Executive Summary

• Robert Berliński, General Manager at BCA Engineering Group, looks at why the data centre boom demands a different commissioning mindset from the legacy era.
• The commissioning model that worked for 10 or 20 MW facilities was designed for a different generation of projects. As campuses grow in scale and delivery schedules continue to compress, planning for commissioning must be considered from day one and AI campuses and high-density deployments and 70 MW+ facilities require a fundamentally different approach.
• Designing effective commissioning scenarios requires more than knowledge of commissioning procedures. There must be an understanding of electrical infrastructure, control philosophy, GPU power behaviour, liquid cooling and the interaction between all major building systems.

The data centre industry has a commissioning problem. Not because projects are tested too little or standards are inadequate. Many commissioning teams and strategies still reflect delivery models developed for facilities that were significantly smaller and delivered sequentially. AI campuses and high-density deployments and 70 MW+ facilities require a fundamentally different approach.

The AI race has fundamentally changed project economics. Every month that a GPU campus is delayed represents millions in unrealised computing capacity. Delivery speed has therefore become an engineering parameter, not simply a commercial objective.

The cost of failure reflects this reality. According to Uptime Institute’s Annual Outage Analysis 2026, 57% of respondents reported their latest major outage cost more than $100,000, while one in five exceeded $1 million.

The industry does not need faster commissioning. It needs a different commissioning model. The challenge is no longer performing more tests but designing a strategy that allows multiple systems to be verified simultaneously without compromising quality.

Complexity has moved from systems to scenarios

The biggest shift is the pace at which customers expect the infrastructure to be delivered. If we need to go from the first power-on to the completion of the IST within weeks rather than months, it’s not enough to simply accelerate the same activities. The entire delivery model must change.

But it does not mean reducing tests, compromising quality or bypassing procedures. Modern commissioning validates scenarios rather than isolated assets. Designing these is significantly more challenging. It requires construction sequencing, energisation strategy, controls integration, vendor availability, engineering resources and temporary infrastructure to converge at exactly the right moment.

A single load test should simultaneously verify UPS performance, switchgear, busbars, cooling response, BMS, EPMS and control logic instead of recreating the same operating conditions multiple times for different disciplines.

The complexity has shifted from executing individual tests to orchestrating an entire project. That orchestration must begin during project planning, not once construction is underway. Energisation strategy, construction sequencing, resource planning, test scripts must be developed together from the earliest project stages, aligned with parallel work execution. Without it it’s impossible to meet the schedules that AI infrastructure providers now expect.

Liquid cooling is rewriting the playbook

Liquid cooling illustrates why traditional commissioning models are being challenged. It changes operating philosophy, system interactions and commissioning scenarios. From our experience on liquid-cooled hyperscale projects in Finland, the engineering principles remain unchanged, but new interactions between power, cooling and control systems require different approach and a much deeper understanding of system behaviour under real operating conditions.

Standards define the objective. Projects define the strategy.

The data centre industry relies on well-established commissioning standards for good reasons. They define what should be verified. However, they do not explain how to commission a 100 MW AI campus delivered under compressed schedules. That requires engineering judgement, understanding the system architecture, and the ability to adapt strategies to each project.

No standard anticipates late design changes, evolving client requirements, contractor interfaces or project-specific constraints. These are engineering challenges that must be managed throughout delivery, not discovered during Integrated Systems Testing.

Designing effective commissioning scenarios therefore requires more than knowledge of commissioning procedures. Engineers must understand electrical infrastructure, control philosophy, GPU power behaviour, liquid cooling and the interaction between all major building systems. Only then can engineering teams design integrated test scenarios that validate multiple systems simultaneously while maintaining the level of confidence required for mission-critical facilities.

The commissioning model that worked for 10 or 20 MW facilities was designed for a different generation of projects. As campuses grow in scale and delivery schedules continue to compress, the industry must rethink not only how it tests infrastructure, but how it plans commissioning from day one.

Commissioning is becoming a project planning discipline. The next generation of AI data centres will not be delivered by organisations that simply execute commissioning more efficiently. They will be delivered by organisations capable of redesigning the entire delivery model around integrated engineering, parallel execution and intelligent validation from day one.