Wind-Powered Hybrid Cooling Efficiency in the San Francisco Bay

Summary

A preliminary feasibility study proposes a hybrid cooling system for a 270 kW data center to be built on the water in the San Francisco Bay. The design utilizes ocean breezes as the primary heat sink, which is active 84.8% of the time. Variable-speed seawater pumps operate as a backup for 15.1% of the period, while a small chiller handles the remaining fraction. The system maintains a mechanical PUE between 1.08 and 1.12 while pumping approximately 41 times less water from the bay compared to a seawater-priority design.

Why it matters

Data center cooling is taking up an increasingly significant share of California's electricity and water budgets. The strict 2.2 K thermal discharge limit in the San Francisco Bay has already hindered seawater cooling proposals. A design that relies on the free ocean layer and only taps into seawater when breezes cease could enable coastal siting without triggering the regulatory and ecological objections that stalled previous concepts. The 63–67% savings in cooling energy compared to a traditional air-cooled chiller facility represents a significant difference at scale.

Highlights

• Breezes alone maintain the 21 °C setpoint 84.8% of the time. Seawater is used for 15.1% of the time, and the chiller for 0.14%.
• The hybrid system pumps approximately 27,000 m³ of water from the bay annually. In a seawater-priority loop, this figure would be approximately 1.13 million m³. Thus, the hybrid system uses 41 times less water.
• When the supply setpoint is raised to 24 °C within the ASHRAE range, pump runtime drops to approximately 4.4% of the time. However, full cold-aisle containment is required to keep rack inlet temperatures below 25.5 °C.

Critical take

This study is a single-author, non-peer-reviewed preprint prepared for a 270 kW room. The headline PUE value relies on a heat exchanger efficiency of approximately 0.75, as acknowledged by the author. If this value drops to a more typical 0.60, pump runtime triples. The required plate heat exchanger stacks are also eight times larger than in the initial draft, posing a practical space footprint issue. When scaling the module to 1–10 MW, this becomes a hard spatial constraint. The lack of 2022–2024 bay climate window data, marine ecology, and structural engineering analyses leaves the concept's real-world feasibility largely untested.