Several businesses can share one AI data centre. One company may own the land and building and lease space and power. Another may run the facility and rent space to many customers. A cloud company may fill the halls with AI chips and sell computing time. An AI company may buy that time to train and run its models. Each layer earns in its own way, through rent, service fees or computing charges, and each carries its own risks.
The layers
| Layer | What it provides | How it typically earns |
|---|---|---|
| Building owner or developer | Land, building, power connection and cooling | Long leases of space and power to large customers |
| Colocation operator | Space, power, cooling and connections shared by many customers | Monthly recurring charges and connection fees |
| GPU cloud provider | AI chips, networks and software in leased space | Committed multi-year contracts, plus pay-as-you-go use |
| AI company | Models and services for end users | Subscriptions, usage charges or advertising |
How some listed companies describe their revenue
Equinix, a colocation operator, reports that its customers are generally billed on a fixed and recurring basis each month for contracts generally one to five years long, and that recurring revenues have been more than 90% of its total revenues over the past three years. In 2025 it reported total revenues of US$9,217 million, of which colocation was US$6,475 million and interconnection US$1,655 million.
Digital Realty describes two kinds of deployment: smaller ones of up to 1 MW, with contracts generally 2 to 5 years, and larger ones above 1 MW, with contracts generally 5 to 10 or more years. It reported total operating revenues of about US$6.1 billion for 2025, mostly rental and related services.
CoreWeave, a GPU cloud provider, says customers generally buy capacity through multi-year committed contracts on a take-or-pay basis, and that committed contracts were over 98% of its revenue in 2025. It also reports that it leases space in third-party data centres, so it is itself a customer of the layer below.
Where the risks sit
Each layer's risk follows its contracts. A building owner with long leases depends on its tenants staying solvent. A colocation operator depends on keeping many customers. A GPU cloud provider owns fast-ageing equipment and may depend on a few large customers: CoreWeave reports that about 67% of its 2025 revenue came from its largest customer.
What is not certain
- Company figures are as each company reports them in its filings; business models change, and past results say nothing about future ones.
- Many AI data centres are owned and operated by a single large company, so the layers are not always separate businesses.
Sources and method
- Equinix, Inc., Annual Report on Form 10-K for the year ended 31 December 2025, filed with the US Securities and Exchange Commission. https://www.sec.gov/Archives/edgar/data/1101239/000110123926000032/eqix-20251231.htm
- Digital Realty Trust, Inc., Annual Report on Form 10-K for the year ended 31 December 2025. https://www.sec.gov/Archives/edgar/data/1297996/000110465926015365/dlr-20251231x10k.htm
- CoreWeave, Inc., Annual Report on Form 10-K for the year ended 31 December 2025. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000104/crwv-20251231.htm
Method: each link was opened on 29 October 2026 and every figure was taken from the page or filing cited beside it, in the currency stated there. No figure was estimated. Company examples describe what those companies report about themselves; they are not recommendations.
Read next
This article provides general information, not investment advice or an offer of securities.