A regional cloud provider needed compute placed inside eleven metropolitan sites to hold application latency under five milliseconds, and it needed those sites live within a single financial year. Building conventional plant rooms would have consumed most of that year in each location. The delivered answer was a prefabricated micro power and cooling module dropped into leased shell space, and the measured outcome across the first four sites was an annualised power usage effectiveness of 1.38 against a 1.92 baseline at the operator's older edge rooms, with each site energised roughly ten weeks after the order was placed.
Edge sites punish the assumptions that work in a large data hall. Each room held between four and eight racks, an average of 6 kW per rack and no on-site staff. The existing estate used tower uninterruptible supplies fitted years earlier for a much smaller load, wall-mounted split air conditioners fighting one another across an uncontained room, and distribution assembled incrementally by whoever was on site that week. Two consequences followed. Overhead energy exceeded the energy actually reaching the servers, and a single failure took the room down because nothing had redundancy that could be serviced without an outage.
The commercial pressure was equally blunt. Each week of delay in bringing a metropolitan site online postponed the revenue that justified it, and eleven bespoke designs would have needed eleven sets of drawings, approvals and contractor mobilisations.
The decisive argument was repeatability rather than novelty. A factory-built module fixes the design once. Power train, cooling, containment, distribution, monitoring and cabling are assembled and tested on a production line, so what arrives on site is a known object that has already been powered up, load-banked and signed off. Site work reduces to positioning, connecting the incoming supply, connecting condenser pipework and commissioning.
That shifts most of the programme risk from the site to the factory, where it is far cheaper to manage. It also means the eleventh installation benefits from every lesson learned on the first ten, which is never true of eleven separately engineered rooms.
Each module comprises six racks in a cold aisle containment arrangement, in-row cooling units sized for the enclosed volume rather than the whole room, and a modular uninterruptible supply configured N+1 from 25 kW power modules. Distribution runs through busway to intelligent rack power strips with per-outlet metering. Environmental sensing, leak detection, door access and power telemetry all report to a single controller that feeds the operator's existing monitoring platform.
Two details did most of the work on the efficiency figure. Containment stopped hot exhaust air from recirculating into cold intakes, which allowed the supply air temperature to be raised from eighteen degrees Celsius to twenty-four within the equipment manufacturers' own specifications. Sizing the uninterruptible supply in increments meant it operated between forty and sixty percent of installed capacity, close to the peak of its efficiency curve, rather than at the twelve to eighteen percent load where the legacy tower units had been idling.
Across the first four commissioned sites, annualised power usage effectiveness settled at 1.38, compared with 1.92 measured on the operator's older edge rooms carrying comparable load. For a site drawing 40 kW of information technology load, that difference removes about 21,600 kilowatt-hours of overhead consumption per site each year. Deployment time from purchase order to live load averaged ten weeks against the twenty-six weeks the operator had budgeted from experience with conventional fit-outs. Unplanned downtime attributable to power or cooling across the first nine months of operation was zero, against four incidents in the preceding nine months on the legacy estate.
Two lessons are worth recording. First, condenser placement drove the site survey more than the module footprint did, and on two sites the roof route added a fortnight that better early information would have removed. Second, the operator initially specified fifteen minutes of battery autonomy at every location. Reviewing the actual generator arrangements site by site showed that five locations needed only five minutes to cover a start sequence, and reallocating that battery investment to the four genuinely unsupported sites improved resilience without increasing spend.
Prefabricated modules make most sense where the same requirement repeats across several locations, where programme time carries real commercial value, and where no permanent technical staff will be present. A single one-off room inside an existing well-run facility is often better served by conventional equipment. The way to tell them apart is to count the sites and price the weeks.
If you are planning an edge rollout and want a like-for-like comparison against a conventional fit-out, send the rack count, the load per rack and the target availability for each location. Our prefabricated deployment options are outlined at https://www.upsboss.com/micro-module/, and the uninterruptible supply and power distribution ranges used in these builds are listed at https://www.upsboss.com/products/.
Key takeaway: the efficiency gain at the edge came from containment and from loading the uninterruptible supply where its efficiency curve is highest, while the programme gain came from building the same tested module eleven times instead of engineering eleven rooms. Repeatability, not exotic technology, produced both numbers.
How small can a prefabricated power module usefully be?
Around three racks is the practical floor. Below that the fixed cost of containment, in-row cooling and monitoring is hard to justify, and a well-specified single rack with a rackmount uninterruptible supply usually serves better.
Does a module lock the operator into one supplier?
Not for the information technology equipment, which mounts on standard nineteen-inch rails. The power and cooling components inside the module are matched as a system, so mid-life substitution of those parts should be planned with the original designer.
What maintenance does an unstaffed edge site require?
Typically two attended visits a year covering filter replacement, thermographic inspection of connections and a battery capacity check, supported by continuous remote telemetry with alarms routed to the operator's existing platform.
Contact: Frank Zhang
Phone: +86-135 5688 8641
Email: frank@upsboss.com
Add: Jufeng Road, Guangming Street, Guangming District, Shenzhen City, Guangdong Province, China
