land, secured megawatts and equipment queues, the owners of power-secured capacity sit on the short side of both strategies at once. That asymmetry — two demand channels, one scarce input — is the core of our thesis, and the environment our 3.6 GW pipeline is being built into.
The split strategy will keep shifting with rates, chips and demand. The buildings will not. Whoever wins the own-vs-rent debate, the facility gets built — and someone collects the value of owning it.
The own-vs-rent split — questions
- What is the own-vs-rent split in the AI buildout?
- Hyperscalers are diverging on how they acquire datacenter capacity. Some, like Amazon, lean toward building and owning their own campuses — controlling the land, power, shell and silicon end to end. Others, like Microsoft, lean on renting: leasing capacity from third-party developers and specialized GPU clouds to add compute faster than their own construction pipeline allows. As CNBC's Kate Rooney reported, the split is now visible in the companies' spending plans.
- Why would a hyperscaler choose to build and own?
- Control and cost at scale. Owning the campus means owning the power position, the cooling design, the security perimeter and the upgrade path — and integrating custom silicon and networking without a landlord in the loop. Over a multi-decade horizon, owning a fully utilized facility is generally cheaper than leasing it, and the land and shell hold value across hardware generations.
- Why would a hyperscaler choose to rent?
- Speed and flexibility. Leased capacity arrives on someone else's construction timeline, which matters when demand is compounding faster than interconnection queues clear. Renting also moderates balance-sheet intensity and hedges demand uncertainty: if AI workloads shift or slow, a lease rolls off — a campus does not. The trade is paying a developer's margin and competing for scarce third-party space.
- Which strategy is winning?
- Neither, and that is the point — both strategies are expanding at once because demand exceeds what either path can deliver alone. Even the most committed self-builders lease when queues run long, and even lease-heavy operators keep building flagship campuses. The constraint is not strategy; it is deliverable, powered capacity.
- What does the split mean for datacenter owners and developers?
- Every 'rent' decision creates a landlord. A hyperscaler that leases capacity is signing the multi-year, creditworthy tenant commitments that make third-party facilities financeable — the anchor contracts that institutional capital underwrites. The lease-heavy strategy is, mechanically, the demand side of the independent datacenter market.
- Does the split change what gets built?
- It changes who builds it more than what gets built. Both paths need the same physical inputs — entitled land, secured megawatts, transformers, cooling and fiber — in the same power-constrained markets with record-low vacancy. Whether the capex flows through a hyperscaler's own construction arm or a developer's balance sheet, it converges on the same scarce inputs.
Related Whyte Consolidated research on the datacenter buildout and the capital forming around it:
- Whyte Consolidated — NVIDIA's $500 billion bet: turning compute into an asset class
- Whyte Consolidated — Schmidt told lawmakers: 1–10 GW data-center projects, scored sixteen months later
- Whyte Consolidated — The coming infrastructure economy: physical assets, programmable ownership, and the new capital stack
- Whyte Consolidated — When your bank becomes a datacenter
Prompted by CNBC's reporting (Kate Rooney) on hyperscaler spending strategies, August 2026. Market figures from hyperscaler earnings guidance and CBRE North America Data Center Trends as published on this site's market page. Characterizations of individual companies' strategies reflect public reporting and may evolve with future disclosures. For informational purposes only. Not investment, legal, tax or accounting advice.