Fish extract oxygen from water using gills — sheets of tissue only microns thick, folded to give a tuna roughly 1 m² of exchange surface per kilogram of body mass. A submerged city could scale up the trick: a tower packed with hollow-fiber membranes, pumping seawater past them and delivering breathable air to residents 300 meters down. No surface umbilical, no compressed-air cache, no plants. Just the ocean, filtered.

The oxygen budget. A resting adult metabolizes ~550 L of O₂ per day (STP). Seawater at 20 °C holds about 5.5 mL of dissolved O₂ per liter — roughly 1/30th what an equal volume of air carries. So per person, at a generous fish-like extraction efficiency of 80%:

(550,000 mL/day) / (5.5 mL/L × 0.8) ≈ 125,000 L seawater/day

That's 1.45 L/s per person. For a modest undersea city of 10,000, you need 14.5 m³/s flowing through the gill — roughly the discharge of the Thames at low water — pumped continuously, forever.

Membrane area. State-of-the-art medical ECMO oxygenators use hollow-fiber polymethylpentene bundles with O₂ flux around 0.03 mL·cm⁻²·s⁻¹ under favorable gradients. Total transfer needed:

10,000 people × 6.4 mL O₂/s = 64,000 mL/s → ~2.1 × 10⁶ cm² = 210 m² (ideal)

But boundary-layer resistance on the seawater side dominates. Real designs need 10–20× overhead, plus biofouling margins. Call it ~50,000 m² of active membrane — comparable to a football field's worth of fiber, easily housed in a 200 m tower packed with cartridges 3 m tall by 1 m wide.

Pumping power. Pushing 14.5 m³/s through fine-fiber cartridges at, say, 50 kPa pressure drop:

P = Q × ΔP = 14.5 × 50,000 = 725 kW

Add prefiltration, backflushing, and CO₂ stripping (the return leg has to expel exhaled carbon dioxide back into seawater — easier, since CO₂ is 30× more soluble than O₂), and total parasitic load lands near 2 MW. Manageable with an SMR or a hard tether to shore.

The nitrogen problem. Gills only harvest O₂. Humans breathe air that's 78% N₂ as a diluent — pure O₂ atmospheres are flammable death traps (see: Apollo 1). You'd need a closed-loop nitrogen reservoir, topped up occasionally from electrolyzed dissolved N₂ or delivered by supply capsule. Losses through airlocks are the real drain.

The killer: biofouling. A submerged membrane at 15 °C is a five-star hotel for barnacles, tunicates, and biofilm. Desalination plants replace RO membranes every 3–5 years despite chlorination. An artificial gill can't chlorinate — chlorine ruins the fibers and poisons the residents. Options: UV pretreatment (kilowatts more), copper-ion-loaded fiber coatings, or periodic thermal shock cycles. Expect membrane life measured in months, not years.

Vertical logistics. Build the gill as a slender 200 m tower rising through the habitat's core, with radial intakes at multiple depths (deeper water is colder → more dissolved O₂: ~9 mL/L at 4 °C, a 60% bonus). Waste-water plumes discharge laterally to avoid recirculation. The whole apparatus masses ~5,000 tonnes — a fraction of a real skyscraper.

Feasible? Physically, yes. Economically, only if the alternative — a snorkel to the surface — is unavailable.