What exactly does "heat as a seed" mean?
You use the phrase "heat as a seed" to describe what happens when industrial waste heat is treated seriously. What does that mean in practice?
A seed needs three things: energy, a substrate, and time. Industrial waste heat provides the first of those reliably — continuous, stable, low-grade thermal energy that runs twenty-four hours a day, three hundred sixty-five days a year, at temperatures that are essentially useless for generating electricity but ideal for biological and physical processes that reward patience.
The question I keep returning to is not how we get rid of this heat. The question is what this heat makes possible.
That answer already exists in the ground. In Wittenberg, Saxony-Anhalt, a fifteen-hectare greenhouse operation — Wittenberg Gemüse — has been growing vegetables year-round since 2013, heated entirely by waste heat from the adjacent SKW Stickstoffwerke Piesteritz, an ammonia and fertiliser plant. In Rockstedt, Lower Saxony, a spirulina algae farm run by Roval GmbH draws heat from a local biogas plant to maintain the stable water temperature that makes year-round commercial algae cultivation viable in northern Europe. And in Rjukan, Norway, Hima Seafood opened a land-based trout farm in late 2025, drawing piped heat from the Green Mountain data center 800 metres away. These are not experiments. They are operating businesses. The seed is already germinating. The question is why it is not everywhere — and the answer is that every one of them still had to be wired up by hand. Standardize that connection, which is what the Thermal Plug does, and "why not everywhere" stops being a question.
The list is longer than expected
What can an industrial site or a data center actually connect to its waste heat right now, without novel technology?
The list is longer than most people expect, and the common thread is simpler than it looks: every application on it needs stable heat, not high-grade heat. Greenhouses need ambient growing temperatures in the 25 to 35 degree Celsius range — supply arrives at higher temperatures from an industrial source and is brought down to the target range via a heat exchanger at the boundary. The Thermal Plug standardises that connection point. Algae, aquaculture, pharmaceutical processes — each has its own window. The selection below is representative, not exhaustive.
· Greenhouse / horticulture: growing temp 25–35 °C (TP supply ≥ 60 °C, via HX)
· Algae cultivation (Spirulina, Chlorella): 25–35 °C
· Tilapia aquaculture: 20–32 °C
· Trout aquaculture: 12–16 °C (lower-grade source or loop bypass)
· Pharmaceutical drying / process heat: 40–60 °C
Representative selection. Full classification across the more-than-24,000 PfA sources in i5.
Aquaculture separates by species: tilapia is a direct thermal match for data center waste heat at most cooling loop temperatures. Trout — a colder-water species — prefers 12 to 16 degrees, which requires a lower-grade source or loop bypass. The Aalsmeer Energy Hub in the Netherlands has been supplying a plant nursery from a data center since April 2021. Aurubis in Hamburg contracted 60 megawatts of copper smelter waste heat for the city's district network in 2017, with initial heat flowing by 2018. None of these require novel technology. All of them, today, require bespoke engineering to connect. That is the gap the Thermal Plug closes.
"Stable heat is what makes marginal land viable. Germany has a public register of more than 24,000 waste heat sources. The question is not whether it can grow something. The question is what."
Dimitri WolfThe heat doesn't stop at the fence
Once you decide the heat matters, does that change how you design the campus itself — not just the thermal connections but the physical site?
Fundamentally. And the change has to happen before the first concrete is poured, because almost none of it is retrofittable.
In North Rhine-Westphalia, the prevailing wind comes from the southwest to west — between 220 and 270 degrees, averaging 3 to 6 metres per second at ten metres height. That is a documented physical fact, verified across DWD weather stations in Düsseldorf, Essen, and Cologne. If you know that, you orient the building so the intake faces into that wind and the exhaust faces away. You shape the terrain on the upwind side — a berm of 2 to 8 metres can accelerate approach velocity at the intake by 10 to 30 percent under the right geometry. You place the perimeter vegetation where evaporative cooling and plume dilution do the most work.
These three inputs — topography, solar orientation, prevailing wind — are currently treated as three separate specialist consultancy reports assembled into a sustainability document at the end of the design process. The solar-optimal orientation in NRW and the wind-optimal orientation conflict by 45 to 90 degrees. In the siloed approach, that tension is resolved arbitrarily. Stuttgart demonstrated that ventilation corridors can be legally protected — the Stuttgart Regional Plan has embedded cold-air corridors with a minimum width of 100 metres since 1998. The design logic is established. What does not yet exist is a methodology that integrates all three inputs as a unified brief from day one of site design — one that resolves, explicitly and quantitatively, the trade-off between solar and wind optimisation, and connects both to the thermal export target. That gap is real, it is confirmed absent from any named German practice or standard, and it is the opportunity. The methodology described in this answer maps directly to a structured pre-design protocol — a preflight check any project team can run before the first planning submission. The earlier it is completed, the cheaper the corrections. After planning freeze, most of it is unavailable. Every one of these site decisions orients around a single fixed point: where the heat enters and leaves — the Thermal Plug interface. Fix that point at the drawing stage and the rest of the site plan follows from it.
Slower than electronics — not softer
Vegetation, earth tubes, terrain shaping — these sound soft compared to the engineering in the other interviews. How substantial are the actual effects?
The physics is not soft. It is simply slower than electronics and measured over longer timescales.
Mature native tree canopy — Acer pseudoplatanus, Quercus robur, Betula pendula, the species native to NRW and high in evapotranspirative output — reduces local air temperature by 1 to 3 degrees Celsius within 50 metres of the canopy, confirmed across peer-reviewed meta-analyses. On a data center campus with well-established perimeter planting, that translates to 0.5 to 1.5 degrees Celsius at the air intakes. Not large in isolation — but it is continuous, requires no electricity, and strengthens over the campus lifetime as the trees mature. A mature tree in full summer transpiration moves 200 to 400 litres of water per day through evapotranspiration; the latent heat absorbed in that process is substantial and distributed across the campus boundary. The campus that plants for thermal performance today is running a cooler baseline in ten years.
Earth-air heat exchangers add a more direct intervention. Tubes buried 1.5 to 3 metres deep in NRW soil — which holds 10 to 14 degrees Celsius year-round at that depth, regardless of surface temperature — pre-cool intake air by 2 to 5 Kelvin per pass at realistic data center airflow rates. Not the 5 to 10 Kelvin figure from residential literature, which applies at airflow rates a tenth of what a DC operates and degrades rapidly under sustained load. The DC-scale figure is more modest. It still meaningfully extends free-cooling hours over a NRW summer. The hard constraint is groundwater: in the Rhine alluvial zone around Cologne, the water table sits 1 to 3 metres below surface. Earth tubes there are limited to about 1.5 metres depth — the technique still works, the thermal margin is tighter. Site-specific ground investigation is not optional. None of this replaces liquid cooling. All of it reduces what liquid cooling has to do, without consuming electricity.
The barrier to planting is organisational, not physical. None of the vegetation described here is technically complex to install. The complication is timing and ownership: trees planted after construction find their positions compromised by drainage routes, buried services, and access roads. The decision is a pre-construction annotation on the site plan — a zone marked before the foundations go in. The planting itself can come later. Whether the operator plants them, a municipality does, or a third-party organisation takes that role — even one with an interest in future timber or biomass — matters less than making the decision early. The infrastructure for this is a line on a drawing. Thinking about it upfront costs nothing.
Running with the biology
You mentioned something about scheduling compute workloads to the morning. Explain what that is about.
As sunrise arrives and vegetation enters its daily transpiration cycle, evapotranspiration rises sharply from around 09:00, and the local microclimate around a well-planted campus becomes measurably cooler than at dawn or midday. A data center with established perimeter planting gets progressively more passive cooling through that morning window — from the surrounding biology, not from mechanical systems.
Batch AI workloads carry no latency requirement. Training runs, model evaluation, overnight preprocessing — these can be moved. Shift that bursty load into the 09:00-to-midday window and you are running hardware when the building's passive thermal environment is most effective, when ambient temperatures are still below their daily maximum, and when EPEX SPOT electricity prices are falling as solar feed-in ramps up. The thermal and economic cases converge. The boundary matters: the 06:00-to-09:00 band is the German grid's ramp-up period — demand rising before solar arrives, prices high, grid stress at its worst — so 09:00 is where the case holds on both. Real-time inference cannot move. Batch can.
And there is a consequence that connects straight back to the Thermal Plug: morning load means morning waste-heat export. A Fernwärme network without thermal storage may not absorb a morning surplus that exceeds its instantaneous demand. Site-level optimisation and network-level intelligence are one connected system, not two separate problems — which is exactly where the next interview goes.
Terraforming is just site design with a longer timescale
You've taken us from what's operational today through campus physics and load scheduling. What is at the far end of this logic?
Apply this thinking at district scale, over 10 to 20 years, and you are no longer managing a campus. You are shaping a microclimate.
Perimeter vegetation that started as a thermal buffer becomes habitat. Retention ponds designed for heat rejection and condensate reuse become biodiversity corridors. Terrain shaped for the prevailing wind becomes landscape. These transitions are not planned — they are the physical consequence of sustained intentional design compounding over time. The physics is the same at every scale. The scope is different.
At the far end of that logic is a word most people associate with Mars. Terraforming. It belongs in this conversation because it is the honest endpoint of the sequence — not a science fiction claim, but the name for what you are doing when you scale deliberate site design to district scale and extend the timescale from years to decades. The lesson from Iceland is not that geothermal is special. The lesson is that stable, accessible thermal energy changes what is possible. Germany has a public register of more than 24,000 waste heat sources, attributed with address, output temperature, and monthly heat profile. The heat is documented. The biology follows when the infrastructure makes the connection. It always has.
Nothing in this interview is specific to data centers. Any industrial building that generates substantial waste heat — a pharmaceutical plant, a steel mill, a ceramics kiln — operates under the same logic. The data center is the most visible current case because of its scale and growth rate. The principle is not sector-specific. The Thermal Plug is named for what it does, not for who builds it.