Breeze First, Seawater Second: Hybrid Cooling for an Above-Water Offshore Data Center on San Francisco Bay Within California's Discharge Limit

Abstract

Coastal sites offer data centers two free heat sinks: cool marine air and seawater. Existing Bay-adjacent concepts pump water continuously, which has already drawn regulatory opposition over thermal discharge and marine-life entrainment. This screening-level feasibility study evaluates a hybrid, above-water design for a 270 kW server room on a pile-supported San Francisco Bay pier. An indirect air-to-air heat exchanger uses the marine-layer breeze as the primary heat sink, an air-to-seawater coil is pumped only when the breeze cannot hold the 21 °C supply-air setpoint, and a small chiller trims the remainder. A room-level CFD model (Ansys Icepak) sets the airflow and hot-aisle return temperature (33.25 °C); an hourly ε-NTU model of the two-stage heat-exchanger chain is then driven by three years of hourly Bay air data (2022–2024) and measured Bay water temperature; fan, pump and chiller power are then added to estimate annual energy and mechanical PUE against two baselines; finally, the air-to-air exchanger is sized from first principles to replace the assumed pressure drop.

The breeze alone holds the setpoint in 84.8% of hours, seawater is needed in 15.1%, and the chiller in 0.14%. With an assumed compact exchanger (200 Pa per side) the hybrid uses 245–275 MWh of cooling electricity per year (mechanical PUE 1.10–1.12), about 63–67% less than an air-cooled chiller plant. California’s Thermal Plan prohibits new cooling-water discharges into San Francisco Bay more than 2.2 K (4 °F) above the natural water temperature. A fixed-flow seawater loop breaks this limit, so the hybrid and a seawater-first baseline are both re-modelled with a variable-speed pump that holds the rise at 2.0 K. Under that rule, seawater first pumps about 1.13 million m³ of Bay water a year and uses 182–219 MWh (PUE about 1.08), depending on pipe friction. With the air-to-air exchanger sized as a plain aluminium plate stack (3 mm gap, 1.2 m plates, 67 Pa per side), the hybrid uses 187–217 MWh (PUE 1.08–1.09): about the same energy, while pumping about 27,000 m³, about 41 times less Bay water and heat. With commercial epoxy-coated counterflow cores (two per position, 84 Pa and 81% dry effectiveness in the maker’s selector), it uses about 182 MWh and pumps about 18,000 m³, about 60 times less. With the seawater coil selected by its manufacturer, which is less effective than assumed at low flow, it uses about 183 MWh, pumps about 21,000 m³ (about 50 times less) and needs a trim chiller about 41 hours a year. The plain-plate stacks are about eight times larger than first drawn, and the result depends on an exchanger effectiveness near 0.75; at 0.60, typical of some commercial cores, pump runtime more than triples. A commercial counterflow core checked in its maker’s selection tool reaches 81–85%, so 0.75 is a conservative assumption. Raising the supply setpoint to 24 °C, within ASHRAE’s recommended range, cuts pump runtime to about 4.4% of hours; room CFD confirms that, with full cold-aisle containment, rack inlets stay at or below 25.5 °C at that setpoint. Treating the room as a repeatable module, the percentages carry over to 1–10 MW facilities, where exchanger space, exhaust recirculation (each 1 K of intake warming adds 7–11 points of pump runtime) and total Bay withdrawals become the binding constraints. A screening cost estimate, using the maker’s quote for the cores (about $35,000 per module) and a published chiller cost, gives a simple payback against an air-cooled chiller plant of about 1.5–8 years at $0.125/kWh and 1–3.5 years at the California industrial tariff (about $0.25/kWh); the outdoor-air system and seawater loop are still placeholders.