DataCentersX > Business Models

Data Center Business Models

The data center industry is structured around different business models that divide or integrate facility ownership, facility operation, IT hardware ownership, and workload execution. Understanding the distinctions between hyperscalers, colocation operators, REITs, AI neo-clouds, build-to-suit developers, and enterprise self-hosters is essential to grasp how compute infrastructure is financed, built, leased, and consumed. The same physical facility can be owned by one entity, operated by another, populated with hardware owned by a third, and host workloads run by a fourth. Sorting out who plays which role is the foundation for understanding the industry's economics and competitive dynamics.

The roles

Hyperscaler self-operators

Hyperscalers own and operate facilities, populate them with hardware they procure directly, and run cloud services for external tenants alongside their own workloads. AWS, Microsoft Azure, Google Cloud, and Meta dominate this category in the West; Alibaba, Tencent, and Baidu in China. The model is the most capital-intensive in the industry and the most vertically integrated. Hyperscalers also lease wholesale colocation space when build timelines for self-built capacity exceed demand growth, which makes them the largest customers of the wholesale colo category.

Colocation operators

Colocation operators own and operate facilities they lease to tenants who bring their own hardware. The model splits into two sub-types. Wholesale colocation leases large blocks (entire data halls or megawatt-scale capacity) to a small number of large tenants, predominantly hyperscalers and large enterprises. Retail colocation leases racks, cages, and partial cabinets to many tenants, with the value proposition centered on carrier neutrality, cross-connect density, and interconnection ecosystems.

REITs

Several major colocation operators are structured as Real Estate Investment Trusts, including Equinix, Digital Realty, and Iron Mountain. The REIT structure has tax advantages for facility ownership at scale and pushes operators to maximize lease revenue and asset values. The structure shapes capital allocation toward acquisition, leaseback, and development of new facilities, and away from operating businesses that REIT rules limit (such as direct workload hosting). Several large operators are not REITs, including CyrusOne (private under KKR), QTS (private under Blackstone), Compass Datacenters (private), and most build-to-suit developers.

Build-to-suit developers

Build-to-suit developers design and construct facilities to single-tenant specifications, typically under long-term lease arrangements with hyperscalers or large enterprises. The developer carries construction risk, completes the build, and hands operations to either the tenant or a third-party operator. The model has emerged as a significant share of new hyperscale capacity additions because hyperscalers cannot build fast enough internally and prefer leases to ownership in capacity-constrained markets.

AI neo-clouds

AI neo-clouds emerged as a new business model around 2023 to 2024 when GPU scarcity created a market for purpose-built AI infrastructure rented at hourly or term rates outside the major hyperscaler clouds. Neo-clouds own or lease facilities, procure GPU hardware (often via venture-debt-backed acquisition rounds), and rent compute capacity to AI startups, research labs, and enterprises that cannot get hyperscaler allocation. The model is capital-intensive on the hardware side, with debt financing typically secured against GPU inventory, and competes with hyperscaler AI services on price, allocation availability, and specialized hardware configurations.

Enterprise self-hosted

Enterprise self-hosted operators own facilities, operate them, populate them with hardware they procure, and run their own workloads. The model has shrunk substantially over the past two decades as cloud migration accelerated, but persists in industries where regulation, data sovereignty, latency, or cost favor on-premise compute. Banks, hospitals, utilities, telcos, government agencies, and large industrial enterprises remain significant self-hosted operators, often with hybrid IT architectures that combine on-premise and cloud workloads.

Tenants and consumers

The other side of the value chain consists of tenants and consumers who do not own or operate facilities. Cloud tenants run workloads on hyperscaler or neo-cloud infrastructure without touching physical hardware. Colocation tenants own their hardware but lease facility space, typically because they want operational control of their compute without the capital burden of owning a building. Most enterprises today operate as some combination of cloud tenant, colocation tenant, and enterprise self-hosted, with the mix shifting toward cloud over time but with meaningful exceptions in regulated and latency-sensitive industries.

Sovereign and government

Sovereign and government operators run facilities owned and operated by national governments or under sovereign cloud arrangements that meet specific national data residency and access requirements. US federal facilities under FedRAMP authorization, European sovereign clouds (Bleu, S3NS, Delos), Saudi PIF-backed AI infrastructure, and similar arrangements in China, India, and the Middle East represent this category. The model overlaps with hyperscaler and enterprise self-hosted depending on whether the sovereign operator builds its own facilities or contracts a hyperscaler partner under sovereign-region terms.

Strategic considerations

Capital intensity. Hyperscaler self-build and AI neo-cloud are the most capital-intensive models, with multi-billion-dollar facility and hardware commitments. Colocation REITs are capital-intensive on the facility side but offload IT hardware capex to tenants. Cloud tenants and colocation tenants have the lowest infrastructure capital requirements.

Vertical integration. Hyperscalers and enterprise self-hosted are the most vertically integrated, owning everything from facility to workload. AI neo-clouds are vertically integrated on hardware and workload but often lease facility. Colocation operators are deliberately horizontal, leaving hardware and workload to tenants.

Geopolitical concentration. US hyperscalers dominate global cloud market share, which has driven sovereign cloud responses in the EU, Middle East, and elsewhere. Chinese hyperscalers serve the domestic Chinese market under different regulatory and supply chain conditions. AI neo-clouds are heavily concentrated in the US, with European and Middle Eastern alternatives emerging.

Emerging dynamics. AI neo-clouds disrupted the traditional cloud-versus-colocation split by offering a third option: rented AI compute without the broader cloud platform. Behind-the-meter nuclear PPAs are creating a new sub-category of energy-integrated facilities. Sovereign AI strategies are creating a new sub-category of nationally-aligned infrastructure outside the hyperscaler footprint.

Where this fits

This page covers business roles in the data center value chain. It complements Types, which covers the physical and architectural constraint set defining each kind of facility, and Sites, which covers specific named deployments. The same physical facility (a Type) can be operated under different business models, and a single business model can host multiple types of facilities. Reading the three pillars together gives the complete picture of who owns what, builds what, and runs what.

Related coverage

Types | Hyperscaler DCs | Colocation DCs | Enterprise DCs | AI Factory | Sites | Workloads | GRC