Active Discussion

Heat, Water and Carbon: A Technical Policy Supplement for Alberta's Largest Data Centres

L
lisa
Posted Sun, 1 Nov 2026 - 12:35

This supplement turns a public-interest concept into testable design and approval questions. Its calculations are illustrative scenarios based on publicly reported information, not a final engineering assessment of Wonder Valley or any other project.

Purpose and boundary

This paper asks what Alberta should require if it approves a multi-gigawatt data-centre and self-generation complex. It does not assume that approval is inevitable, and it does not conclude that industrial symbiosis can make every proposed scale acceptable.

The framework has four goals:

  1. account for the complete energy and water system rather than isolated components;
  2. prioritize measures by their effect on fuel use, emissions and public risk;
  3. make residual heat, water and carbon physically connectable to neighbouring industry;
  4. preserve an enforceable stop mechanism when early phases fail to meet their conditions.

The public figures remain unsettled. Wonder Valley was initially described as a 7.5-gigawatt campus, while a June 2026 update described up to nine gigawatts of on-site generation at full build-out.[2] A preliminary analysis prepared for Sturgeon Lake Cree Nation estimated approximately 18 gigawatts of eventual heat release and roughly 30 million tonnes of annual carbon-dioxide emissions.[1]

Those figures should be treated as screening inputs. A project-specific assessment would need confidential and public engineering information that is not available here: turbine models, lower- versus higher-heating-value efficiency, ambient derating, heat-recovery steam-generator design, computing load, power-usage effectiveness, cooling technology, capture configuration, parasitic load, dispatch profile and water balance.

1. Define the system boundary before counting heat

A gas-powered computing campus converts energy in sequence:

fuel → electricity plus generation losses → computation plus low-grade heat

If the assumed fuel input is 17 gigawatts and the electrical output is 7.5 gigawatts, the simple-cycle-or-plant-level electrical efficiency implied by those two figures is approximately 44 per cent. About 9.5 gigawatts would leave the generating stage as heat and other losses. The 7.5 gigawatts delivered to computing would then also become heat after doing computational work. Across the final site boundary, the total heat rejection could therefore approach the original fuel input.

That is not double-counting if the two stages are identified clearly. It becomes misleading when “18 gigawatts of waste heat” is presented as one homogeneous stream available to one customer.

A regulatory filing should show a Sankey-style energy balance with at least these nodes:

  • fuel energy entering each generating block;
  • gross and net electricity output;
  • generation exhaust and cooling losses by temperature;
  • capture and compression energy;
  • electricity delivered to information-technology equipment;
  • cooling, pumps, networking, storage and other auxiliary loads;
  • heat delivered to external users;
  • residual heat rejected to air or water.

All values should include annual energy as well as peak power. A gigawatt available for a few hours is not equivalent to a dependable gigawatt-year.

2. Separate thermal grades

High-temperature generation heat

Gas-turbine exhaust may be valuable for additional power generation, steam production, absorption cooling or industrial processes. The first policy question is whether that heat should be used inside a high-efficiency combined-cycle configuration before any external sale.

Exporting high-temperature heat to a low-value customer while operating an inefficient generating block could increase rather than reduce the overall opportunity cost. Internal efficiency therefore comes before industrial symbiosis.

Low-temperature computing heat

Most electricity entering computing equipment eventually appears as heat at a relatively low temperature. The useful outlet temperature depends on cooling architecture, server design and allowable chip temperature. Air cooling usually makes heat collection less valuable than warm-water liquid cooling.

Low-temperature heat may support aquaculture, soil heating, snowmelt, low-temperature district systems or heat-pump-assisted applications. Its economic value depends on distance, temperature lift, annual load factor and the cost of backup systems.

It should not be described as economically convertible back into electricity. As an illustration, a heat engine working between 60°C and 5°C has a theoretical Carnot ceiling of about 16.5 per cent before real-world losses. Practical output would be much lower.

3. Put fuel and emissions ahead of heat reuse

Suppose, only for illustration, that a plant producing 7.5 gigawatts of net electricity requires 17 gigawatts of fuel input. Raising net electrical efficiency from the implied 44 per cent to 60 per cent would reduce fuel input to approximately 12.5 gigawatts.

That is a fuel reduction of roughly 26 per cent. If the preliminary 30-million-tonne emissions estimate scaled directly with fuel use and nothing else changed, the illustrative reduction would be approximately eight million tonnes per year.

This is not a forecast. It excludes ambient conditions, plant auxiliary load, carbon-capture energy, cycling and the possibility that the published inputs describe different project boundaries. It demonstrates why verified generation efficiency matters more to total emissions than finding another customer for a small share of low-grade server heat.

The approval standard should therefore be performance-based:

  • maximum fuel input per net megawatt-hour;
  • maximum carbon-dioxide-equivalent emissions per net megawatt-hour;
  • separate disclosure of capture rate and avoided emissions;
  • minimum annual availability;
  • no credit for heat “available” unless a defined methodology shows what external fuel or electricity it displaced.

Alberta's current hardware levy moves in the opposite direction: it applies at two per cent to grid-connected data centres of at least 75 megawatts, falls to one per cent where a grid-connected centre generates its own power, and falls to zero off-grid.[4] This rewards a supply arrangement rather than measured emissions intensity.

A revised agreement could retain protection for ratepayers while tying relief to independently verified performance.

4. Treat carbon capture as a staged operating condition

Carbon capture is the only proposed mechanism in this framework capable of deeply reducing the remaining combustion emissions, but a paper promise is not enough.

Large gas-turbine exhaust streams have low carbon-dioxide concentration and high volumetric flow compared with many industrial capture sources. Capture also imposes energy, cooling, compression and water requirements. The regulator should require measured performance from an earlier block before permitting later blocks.

A staged condition should disclose:

  • inlet exhaust composition and flow;
  • capture rate during stable and changing load;
  • net emissions after capture energy and compression;
  • solvent degradation products and air emissions;
  • water withdrawal and consumption;
  • transport route, injection location and storage liability;
  • downtime rules and annual capture availability;
  • treatment of carbon used in greenhouses or products versus permanently stored carbon.

Carbon supplied to a greenhouse should not be counted as permanent sequestration. It may displace purchased commercial carbon and support food production, but most of it will return to the atmosphere.

Likewise, “food-grade” must be an output specification. Captured gas requires purification, monitoring, compression, storage and quality control before it can be safely sold into food or greenhouse markets.

5. Resolve the water balance before approval

Public reporting says the Municipal District of Greenview holds a licence for up to six million cubic metres per year and a preliminary certificate for up to 24 million cubic metres, while the proponent has estimated maximum project demand below six million cubic metres annually.[2]

Those numbers should be reconciled with the proposed cooling system, capture process and operating phases in one auditable water balance.

For scale:

  • six million cubic metres per year averages about 16,400 cubic metres per day;
  • 24 million cubic metres per year averages about 65,800 cubic metres per day.

A commonly used screening estimate of 1.6 litres evaporated per kilowatt-hour of heat rejection would consume about 38,400 cubic metres per day for each continuously evaporated thermal gigawatt. Under that assumption, six million cubic metres per year would cover only about 0.43 thermal gigawatts, while 24 million would cover about 1.71 thermal gigawatts.

That apparent mismatch does not prove the published water estimate is wrong. It may indicate predominant air cooling, closed-loop systems, hybrid operation, a different consumption factor or a smaller water-cooled load. It does show that the cooling architecture must be disclosed before broad claims about heat and water can be assessed together.

The water filing should separate:

  • gross withdrawal;
  • consumptive use;
  • return flow and temperature;
  • blowdown and treatment residuals;
  • seasonal and drought restrictions;
  • cooling demand;
  • carbon-capture demand;
  • recovered condensate;
  • potable, process and irrigation-quality streams;
  • each development phase rather than only the final average.

6. Examine combustion-water recovery without promising it

Methane combustion produces water vapour. In ideal stoichiometric terms, the reaction produces 36 units of water mass for every 44 units of carbon-dioxide mass. Applied mechanically to 30 million tonnes of carbon dioxide, that corresponds to about 24.5 million tonnes—or approximately 24.5 million cubic metres—of water vapour in the exhaust.

That is a chemical mass balance, not a recoverable-water forecast.

Actual recovery depends on fuel composition, excess air, exhaust humidity, dew point, cooling temperature, capture equipment, corrosion control, treatment requirements and the energy needed to condense and move the water. A carbon-capture system may cool exhaust and recover some condensate, but the recoverable share requires project-specific engineering.

Any recovered water should be credited only after measuring:

  • annual quantity under real weather and load conditions;
  • contaminant profile;
  • treatment energy and chemical use;
  • discharge or beneficial-use permit requirements;
  • the fresh-water withdrawal actually displaced.

7. Design a standard resource interface

The operator's obligation should be to make defined resources physically available at the site boundary. External customers then choose whether to connect.

Thermal connections

At minimum, the design should reserve routes and connection points for:

  • low-temperature water from the computing loop through isolation heat exchangers;
  • medium- or high-temperature water or steam where the generating configuration can provide it without reducing higher-value internal efficiency;
  • return-water connections and metering;
  • expansion capacity for future customers.

The interface specification should publish temperature bands, maximum flow, pressure, annual availability, outage notice, water chemistry, metering rules and whether supply is firm or interruptible.

Carbon connection

A carbon header should be built only when capture is operating and the product meets a defined specification. The interface should disclose purity, contaminants, pressure, storage volume, delivery variability and the accounting treatment for utilization versus storage.

Treated-water connection

Recovered condensate or treated process water should have separate quality classes. “Treated water” is not a sufficient specification for irrigation, aquaculture, boiler feed or food production.

Reliability boundary

The data centre must retain enough heat-rejection capacity to operate safely with zero external demand. Offtakers should retain backup heat or accept an interruptible contract. Neither side should be able to threaten the other's safe shutdown.

8. Match candidate industries to demand profiles

The province or municipality should zone surrounding land as a heat-and-resource park, but it should not forecast uptake as though every candidate were commercially proven.

Potential customers include:

Candidate Useful resource Principal constraint
Greenhouses Medium-temperature heat, electricity, purified CO₂, irrigation water Summer heat demand, labour, lighting and market competition
Aquaculture Stable low-temperature heat Market size, species fit, water quality and biosecurity
Lumber and panel drying Medium-temperature process heat Mill proximity and retrofit economics
Grain and oilseed drying Seasonal medium-temperature heat Short annual season and distance to producers
Absorption cooling Higher-temperature heat Capital cost and seasonal operating profile
Snowmelt and soil heating Low-temperature heat Low economic value
Direct-air-capture systems Heat and electricity Technology, cost and storage uncertainty at this scale

No single customer should anchor the environmental case. The cluster should be evaluated as a portfolio with explicit low-, expected- and high-uptake cases.

9. Use staged approvals and measurable gates

A four-phase campus should not receive one irreversible approval based on final-build promises. Each phase should have a public performance gate.

Gate A: before construction

  • complete cumulative-effects and rights assessment;
  • disclosed energy and water balances;
  • financial security for reclamation;
  • consultation funding and technical-access arrangements for affected First Nations;
  • reserved corridors and design for the resource interface.

Gate B: before commercial operation

  • commissioned meters and public data schema;
  • tested efficiency, water and emissions performance;
  • safe full-load heat rejection without external customers;
  • emergency and drought operating plans.

Gate C: before the next generating block

  • at least one full seasonal operating record;
  • compliance with emissions and water intensities;
  • capture performance where required;
  • updated cumulative-effects assessment;
  • verified reclamation funding;
  • explanation of any difference between offered and actually used resources.

Gate D: before final build-out

  • independent review of all previous blocks;
  • revised regional grid, air, water, labour and infrastructure effects;
  • updated Indigenous consultation and partnership decisions;
  • proof that liabilities have not outgrown security.

10. Public reporting must distinguish capacity from outcomes

Every year, the operator should publish:

  • fuel input and net electricity by generating block;
  • computing and non-computing electricity consumption;
  • emissions before and after capture;
  • capture uptime and permanent storage;
  • gross water withdrawal, consumption and return;
  • heat available by temperature grade;
  • heat contracted, delivered and rejected;
  • carbon and water delivered to external users;
  • outages and permit exceedances;
  • reclamation security and ownership changes.

Heat infrastructure should not be credited merely because a pipe exists. Reports should separate:

  1. theoretical resource;
  2. engineered availability;
  3. contracted quantity;
  4. delivered quantity;
  5. verified displacement of another energy or water input.

11. Preserve the no-build and smaller-build options

Industrial symbiosis is not an entitlement to approval. The province exempted Wonder Valley from a provincial environmental impact assessment while retaining permit requirements.[3] Given the preliminary scale estimates and the continuing disputes over consultation and cumulative effects, a comprehensive whole-system review remains the prudent standard.

The approval process must be able to conclude that:

  • the full build-out is unacceptable;
  • a smaller configuration is acceptable;
  • a different generation mix is required;
  • later phases must wait for new evidence;
  • the project should not proceed.

The strongest opposing case is that a resource park could become approval cover for an emitter whose footprint overwhelms the benefits around its fence. The framework answers that objection only if efficiency, emissions, water, rights and decommissioning conditions are binding—and if failure can actually stop expansion.

Questions requiring independent study

Before these concepts are converted into permit conditions, Alberta should commission or require independent work on:

  • the actual generation and computing configuration;
  • combined-cycle and carbon-capture performance in northern Alberta conditions;
  • regional carbon-storage geology and transport;
  • Smoky River seasonal and cumulative water effects;
  • recoverable combustion condensate;
  • thermal-network routing and loss;
  • realistic regional greenhouse, aquaculture, forestry and agricultural demand;
  • labour, housing, road and emergency-service effects;
  • electricity-market treatment of sales to neighbouring users;
  • Indigenous rights, governance and optional ownership structures;
  • full-site decommissioning cost.

Conclusion

The engineering concept is not that every joule, litre or molecule can be profitably reused. It is that Alberta should know where each one goes, reduce the largest harms first, make useful residual resources connectable, and prevent later phases from advancing on promises that earlier phases failed to meet.

The correct policy sequence is:

assess → minimize demand → improve generation → reduce and capture emissions → manage water → connect residual resources → verify outcomes → decide whether expansion is justified.

That sequence can support industrial development without confusing industrial symbiosis with environmental absolution.

Sources

[1] https://www.cbc.ca/news/canada/edmonton/wonder-valley-slcn-court-physicist-9.7403677 — Proposed data centre would release ‘23 Hiroshima-sized atomic bombs worth of energy’ a day, physicist says > "At full operation, it would release around 18 gigawatts of heat a day and about 30 million tonnes of carbon dioxide a year — more than four per cent of Canada’s total emissions, said Davies, whose assessment was based on publicly available information." [2] https://www.cbc.ca/news/canada/edmonton/wonder-valley-newsletter-open-house-9.7217459 — Updated info about water use, timelines for proposed Wonder Valley AI project in Alberta > "It would inhabit approximately 64 square kilometres in the Municipal District of Greenview and will include up to nine gigawatts of on-site power generation capacity when all phases are completed, a spokesperson for the open house confirmed in an email to CBC News." > "The newsletter said Wonder Valley would draw water from the nearby Smoky River for the facility, and noted that the M.D. of Greenview holds a Water Act licence to use up to six million cubic metres of water per year. It also has a preliminary certificate to withdraw up to 24 million cubic metres of water per year." [3] https://www.cbc.ca/news/canada/edmonton/wonder-valley-data-centre-environmental-impact-assessment-9.7158526 — Wonder Valley AI Data Centre exempt from provincial environmental impact assessment, still needs permits > "The Ministry of Environment and Protected Areas says the Wonder Valley AI Data Centre Park project was reviewed by an independent approvals officer who deemed it exempt from the assessment due to its proposed use of standard power and water systems." [4] https://www.blg.com/en/insights/2026/07/data-centre-regulation-in-alberta — Data centre regulation in Alberta: key considerations > "The Financial Statutes Amendment Act, 2025 (No. 2) introduced a 2 per cent levy on computer hardware that applies to grid-connected data centres drawing 75 MW or more."

CanuckDUCK Provenance Package

Status: Verified. This article has a lightweight authenticity reference anchored to CanuckDUCK local Hedera; full content remains in Pond.

  • Package: bb245cf9-c222-4a19-8052-e419059eafaf
  • Hedera topic: 0.0.1068 / sequence 10456
  • Graph pathway and Consensus binding: present as package sections; Consensus may show no linked poll until a public input surface is created.

Verify package · Machine-readable JSON

--
Consensus
Calculating...
0
perspectives
—
views
Constitutional Divergence Analysis
Loading CDA scores...
Perspectives 0