Vladimir Paul, CxA · Commissioning Agent, SGM Engineering Inc. · Florida
Vladimir Paul is a Certified Commissioning Authority and licensed Mechanical Contractor based in Florida, with over two decades of experience across the full lifecycle of mechanical and HVAC systems. His background runs from hands-on field work — technician, foreman, superintendent — through project management and company leadership, including CEO of a mechanical contracting firm, to his current focus in commissioning: the discipline where systems are tested, validated, and verified against design intent in real-world conditions.
What defines Vladimir's perspective is a combination that is genuinely rare: deep practical installation knowledge from years in the field, combined with high-level system commissioning expertise that operates at the intersection of engineering design and operational reality. He sees both what was specified and what actually happens when the system runs. The technical framework for Port 3 emerged from that vantage point — unsolicited, precise, and grounded in decades of watching HVAC systems perform and fail in the field. He is currently pursuing advanced research interests in modified absorption cycles for sub-70 °C driving temperatures. Connect with Vladimir on LinkedIn →
What is Port 3 — and why does the data centre benefit from it?
Ports 1 and 2 of the Reverse Thermal Plug export waste heat outward — to bio-processes and district heating networks. You proposed something structurally different. Walk me through the logic.
Ports 1 and 2 treat heat as a product to be sold or transferred. Port 3 treats it as a fuel. The idea is straightforward: instead of exporting the data centre's waste heat to an external off-taker, you route it into an absorption chiller that produces chilled water — and that chilled water goes back into the data centre's own cooling loop. The data centre partially cools itself using the heat it generates. The thermal liability becomes the cooling asset.
In a conventional data centre, cooling is an electrical cost — compressors, variable-speed drives, refrigerant cycles running continuously. An absorption chiller replaces the compressor with a thermochemical process. The electricity demand for cooling drops. The waste heat that was previously rejected to the atmosphere now does mechanical work before it leaves the building. The interface that enables this is Port 3 of the Reverse Thermal Plug — a third output on the same standardised boundary, feeding the chiller's driving circuit.
How does an absorption chiller actually work? No textbook — explain it like an engineer on a whiteboard.
Most people understand compression cooling. Almost nobody understands absorption. Make it concrete.
Instead of a compressor, you have a bucket of salt. Lithium bromide — LiBr — dissolved in water. That salt solution has an extreme affinity for water vapour. It pulls moisture out of the surrounding air with significant force — think of it as a chemical vacuum pump, except the only energy input is heat, not electricity.
Here is the cycle. The evaporator operates under slight vacuum — water does not need to reach 100 degrees Celsius to boil under reduced pressure. At a few millibars of vacuum, water boils at 4 to 5 degrees Celsius. That phase change absorbs heat from the surrounding circuit, producing chilled water at precisely the temperature a data centre cooling loop needs. The water vapour produced by that boiling is immediately pulled into the absorber chamber by the LiBr solution waiting there.
Now the salt solution is diluted — it has absorbed water and lost its pulling power. You pump it to the regenerator, where the free heat input — your data centre waste heat — drives the water back out of the solution. Concentrated LiBr is restored. The expelled water vapour moves to the condenser, returns to liquid, and cycles back to the evaporator. The compressor has been replaced entirely by the heat-driven regeneration step. No refrigerant. No high-pressure compression. The only moving parts are low-pressure pumps.
Working pair: Lithium bromide (absorbent) + water (refrigerant)
Evaporator output: 4–7 °C chilled water · operating pressure: ~6–10 mbar
COP: 0.7–0.8 (single-effect, chiller unit only) · no compressor, no refrigerant circuit
Net system COP (including vacuum pumps, solution pumps, controls): est. 0.45–0.65 under variable-load conditions
Driving temperature required: 70–90 °C (standard) · low-temp variants: 65–70 °C minimum
Commercial reference: Yazaki WFC series (Japan) — low-temperature absorption chillers
Also produced by: Carrier, Trane, Broad — primarily for cogeneration / turbine exhaust applications
Data centres reject heat at 40 to 60 degrees. The chiller needs 70 to 80. How do you bridge that gap?
This is the central constraint. The technology exists. The temperature does not match. What are the realistic paths?
The gap is real and must not be minimised. Standard data centre waste heat — condenser water return on an air-cooled campus, warm-water loop on a liquid-cooled campus — sits at 40 to 60 degrees Celsius. Even the most temperature-optimised low-end absorption chillers, like Yazaki's WFC series, need a minimum of 70 degrees, ideally 80, at the driving circuit inlet. You are 15 to 30 degrees short depending on the specific installation.
There are two engineering bridges. The first is a heat pump uplift — a relatively small unit sized only for the temperature boost, not for the full cooling load. At a coefficient of performance between 3 and 5 for a lift of this size, you spend roughly one kilowatt-hour of electricity to move three to five kilowatt-hours of heat from 55 to 80 degrees Celsius. The net cooling output from the absorption chiller still exceeds the uplift cost — but only if the waste heat itself is genuinely free. In a data centre, it is. The second bridge is solar thermal collection. In a high-irradiance climate, flat-plate or evacuated-tube collectors can deliver fluid at 70 to 85 degrees Celsius with no electricity cost. The data centre waste heat provides the stable thermal baseload; solar provides the temperature top-up. A buffer storage tank between the two smooths the solar intermittency and feeds the chiller a consistent driving temperature.
Florida makes this argument almost embarrassingly obvious. Vladimir's observation is precise: a car parked in the Florida sun reaches interior temperatures where you could brew tea within two hours. That is not metaphor — that is a solar thermal collector operating at ambient conditions with no engineering whatsoever. With actual collectors and thermal mass, 80 degrees Celsius is a conservative target.
DC waste heat available: 40–60 °C (air-cooled) · 50–65 °C (liquid-cooled)
Chiller driving requirement: 70 °C minimum · 80 °C optimal
Bridge option 1 — Heat pump uplift: ΔT +15–30 °C · COP 3–5 (at this lift) · electrical cost: ~0.2–0.3 kWh per kWh of chilling
Bridge option 2 — Solar thermal: 70–85 °C output · free fuel cost · intermittent, requires buffer storage
Hybrid: DC waste heat (baseload) + solar thermal (temperature boost) + insulated tank (buffer) → stable 75–80 °C driving circuit
Where does the system break down? What is the honest engineering challenge that is not yet solved?
The chemistry is proven. The hardware exists. Yazaki is a real company with a real product line. So why is this not already running at data centre waste heat temperatures everywhere?
Because three problems intersect at the system level in a way that none of them does in isolation. The first is thermal storage at small scale. The chiller wants a stable, continuous driving temperature. Data centre waste heat output varies with compute load — predictably over hours, less predictably over minutes. Solar input varies with weather. The buffer tank design that smooths both inputs simultaneously at small scale — 50 to 500 kilowatt thermal range — has not been solved in a cost-effective, field-validated configuration. Large district heating networks have solved this at megawatt scale. The small-scale version remains an open engineering problem.
The second is control. The absorption cycle has a critical timing decision: when do you trigger regeneration — the step where you drive water back out of the diluted LiBr solution using heat? Too early and you waste driving energy regenerating a solution that still has capacity. Too late and the solution is so diluted that cooling output collapses. In a laboratory with stable heat input and stable load, this is manageable. Under variable driving temperature from a real data centre load profile combined with variable ambient conditions, it becomes a real-time optimisation problem. No robust, low-cost adaptive controller for this specific configuration exists in commercial form today.
The third is balance-of-plant cost. The chiller itself — the LiBr unit, the evaporator, the absorber — is not the expensive component at this scale. The heat exchangers, the buffer storage, the control system, the water treatment circuit, and the installation engineering are. The capital cost of the balance-of-plant currently undermines the business case for deployments below roughly 500 kilowatt thermal driving input. Above that threshold, the economics work. Below it, they do not yet.
Crystallisation risk: sudden drop of driving temperature below ~55 °C causes LiBr solution to crystallise in circuit — requires controlled shutdown logic and buffer storage as protective layer
Corrosion: LiBr is highly corrosive to ferrous metals — all wetted components must be stainless steel or specified corrosion-resistant alloys
Vacuum integrity: any air ingress into the low-pressure circuit degrades COP rapidly — requires routine leak-testing as scheduled maintenance
Supply chain note: LiBr is produced primarily in China and Japan — EU critical materials exposure exists at scale; no immediate scarcity risk in 2026 but a relevant consideration for 20-year infrastructure planning
Fault handling: each connected off-taker maintains independent backup capacity — the interface is valve-controlled and isolatable in seconds. Full redundancy model defined in Interview II.
You mentioned a modification that has not been tried commercially. What is it?
You described an idea during our conversation that you said you have never seen implemented. It is worth stating precisely.
The idea is to apply a slight vacuum to the condenser chamber — not the evaporator, which already operates under vacuum, but the condenser side. In a standard absorption chiller, the condenser operates at near-atmospheric pressure. The regenerated water vapour condenses there before returning to the evaporator. If you reduce pressure in the condenser, the LiBr solution releases water vapour at a lower temperature during regeneration. That shifts the minimum effective driving temperature downward — potentially by 8 to 15 degrees Celsius, from the 70-to-80 band toward roughly 65 to 70 degrees Celsius — which narrows the gap to the upper edge of the data centre waste heat band and sharply reduces, though does not always eliminate, the uplift requirement.
The engineering constraints are real: the vacuum system adds complexity and energy cost, and there is a threshold below which the vacuum energy demand exceeds the benefit of the lower driving temperature. The lower bound is set by the cooling water — the condenser must still stay warmer than the cooling-water return, so the achievable reduction is climate-dependent. That threshold has not been rigorously characterised for this configuration. A modest reduction from the standard ~65 mbar condenser pressure down toward roughly 40 to 50 mbar absolute might shift the driving temperature enough to be worthwhile at a net energy cost that still makes the system viable. This needs to be modelled first — MATLAB or equivalent — before it is built. But the thermodynamic logic is sound, and to the best of our knowledge, no commercial system has implemented it at data centre waste heat conditions.
Hypothesis: reducing condenser pressure from the standard ~65 mbar toward ~40–50 mbar lowers the regeneration driving temperature by ~8–15 °C (from ~80 °C toward ~65–70 °C); lower bound constrained by cooling-water temperature
Required: modelling (MATLAB / EES) of net COP vs. vacuum energy cost across condenser pressure range
If validated: narrows the temperature gap toward the DC waste-heat band, reducing or removing the heat-pump / solar uplift
Status: thermodynamically plausible · not commercially implemented · PhD-worthy research scope
Dynamic system model (Modelica / MATLAB) under development with Vladimir Paul · expected Q3 2026 · contact via LinkedIn for collaboration
Contact: Vladimir Paul on LinkedIn — pursuing this research direction
How does Port 3 connect to the existing Reverse Thermal Plug architecture?
Ports 1 and 2 are already defined — bio-processes at 35 to 55 degrees Celsius, district heating at 60 to 80 degrees Celsius. Where does the absorption chiller sit on that same interface?
Port 3 taps the secondary loop at the highest available temperature — the point closest to the isolation boundary where the fluid is hottest before any downstream off-take has extracted energy from it. Depending on the driving temperature available and whether a heat pump uplift is included, the chiller's driving circuit receives fluid at 70 to 80 degrees Celsius. The chilled water output — at 6 to 10 degrees Celsius — is returned directly to the data centre's own cooling distribution system, reducing the load on the mechanical chiller plant.
The three ports now cover the full thermal output spectrum of the interface in a logical cascade. Port 3 at the high end extracts value from the highest-temperature portion of the secondary loop. Port 2 — district heating — takes the mid-temperature band. Port 1 — bio-processes — uses the lowest temperature band where biology operates. Nothing is wasted at any temperature level. The interface dispatches heat to the most thermodynamically appropriate use at each point in the cascade. That is what source-agnostic design actually means when fully realised: not just compatibility with different heat sources, but full utilisation of the available temperature gradient across every connected off-taker simultaneously.
Port 3 — Absorption chiller driving circuit: 70–80 °C · output: 6–10 °C chilled water back to DC
Port 2 — District heating supply: 60–80 °C · output: heat delivered to external network
Port 1 — Bio-processes (biogas fermenters, algae): 35–55 °C · direct temperature match
Cascade logic: highest temperature → most thermodynamically demanding use first
Result: closed thermal loop — DC waste heat partially returns as cooling to the DC itself
Boundary definition: the interface ends at the metered handover skid — ownership, liability, and control responsibility are defined in full in Interview II.
Ports 1 and 2 sell the heat. Port 3 puts it to work — and sends the result back. One interface. Three directions. Nothing wasted at any temperature level.
Why not simply export all the heat externally? Heat export remains an option — it is Port 2. Port 3 answers a different question: what if the data centre does not want to depend on an external off-taker for its cooling economics? Self-cooling via Port 3 lowers PUE directly at the IT load, reduces grid dependency, and eliminates the negotiation with external heat networks. The operator chooses the proportion. The interface supports both simultaneously.
Dimitri Wolf · Reverse Thermal Plug · Port 3-
01 · Finland · ActiveVatajankoski + E-Heat — Merikarvia / HonkajokiA containerised 1.5 MW data centre is physically plugged into the local district heating network. Waste heat exits at approximately 80 degrees Celsius as hot water and is piped directly to homes and municipal pools. Energy Reuse Effectiveness reaches up to 96 percent. HPC compute is sold as a service; heat is sold as a byproduct. One standardised interface, two revenue streams, no cooling towers. The nomenclature is different. The architecture is identical.Verdict: Thermal Plug — heat export mode · ERE 96%
-
02 · Finland · Live since November 2025atNorth FIN02 + Kesko — EspooColocation waste heat from FIN02 is piped directly to a Kesko retail store, displacing gas heating for the building. No conversion, no intermediate storage, no new energy source — direct low-temperature thermal export from server exhaust to commercial building circuit. The data centre is the boiler. The Kesko store is the off-taker. The pipe between them is the plug.Verdict: Thermal Plug — direct commercial heat off-take · gas displacement confirmed
-
03 · United Kingdom · Active since 2024Queen Mary University London + Schneider Electric — LondonA Tier 2 data centre's server exhaust is captured and rerouted into campus hot water and space heating systems. Running since 2024. Annual saving of £240,000 on gas. The university campus connected its data centre cooling output to its own heating circuit — the shortest possible interface path, owned and operated by one institution. The economics are audited and public. The concept is not theoretical.Verdict: Thermal Plug — campus-integrated heat recovery · £240k/yr gas saving
-
04 · Finland · Ongoing expansionMicrosoft Finland Region — Helsinki / Espoo District HeatingWaste heat from hyperscale data centres is being fed into the Helsinki and Espoo district heating networks via Fortum. Around 75 percent of the data centres' annual waste heat is projected to be captured and delivered to residential users at city scale — up to 350 MW thermal, eventually covering roughly 40 percent of regional district-heat demand. Fortum's heat-pump plants at two Microsoft sites began operating in May 2026; full waste-heat integration phases in alongside Microsoft's construction schedule. At this density, the data centre becomes a primary heat utility — not a supplementary contributor. The grid operator plans around it. That is not an analogy to a Thermal Plug at hyperscale. That is its definition.Verdict: Thermal Plug at hyperscale · ~75% projected heat capture · primary heat utility function
-
05 · France · ActiveEquinix Paris — District Grid IntegrationServer heat integrated into local district heating grids under EU Energy Efficiency Directive mandates. Equinix is operating under regulatory pressure what others are exploring voluntarily. That distinction matters: when the law pushes the interface toward existence, the infrastructure cost moves from speculative investment toward compliance cost. The revised EU EED now requires all data centres above 1 MW to assess waste-heat utilisation and to implement it where technically and economically feasible — a feasibility test, not a blanket mandate, but one that makes the Thermal Plug interface a near-default design consideration in European markets. Germany's national EnEfG goes further, setting minimum reuse quotas.Verdict: Thermal Plug — regulation-driven implementation · EU EED reuse-where-feasible
-
06 · United States · PPA signed February 2026Google + Ormat Technologies — NevadaA 150 MW geothermal power purchase agreement signed in February 2026, with new Ormat projects across Nevada supplying Google data centres via NV Energy's Clean Transition Tariff. First operations expected 2028. This is the closest existing configuration to the Deep Thermal Plug — geothermal energy contracted at the data centre edge. It is structured as an electricity-only agreement. The thermal cascade — the full ORC-plus-Thermal-Plug architecture — does not yet exist in this deployment. That is the gap. That is the opportunity.Verdict: Partial Deep Thermal Plug · power only, thermal cascade not yet implemented · gap identified
The projects being designed in 2026 will still be standing in 2050. The thermal interface decisions embedded in those designs will determine the bankable optionality of those assets for the next thirty years. The Deep Thermal Plug is not a retrofit. It is a design-stage decision.
Dimitri Wolf · Deep Thermal Plug