Case study · Commissioning
Confidential Hyperscale AI Campus
Engaged for a single power-module scope. Within seven weeks, running full commissioning oversight across all seven buildings of a 260 MW AI campus, scaling the team from four people to fourteen.
The challenge
A 260 MW campus running against the clock.
Seven buildings on one campus, built for high-density AI compute, with multiple builders, multiple equipment vendors, and fixed energization dates that did not move. At this scale the quality and documentation load alone can outrun any team.
What the owner needed, and did not yet have, was a single team that could see the whole campus at once, catch what was slipping past sign-off, and prove the building was actually ready before crews were put at risk or systems were energized.
Our role
One scope, then the whole campus.
Within seven weeks of mobilizing, the owner had expanded our scope to all seven buildings, five disciplines, and seventeen categories of deliverable: full commissioning documentation, independent auditing, peer review, root-cause analysis, troubleshooting, and short-circuit coordination oversight.
We scaled from four people to fourteen over those seven weeks, using an assessment-first ramp so every new person arrived against a known gap rather than a guess. The same core team held continuity across all seven buildings, which is what turns a problem found in one building into an early warning in the next.
Repetitive documentation work was automated so our engineers could spend their time in the field, witnessing the actual work and auditing the record against it.
Outcomes
What happened
- Stopped potentially fatal work. We identified incident energy above the 40 cal/cm² no-work threshold defined by NFPA 70E, peaking near 67 cal/cm², on equipment with no safety plan in place. We withheld work authorization and issued a formal request for information to the engineer of record until it was resolved.
- Audited the protective-device settings. We reviewed roughly 1,100 pages of short-circuit coordination studies covering about 2,300 arc-flash entries and surfaced more than 100 findings. We audited 1,164 protective-device positions and corrected 44 transformers, 132 settings in all, before a fixed medium-voltage energization date. Some of that equipment had already cleared earlier rounds of testing.
- Found $2.0M per year the design was leaving on the table. We built a from-scratch, physics-based thermodynamic model of a 10 MW cooling plant, an 8,760-hour annual simulation, that identified roughly $2.0M per year in utility-rate savings and about $755K per year for every 0.1 point of PUE improvement.
- Resolved issues roughly 4x faster. Across the campus we tracked about 7,700 quality issues and closed roughly 85 percent of them, cutting the median time to resolve a high-priority issue by about four times. More than 100 daily reports and weekly status came out of an automated data pipeline, not a backlog of meetings.
- Efficient asset tracking. Roughly 3,500 assets were tagged and tracked, giving the owner a living, queryable record of the build rather than a stack of files at the end.
What this proves
A second set of eyes is not a formality.
Every one of these findings was on a campus built by experienced contractors and already under commissioning. The problems were there regardless. Systematic, independent oversight finds them before energization and before turnover, while crews are still on site to fix them.