Philadelphia
A liquid-cooled data center built with its host,
returning its heat to the city.
The Presentation
The case, and the options on the table for
everyone in the room.
Three candidate sites.
One we build first.
We are pitching three candidate parcels. The River Site is our recommended anchor for 80 MW and the one we deep-dive next. None is locked; siting follows the power study.
The River Site
The property, and how we propose to build it.
Why the river parcel.
- River frontage, power on or beside the parcel, a heat sink next door.
- Room for a three-building campus, and the clearest reference for the design options.
[Parcel specifics to confirm: area, control, power and heat-sink distances.]
Three buildings.
A data center, a plant, and a public center with a greenhouse. Heat runs from the data center, through the plant, to the public center.
The Data Center.
- 80 MW. Five storey, ground plus four data floors of 20 MW each.
- Direct-to-chip liquid cooling, CDUs at the boundary, cores and risers stacked.
- Behind-the-meter power in. Tier III, concurrently maintainable, N+1.
The Public Center.
- Two storey. Community, office, and a craft brewery, with a greenhouse and a public forecourt.
- It receives recovered hot water from the plant.
- Built here, in the design. Programmed in Part Three, the community track.
Power, built three to make two.
Three buses, sized so any two carry the full load. Dual-corded, 800 V ready, and behind the partner meter.
- A 3M2 distributed redundant topology, Tier III and concurrently maintainable, so any one path can drop for service without touching the load.
- Total facility electrical load of 88 to 92 MW on a 100 MVA service, with a behind-the-meter power cost target below $90/MWh.
- Up to 20 MW drawn behind the partner meter, the balance on a new 60 to 80 MW front-of-meter interconnection at PECO 230 kV into PJM.
Where the loops meet.
The CDU gallery is the boundary between the chip loop and the facility loop. Eleven units at N+1, on a reverse return for balance.
- Direct-to-chip liquid cooling, with the CDU sitting at the boundary so the chip loop and the facility loop never mix.
- DN450 reverse-return headers carry about 355 L/s at a 14 K delta-T on a 25% propylene glycol facility loop, balanced by design.
- No mechanical refrigeration, so the hall runs at a PUE of 1.10 to 1.15 with ERE below 1.0 once the partner takes the heat.
Three loops, one boundary.
Chip, facility, and recovery loops meet at the CDU, so no central separation heat exchanger is needed.
- Recovered heat leaves at 40 to 65 °C (104 to 149 °F), metered across the host boundary before it reaches the partner.
- The dry-cooler reject path sits on the always-open leg, sized for the full load, so IT cooling never depends on heat offtake.
- 20 to 22 MW thermal returned to the district system, displacing about 660,000 Mcf/yr of gas for the host.
Where the heat is measured.
Warm water off the rejection pumps passes a recovery heat exchanger metered to EN 1434, the point where recovered heat is counted before it reaches the partner. The dry coolers take the balance, so rejection is always available.
20 MW on one floor.
A single storey of 893 m² (9,610 ft²). Two halls in the centre, power and battery to the west, plant to the east, with the dry coolers carried outside.
- Ground plus four identical 20 MW floors stack to the full 80 MW, so the block grows by storey rather than by redesign.
- Cores and risers stacked, CDUs at the boundary of each hall, dual-corded down to the floor distribution.
- The dry coolers are carried outside the envelope, keeping the reject path independent of the occupied floor.
The Community
Separate from the build.
Themed on energy and heat recovery.
The data center stands on its own. The community work is its own commitment, connected only by the hot water that leaves the plant.
Heat becomes food.
Recovered heat runs a controlled environment greenhouse at real scale. Shown here against a football field for size, and as the growing array the warm water can support.
