For engineers

You’ve heard “waste-heat reuse” pitched before. Here’s the math behind ours.

No black boxes. This is the method (Q = U·A·ΔT), the per-zone loads, the source-temperature reality, and — the part most decks skip — the open items we deliberately don’t paper over. The bench closing them is assembled: liquid cooling from the chip out, filtration and fertigation, and four decades of ground-source thermal design. We show the math anyway — a project whose pitch is “verifiable benefit” should be the easiest one to check.

Scroll
Straight to the depth
For the non-engineering reader

Here’s the whole page in plain terms: a data center makes an enormous amount of heat and throws it away. A greenhouse next door can use that exact kind of low, gentle warmth to grow food all year — and the match is easy, because a greenhouse wants heat that’s too weak to be useful to almost anyone else. Everything below is us showing our math to the engineers and naming the few things our engineering bench pins down site by site. None of it changes how the data center runs.

01 Why most waste-heat pitches don’t pencil

The offtaker usually wants what the source can’t give.

District heating wants 60–70 °C; air-cooled exhaust strains to deliver it; the load sits miles away; and a heat pump quietly becomes the hero and eats the economics. You know the failure modes. A greenhouse inverts every one of them — it’s the rare offtaker whose demand matches what a data center most easily rejects.

The grade match, stated plainly

Comfortable growing air sits at 18–27 °C; the real heating happens through low-temperature under-bench and floor loops. Captured reject water at 35–50 °C clears every zone setpoint before any lift in all but the coldest hours. And supply isn’t the constraint: a data center yields ~0.69–0.86 MWh of reusable heat per MWh of IT load, and at flagship scale the greenhouse pulls a single-digit-percent slice. The recovery gear stays small because the source always rejects far more than the greenhouse can use.1

02 The loop, and the two non-negotiables

Cooling-agnostic. It never touches your servers.

A sealed secondary loop draws low-grade heat off the facility’s existing reject side through a plate exchanger, buffers it around the clock, and distributes it at 95–122 °F to four growing zones, with a cooled return to the plant. It’s additive to whatever cooling you run — air, liquid, immersion — and changes nothing about how the data center operates, fails over, or meets SLAs.

Full isolation
A plate heat exchanger fully separates the data-center loop. The farm taps a bypassable slipstream upstream of the chillers, which always have the last word on temperature.
3rd-party BTU
Delivered heat is metered by someone other than us — the benefit is auditable, not promised. The entire thesis verified in one sensor.
Worst case = today
If the farm can’t take heat, a bypass routes it straight to the chillers and the plant cools exactly as it does now.
The grade of the heat
A greenhouse wants exactly what a data center throws away.
District heating needs 60–70°C and strains to get it — which is why reuse stalls. A greenhouse’s comfort line sits at 27°C. Pick a cooling type and see where its reject heat lands.
20°40°60°80° GREENHOUSE comfort 18–27°C district heating wants ≥60° REJECT HEAT

Water-cooled reject ~50–60°C. Clears the greenhouse comfort line with room to spare — no lift needed. The low grade that’s a deal-breaker for district heating is the whole point here.

Sourced Reject-temperature ranges from an ACEEE analysis of data-center cooling; greenhouse comfort range is standard horticultural practice.

03 The part most decks skip

The open items — and whose bench they’re on.

The research brief proves supply dwarfs demand across five climates. What a desk can’t close — and what site engineering exists to settle — is the short list below. We state it up front, because a project whose pitch is “verifiable benefit” has to be the first to name what isn’t yet verified — and each item now has a name beside it.

1
The source-temperature floor. Published reject heat runs 50–60 °C water-cooled, but real direct-to-chip loops vary and the exchanger charges 2–5 °C of approach. The contractual minimum supply temperature — and the firm heat-pump size it dictates — is the first calculation on every site. That’s Curtis Breville’s side of the exchanger: the chip-side loop is his home discipline.
2
Delivery at low ΔT. Q = U·A·ΔT sizes what the building loses; it doesn’t prove the loops can emit that load at crop-safe surface temperatures. First pass: four of five climates land within radiant’s reach; the coldest needs supplemental low-temp air handling. The per-zone emitter schedule is Ed Lohrenz’s scope — low-ΔT delivery is forty years of his practice.
3
Buffer sizing. “24/7 store” needs a number. First pass: ~two hours of design-peak draw. Ed validates the volume and stratification against a real diurnal charge/discharge profile — with Cole Candler on the tank itself.

The heat pump stays small throughout — but its adjective is firm, not optional: sized in steel to carry the warmest zone (turmeric & ginger, 82 °F) at whatever floor the host contracts. Every component is off-the-shelf; the value the bench adds is integration, reliability, and a stamp.

For the non-engineering reader

In summer the greenhouse doesn’t need much heat — but the data center is still making plenty. Rather than waste it, that summer heat can run equipment that makes cool, dry air for the plants, and even pulls clean water out of the air as a bonus. The point: the system earns its keep all twelve months, not just the cold ones.

A fourth lever — turning summer into a use

Summer is the one stretch the greenhouse can’t take heat for warmth — exactly when the campus rejects the most, and when the BTU meter would otherwise read near zero. Two low-grade-driven processes close that gap. A silica-gel/water adsorption chiller runs on water-cooled reject heat at roughly 50–61 °C (data-center-validated; COP ~0.3–0.5 at this grade, ~18–22 °C chilled water), and a liquid-desiccant stage regenerates at ~45–70 °C, attacks vapor-pressure deficit directly, and condenses fresh water as a by-product — the same desiccant path the arid config already leans on. Both keep the meter live year-round. Sizing and COP get stamped per site; we name it as a design direction, not a built figure.

And it barely touches the grid

By design the greenhouse is a small electrical load: it heats on the data center’s reject water rather than resistance heat, leans on low- and no-light crops (gourmet mushrooms need none), and shifts what little load it has off-peak behind the buffer. It does not reduce the campus’s own draw — we never claim it does — it simply doesn’t add meaningfully to the strain residents are angry about.

04 The bench

The rare fit sits at an interface. We staffed every side of it.

Waste-heat reuse fails at the seams — between the chip and the loop, the loop and the water, the water and the ground. So we didn’t hire one engineer to stretch across all of it; we assembled a specialist for each seam.

The chip side

Dr. Curtis Breville · IPE Solutions. Three-plus decades in data centers; creator of the industry’s first vendor-neutral liquid-cooling certification. Owns the source: the direct-to-chip loop, the exchanger handshake, and the contractual temperature floor.

The water

Jim Phene · Epiphene, Inc. Netafim and LAKOS alumni; ½-micron, chemical-free filtration. Owns the medium: the sealed loop’s cleanliness and the fertigation lines that feed the crops — second-generation water science.

The delivery

Ed Lohrenz · GEOptimize. Wrote IGSHPA’s Certified GeoExchange Designer curriculum; 105 MW of thermal capacity designed since 1982. Owns the load side: emitter schedules, buffer sizing, and the ground itself.

Behind them stand the operators — the rigs, the tanks, the crews. Meet the full bench on the About page →

05 If the physics interests you

Read it, then check our math.

This page is written to be scrutinized. If you’re an operator’s facilities engineer, a municipality’s reviewer, or a peer who thinks we’ve missed something — read the case and bring the red pen. A 45-minute call with the bench, no NDA needed for the public thesis. If it holds for your site, we scope the pairing together under a mutual NDA.

1 Microsoft reusable-heat analysis; ACEEE reject-temperature ranges, via ReImagine Appalachia (2026). Engineering values and open-item first passes are Intelligent Harvest design analysis, superseded by the engineer of record’s stamped figures.

The Water Architecture

One water system, four climates, one published meter.

Net-zero town-water draw in temperate Virginia, net-positive in the desert, a stormwater sink on the coast — committed where the gear is proven, honest about what's still under validation, and metered everywhere.

See the water architecture →