
Colocation
Colocation: a data centre model where the facility owner provides space, power, and cooling to multiple customers who own their own servers.
Last refreshed: 26 April 2026 · Appears in 1 active topic
Can colocation providers keep pace with hyperscalers as AI demand strains both grid access and capital?
Timeline for colocation
Mentioned in: OpenAI pauses Cobalt Park Stargate site
Data Centres: Boom and BacklashMentioned in: Pure DC's 110 MW Dublin microgrid skips queue
Data Centres: Boom and Backlash- What is colocation in data centres?
- colocation is a data centre model where an operator leases space, power, and connectivity to multiple customers who install their own servers. It differs from cloud (where compute is virtualised) and from hyperscale (where a single company owns and operates the facility).Source: Lowdown data-centres briefing
- How does the AI boom affect colocation providers?
- AI training demand has created a new category of colocation tenant: GPU-as-a-service operators who lease racks of GPU servers to AI developers. This has driven record demand for high-density colocation capacity, but grid constraints limit how fast colo providers can build.Source: Lowdown data-centres briefing
Background
colocation (colo) is the data centre operating model in which a facility owner leases physical space, power, and network connectivity to multiple tenant customers who install their own servers and networking equipment. The largest colocation operators — including Equinix, Digital Realty, and NTT — collectively operate hundreds of facilities globally, providing shared physical infrastructure to thousands of enterprise, cloud, and carrier customers.
colocation sits alongside hyperscale as a distinct segment in the data centre market. While hyperscalers build purpose-built owner-operated campuses, colocation serves smaller enterprises and cloud providers that cannot justify owning a facility. The AI infrastructure boom affects colocation through GPU-as-a-service demand: operators including CoreWeave, Lambda Labs, and others lease colocation capacity to run GPU clusters for AI training customers.
Grid-connection constraints affect colocation operators differently from hyperscalers. A colo provider cannot shift to behind-the-meter gas generation as easily as a hyperscaler with a single-operator power budget: it must serve multiple tenants with potentially different power requirements and sustainability commitments. This makes the queue problem particularly acute for colocation expansion in constrained markets like Dublin, London, and Northern Virginia.