An operating colocation facility can offer a useful starting point for AI infrastructure. It also brings customers, maintenance obligations and equipment that must keep working during a rebuild. Gorilla Technology's proposed US acquisition makes that tension worth examining.
In its October 9 announcement, Gorilla said it had signed a non-binding letter of intent for its first US data center acquisition. The company describes roughly 6 MW of current utility capacity and a path to 36 MW of total site power. It says the utility has provided written support for additional capacity, with an approximately 18-month delivery timetable subject to infrastructure works.
The acquisition has not closed. Gorilla targets completion by the end of November, subject to diligence, approvals and definitive agreements. Its review includes existing tenants and future direct liquid cooling. These are company-reported plans, rather than evidence of commissioned AI capacity.
For infrastructure buyers, the operational question is how a live facility can move through that transition without turning an expansion into a service incident.
Start with the obligations already inside the building
My assessment is that a brownfield AI project should begin with a tenant and dependency map. A floor plan alone will miss the constraints that determine when work can happen.
Ask which customers share electrical distribution, cooling systems, network paths and maintenance access. Identify where a change to an apparently unused area could affect an occupied hall. Record who approves an outage, how much notice is required and which changes must wait for a customer migration.
Commercial diligence and engineering diligence need to meet at this point. A lease may promise service continuity that a proposed construction sequence cannot yet deliver. Discovering that conflict before committing to a deployment date gives the project team a chance to redesign the work.
Define an AI phase around a usable operating envelope
Total site power is useful for describing a development opportunity. A deployment team needs a more specific boundary: the electrical and thermal conditions available to its racks, under both normal operation and the agreed failure scenarios.
A practical phase definition should identify the racks being released, their supported load, the cooling configuration and the equipment that remains shared with existing tenants. It should also state the assumptions behind redundancy. A rack that can operate only while every component is healthy has a different service envelope from one that can sustain a planned maintenance event.
For liquid cooling, procurement should follow an agreed interface specification. Confirm who owns the facility water connection, coolant distribution units, leak detection and emergency response. Assign responsibility for water quality and inspection records. Those boundaries should be settled before delivery teams arrive with equipment built to different assumptions.
Make construction isolation part of the acceptance plan
A live-site upgrade needs explicit separation between construction and operations. That includes access routes, dust control, temporary services and a process for approving changes to shared infrastructure.
The useful acceptance test is broader than switching on new racks. Demonstrate that the new phase can enter service while existing customers remain within their contracted operating conditions. Test the monitoring and escalation paths as well as the physical systems.
A construction incident should reach the operating team with enough information to act. A cooling alarm should identify affected assets and customer services. Define who can stop work, who can authorize a restart and what evidence must be captured before the phase is handed over.
Buy milestones with evidence attached
An infrastructure supplier evaluating this kind of project should ask for a phased delivery record. Separate the transaction, utility works, site upgrades and customer acceptance into distinct milestones. Give each milestone an owner, a dependency and an observable completion condition.
For a compute customer, that creates a clearer basis for contracting. Capacity reservations, equipment shipments and workload migration can follow the phase that is actually ready. For the operator, it exposes schedule conflicts early enough to adjust the plan.
Gorilla's proposal is a timely example of why existing facilities deserve a careful transition strategy. The engineering opportunity lies in connecting acquisition diligence to a defensible operating plan, so the first AI customers can arrive without compromising the customers already there.
Top comments (1)
tr.ee/dev-to