Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesWhen DuPont Fabros Technology (DFT) opened ACC7 in Ashburn, Virginia, in 2014, it introduced its largest facility and a new design intended to bring hyperscale efficiency to wholesale, multi-tenant data centers. The 446,000-gross-square-foot building was planned for 41.6 MW of critical power at full development, with 28 computer rooms. Its lessons were practical rather than theoretical: medium-voltage distribution could reduce feeder materials, water-side economization could limit chiller use, and slab-floor rooms could give customers more flexible layouts—but each gain required new safety, commissioning, airflow, and operating disciplines.
What did DuPont Fabros learn from building ACC7?
DFT’s 2014–2015 accounts identify four connected design choices:
- 4,160-volt medium-voltage distribution.
- Medium-voltage isolated-parallel UPS topology.
- A cooling plant centered on water-side economization, with chillers for warmer conditions.
- Slab-floor computing rooms with overhead cabling, containment, and internal customer segregation instead of raised floors.
The objective was to lower construction and operating costs without sacrificing redundant power, cooling resilience, or the ability to serve customers with different densities and deployment sizes. Scott Davis summarized the governing principle as: “We never save at the cost of reliability or resiliency.”
ACC7 was the first facility to fully use this design. DFT described each base room as approximately 8,500 gross square feet, with flexible critical-load planning in the roughly 1.0–2.0 MW range and room for about 378 standard cabinets. Those figures describe planned capacity and options, not a claim that every room or tenant used the same configuration.
Recommended Free Tools
#1 Best Overall
Why did ACC7 use medium-voltage power?
DFT moved distribution from the more common 480 volts to 4,160 volts. Higher voltage allows the same power transfer with less current, which can reduce conductor quantity, duct-bank size, and space consumed by feeders.
The cited feeder comparison
In a 2015 presentation reported by Data Center Knowledge, Scott Davis compared a 2,500-kVA duct-bank example. A 480-volt design required about 2,700 feet of copper across eight sets of four-wire ducts. ACC7’s 4,160-volt arrangement used about 180 feet of shielded cable in one three-wire duct. Davis said the example reduced feeder-wire length by a factor of seven and reduced duct-bank material and space requirements. This is a project-specific comparison, not a universal result for every medium-voltage installation.
Operational and efficiency claims
The same account cites fewer terminations, cooler-running duct banks, and longer feeders that gave the layout more flexibility. It reports 99.6% efficiency for the medium-voltage PDUs. DFT’s opening announcement also reported oil-filled medium-voltage PDUs and a calculated annualized PUE of 1.15.
Rank #2
The trade-offs
Medium voltage transferred complexity rather than eliminating it. The 2015 account describes a smaller equipment market, less workforce familiarity, labor-intensive cable handling and termination, tight termination spaces, higher electrocution hazards, and more complicated relay protection. A viable design therefore depends on qualified electricians, careful protection coordination, controlled construction procedures, and maintenance plans suited to the equipment. The lower material count is valuable only when these safety and reliability requirements are funded and managed.
How did ACC7 keep cooling efficient without relying on chillers all year?
ACC7 used water-side economization with chiller assistance. Heat exchangers could use favorable outdoor conditions to provide chilled-water production without running mechanical refrigeration continuously; chillers supplied additional cooling during warmer conditions.
Economizer expectations and reported operating range
DFT’s February 2014 design report expected a plate-and-frame heat exchanger to provide the primary cooling source for about 75% of the calendar year. A later 2015 account describes heat exchangers supplying 65–70°F water year-round, with mechanical chillers when required. The 75% figure was an expectation in the planning report, not a measured lifetime result.
Plant equipment and reclaimed water
DFT’s September 2014 opening announcement described an evaporative cooling plant using reclaimed water, 12 centrifugal chillers, heat-exchanger lineups rated at 1,400 tons each, and an 80,000-gallon chilled-water storage tank. These are company-reported specifications at opening.
Airflow discipline
The cooling strategy depended on containment, not just plant efficiency. Cold air was directed to server inlets while hot exhaust returned through chimney racks or contained hot aisles. Davis said containment was mandatory to prevent hot and cold air from mixing and to pursue a low PUE. The feature also reported CRAH fan power 60% lower than traditional units when warmer server operating temperatures were used. That reduction required disciplined hot/cold-aisle management; a leaky or poorly managed room would give back much of the benefit.
What were the trade-offs of eliminating raised floors?
ACC7 used slab floors and ran network and power cabling overhead. Power-distribution equipment stayed outside the rooms, leaving more floor area for customer IT equipment. This simplified a room layout that could be divided and fitted out in stages, but it made overhead pathways, cable management, and containment central to operations.
Flexible room deployment
DFT planned to build the full shell while fitting out computer rooms in smaller phases as demand arrived. Rooms could be subdivided, support variable densities, and—where appropriate—use server containers. These were design options intended to meet different customer requirements, not proof that every tenant received every option.
Customer separation inside wholesale space
Steel mesh fencing allowed equipment to be segregated within a single room. Davis linked this feature to stricter audits among customers serving privacy-sensitive industries, including healthcare, and to smaller deployments seeking wholesale space. He said DFT was seeing requests in the 100–200 kW range after a historical minimum deal of about 500 kW. Fencing can create physical separation, but it does not by itself satisfy a particular audit or regulatory standard; access control, procedures, monitoring, and contractual controls remain necessary.
How was the design tested before customer deployment?
DFT built a proof-of-concept room of about 8,500 square feet containing more than 200 cabinets. The 2015 report says the room included more than 20 cabinet and containment designs from eight vendors. Load banks simulated electrical and heat loads up to 15 kW per cabinet across different rack dimensions and containment arrangements.
Davis described the purpose this way: “It gives us a comfort level to tell our customers that we have designed it, and proven that it works.” The exercise was an engineering validation by DFT, not an independent certification or a long-term operating study. It demonstrated that the proposed combinations could be evaluated under representative loads; it did not establish later whole-facility performance for every tenant configuration.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What do the PUE figures actually show?
| Figure | Status and date | What it means |
|---|---|---|
| Below 1.14 at 75% capacity | Forecast reported in February 2014 | Planning expectation, not a later measured result. |
| Below 1.13 at full utilization | Forecast reported in February 2014 | Projected annualized value at full use. |
| 1.15 annualized PUE | Company-calculated figure in the September 2014 opening announcement | DFT’s stated calculation at opening, not an independent benchmark. |
| Approximately 1.15 expected | Reported in the February 2015 testing coverage | Expectation associated with the new design. |
| 1.28 at earlier Ashburn facilities | Historical comparison in the February 2015 report | Reported comparison with prior DFT facilities. |
PUE varies with IT load, weather, water systems, maintenance state, and measurement boundaries. The cited material does not establish ACC7’s current operator, current configuration, or independently measured long-term PUE.
Which design lessons transfer to another data center?
ACC7 is best treated as a case study and a decision framework, not a universal blueprint. A comparable project should evaluate:
- Delivered critical capacity: how much usable IT power remains after redundancy and losses.
- Capital cost per megawatt: whether savings in conductors or plant equipment outweigh specialized construction.
- Feeder materials and length: voltage, route distance, duct-bank space, and termination count.
- Maintainability and skills: availability of workers trained for medium-voltage equipment and controls.
- Protection and safety: relay coordination, arc-flash exposure, switching procedures, and clearances.
- Cooling energy and water: economizer hours, chiller performance, reclaimed-water availability, and local water constraints.
- Operating density: rack dimensions, heat loads, containment choices, and airflow controls.
- Customer segregation: physical separation and the operational controls needed for audits.
- Phased expansion: shell-first construction and room-by-room fit-out without stranding capacity.
- Measured PUE: values tied to a stated load, weather conditions, boundary, and measurement method.
DFT’s broader lesson was to integrate electrical, mechanical, and customer-layout decisions early. Medium voltage affected protection and staffing; slab floors affected cabling and airflow; economization affected containment and water management. The design worked as a system, not as a collection of independent efficiency tricks.
What ACC7’s experience does—and does not—prove
ACC7 shows how a wholesale data center can pursue hyperscale-style efficiency while retaining redundant infrastructure and tenant flexibility. The reported feeder comparison, economizer strategy, containment requirements, and proof-of-concept testing explain the engineering logic behind the facility.
It does not prove that every project should adopt 4,160-volt distribution, that reclaimed-water economization is suitable in every climate, or that a quoted PUE will recur under all loads. The available accounts are company announcements and trade-press reports from 2014–2015; no independent source in this record verifies long-term operating performance.
