An Intelligent Harvest greenhouse runs four closed loops — heat, water, roof, and soil — and offers an optional fifth, the Campus Catchment, that joins only where the operator signs the roof tie-in. Below is each loop’s official schematic, the verified language for what it does, and the math underneath it — with an honest line, every time, between what’s measured and what’s still a design target for a site we haven’t built yet. Worked examples use real constants and public U.S. climate data — no single site assumed.
“Server heat that would be vented is captured in a sealed loop and carried to the greenhouse — then returns, cooled, to warm again.”
A data center’s cooling problem is the greenhouse’s heating supply. Warm coolant leaving the servers crosses a plate heat exchanger — a stack of thin steel plates where the data-center loop and the greenhouse loop trade heat but never share a drop of water. The greenhouse side carries that warmth to the four zones and returns cooler, ready for the next load. Two fluids, sealed and separate, end to end.
A server is, thermodynamically, a heater that happens to compute. Essentially all the electrical power it draws leaves as heat — so a hall’s heat output is approximately its electrical load. The warmth is already there, already paid for, whether or not anyone catches it.
How much heat a flowing loop moves is fixed physics: mass flow, times the heat capacity of water, times the temperature it gains across the exchanger.
The heat the campus would throw away becomes the harvest the town eats. A Small facility supports about one flagship greenhouse module of reuse — and it’s a sliver of what the campus rejects; the rest of its heat is unchanged.
† Illustrative design targets, scaled from the flagship anchor (727,000 lb per IH-STD-003 / 3,000 MWh per module) by the same model the homepage calculator uses. Real and unconditional: energy conservation, cp = 4.186 kJ/kg·°C. Put on the record once metered: delivered heat and crop output.
The insulated supply-and-return main runs underground and trenchless between the hall and the greenhouse — bored in by horizontal directional drilling rather than cut across the site. A Gladiator G30 rig (≈ 29,000 lb of pullback, 147 hp) pulls the carrier pipe through undisturbed ground. That’s Dallas Rush’s world: moving energy and utilities under a site without tearing it open.
“Rain is captured off the sawtooth roof, the dirty first flush is diverted away, and clean water is stored, filtered, and returned to the four zones below — all one building.”
The same sawtooth roof that lets light in catches the rain that falls on it. The first flush — the dirty opening of a storm that rinses dust and debris off the roof — is diverted to stormwater and never enters the clean system. What’s left runs to a covered reservoir, is filtered and UV-treated, and waters the four zones from above. The reservoir is sealed; the loops never touch the stormwater.
Harvested volume is just roof area, times how much rain fell, times how much of it you actually collect.
Using each region’s NOAA / NCEI 1991–2020 rainfall normal (real, public), at C = 0.85:
| Region | Normal rain | Capture / 1,000 ft²·yr |
|---|---|---|
| Northeast | ≈ 44 in | ≈ 23,300 gal |
| Mid-Atlantic / Coastal | ≈ 46 in | ≈ 24,400 gal |
| Southeast | ≈ 52 in | ≈ 27,500 gal |
| Midwest | ≈ 39 in | ≈ 20,700 gal |
| Southwest | ≈ 13 in | ≈ 6,900 gal |
| Pacific Northwest | ≈ 40 in | ≈ 21,200 gal |
Rain that lands on the sawtooth roof is filtered, UV-treated, and waters the zones from above — never the water that cools the servers. The capture rate is fixed physics and public rainfall; the yearly total is simply that rate across your actual roof. A Small build carries about 108,900 ft² of catchment † across one flagship.
† The rate is real — roof geometry (0.623 gal per inch per ft²), a 0.85 smooth-roof collection coefficient, and representative regional rainfall from NOAA / NCEI 1991–2020 station normals (e.g. Boston, Philadelphia, Atlanta, Chicago, Salt Lake City, Seattle). A region spans a range; your actual site uses its own local normal. The yearly total rides on roof area: a flagship carries 108,900 ft² (2.5 acres) of catchment — a design target until a real roof is built and measured, and the only unit a published figure may use.
The covered reservoir is a tank sized to ride the dry spells between storms: volume = daily draw × days of autonomy. That buffer-tank discipline — specify it, set it, service it — is exactly what Cole Candler’s Gas Station Supply has done across Central Virginia for decades. The tank is what turns an inch of rain into water on demand.
“One sawtooth roof runs every market — it vents and shades through a Southern summer, seals and holds warmth through a Northern winter, and harvests a wet season, while the zones stay on setpoint.”
The roof isn’t passive — it’s the building’s main climate actuator, and the same design runs in Texas, Kentucky, or Virginia. In summer it opens its vents and draws a shade screen to dump solar gain; in winter it seals and draws a thermal screen to hold warmth in; in a wet season its gutters route rain into the water loop. The zones underneath stay on setpoint while the outside swings.
A drawn aluminized screen traps a still-air layer and reflects the greenhouse’s own longwave heat back to the canopy. Published horticultural ranges (UMass Extension) put the effect at a real, measured band:
Drawn in heat, the screen sheds a rated fraction of incoming solar before it ever loads the zones. Shade factor is a manufacturer-rated spec:
The same roof runs in Texas, Kentucky, or Virginia — it is the building’s main climate control. A Small build runs one module (≈30,000 ft² of actuated roof †). The screen percentages are proven horticultural ranges and don’t change with size — more building, the same self-reliance, less bought energy.
† Real: thermal-screen savings 30–50% (UMass Extension); shade factors 40–60% (manufacturer-rated). Roof catchment area is illustrative and scales with module count.
“Spent plants from the zones compost on-site in a sealed, aerobic, biofiltered vessel — no odor, no pests — and come back as soil for the next crop, with the surplus bagged for the town.”
When a crop finishes, its spent plants and trimmings don’t leave the site — they go into a sealed, aerobic, in-vessel composter. Enclosed means no open piles, no odor, no pests; the exhaust air is biofiltered and scrubbed, and the leachate is contained and recirculated — it never reaches the clean water pond. What comes out is finished compost that becomes the soil for the next crop. The surplus is bagged for the town. The loop closes on-site.
Composting is mostly a controlled exhale. Cornell’s compost chemistry: with each microbial pass, roughly two-thirds of the carbon leaves as CO₂ and the rest stays as stable humus; the C/N ratio falls from about 30:1 toward 10–15:1. The result is a real, repeatable yield band:
The fraction is real. The absolute tonnage — pounds of trimmings, bags for the town — rides entirely on crop throughput, which only exists once a real greenhouse is running. So it stays a design target. We’d rather show you the honest mechanism than a number we can’t yet stand behind.
Every finished crop becomes the soil for the next, and the surplus is bagged for the town — across one module on a Small build. The yield fraction is real and repeatable. The absolute tonnage rides entirely on crop throughput, so — by deliberate choice — we won’t print a pound count we can’t yet stand behind.
† Real: compost yield 30–50% of input mass; ~⅔ of carbon respired as CO₂ per pass (Cornell). The tonnage of compost bagged for the town stays a design target until a real greenhouse is running and weighed.
“In-vessel” means exactly that: a sealed tank, contained and serviceable — the same tank discipline Cole Candler’s team brings to the water and heat buffers. The vessel is what makes “no odor, no pests, nothing trucked away” a design spec instead of a hope.
“Rain the campus already sheds is intercepted at grade — downstream of the operator’s code-required drainage — banked in a covered cell, and metered back: the first fill, the hottest hundred hours, and the fire reserve. If our side ever fails, the water simply goes where the permit drawings always sent it.”
This loop is the one that is offered, never required. The four loops above make a greenhouse self-sustaining on water — catchment exceeding worst-case make-up with 1.4× cover in a stress-test dry year †, reported net on the third-party meter. That is the floor, and it stands alone: if a town or an operator prefers the greenhouse simply sustain itself, that is the whole deal, and the gauge still gets published.
Where the operator signs the roof tie-in — a drainage easement and runoff license, an exhibit to the ground lease — the fifth loop switches on. One precast vault in their storm line, at grade, outside their fence: a first-flush diverter, a screen, and a meter. Clean flow banks in the covered wet cell; a one-way, air-gapped line supplies the data center’s commissioning fill (≈400,000 gallons per building, the operator’s own public figure), its hottest-100-hours misting water, and a dedicated fire reserve that no drought is allowed to borrow. Heat crosses the fence one way; water crosses back the other; both meters are published.
The condition is printed, not implied: greenhouse roofs alone cannot fund these promises — the canonical math shows the worst dry-year corner goes negative at campus scale without the data-center roofs, and clears only marginally even at the smallest build. So the tie-in is a condition of the offer at every size, and the design ships two cells — a covered wet cell that wants to be full, and a dry detention cell that must sit empty — because one pond cannot serve both masters. Anything that will ever be misted passes filtration, UV, and residual treatment first.
Four loops, one building — and a fifth on offer where the fence line says yes. The constants are physics; the magnitudes wait for a site. That’s the honest line — and we’ll hold it all the way to the first harvest.
Loop schematics are the official Intelligent Harvest build-hub documents. Worked examples use real constants and public climate data; every site-specific magnitude is marked † as a design target until a site is built and measured.