Canada Faces Water Risks From Hyperscale AI Data Centers

  • Thread Author
On a sun‑baked day in Nanaimo, B.C., neighbours and lifelong residents found themselves facing a new kind of local infrastructure fight — not over a factory or a freeway, but over an incoming hyperscale data centre whose thirst could, opponents warn, strain municipal water supplies already stressed by drought and hotter summers.

Background​

The global race to build the physical underpinnings of artificial intelligence — large, power‑dense data centres that host the GPUs and accelerators realising modern generative AI — has created two paradoxes at once. First, communities are being sold economic promises: construction jobs, tax revenue, and “future” high‑tech employment. Second, those same projects carry quietly large resource demands: electricity measured in megawatts and water measured in millions of litres used for cooling. In Canada, federal and provincial incentives, abundant hydroelectricity in many regions, and a generally cool climate have made the country a magnet for hyperscale investment. But as projects multiply — and as critics point out examples abroad where promised water use estimates were exceeded — community resistance and regulatory gaps are surfacing.
This feature unpacks what data centres consume, why AI‑driven facilities are different, how Canadian projects are being permitted today, and what practical policy and technical paths can reduce risks to water security and community trust.

Overview: Why AI data centres drink water​

What data centres are asking for now​

Modern hyperscale data centres are not the tidy rooms of the early internet era. They are purpose‑built campuses containing thousands of racks of servers and thousands more high‑performance processors humming 24/7. To run those processors reliably you need:
  • Continuous, resilient electrical power (often tens to hundreds of megawatts).
  • Cooling systems that remove heat at extreme density and throughput.
  • Redundant infrastructure for reliability and compliance.
Traditional cooling strategies — evaporative cooling towers, chilled water loops, and direct evaporative systems — use water because it is an efficient medium for moving heat away from tightly packed silicon.

Why AI increases water intensity​

Generative AI workloads change the equation in two ways:
  • Density. AI models push more compute into smaller physical footprints. Higher compute density generates more heat per square metre than general cloud workloads, increasing the need for heavy cooling.
  • Duty cycle. Inference and training workloads for large models can run continuously and unpredictably, limiting the effectiveness of intermittent or seasonal air cooling.
Together, density and duty cycle mean AI‑optimised data centres are more likely to rely on water‑based cooling more often and for longer periods than older facilities designed for sporadic or lower intensity workloads.

What the numbers say — and what they don’t​

Representative figures​

Several independent technical studies and industry filings provide a window into orders of magnitude:
  • Individual medium‑to‑large data centres have been shown to draw millions to hundreds of millions of litres of water per year for cooling, depending on size and cooling design.
  • A single 100‑megawatt facility, using conventional evaporative cooling, can require on the order of a million to two million litres of water per day when operating with water‑intensive cooling systems.
  • Research that focused on the water footprint of AI models — including training and inference — estimated significant consumptive water needs during training runs and projected large aggregate withdrawals as AI demand grows.
These numbers vary widely by cooling design, local climate (air‑cooling works far more of the year in cold places), and whether non‑potable or reclaimed water is used. They are estimates, not immutable facts; the variability and opacity surrounding actual operational figures are part of the problem.

The limits of published figures​

Two practical issues make transparent comparison hard:
  • Industry reporting is inconsistent. Some operators publish aggregate water withdrawals in sustainability reports, but few disclose per‑facility daily or annual consumptive volumes.
  • Water accounting methodologies differ. "Withdrawal" vs "consumption" vs "evaporative loss" are technically distinct: water withdrawn from a utility may be partly returned to the system; what matters to a municipality is net consumptive loss and peak daily withdrawal when supply is limited.
Where figures do exist, they sometimes raise alarms — especially when early estimates provided to local authorities were later shown to understate usage during hotter years or during construction phases.

The Canadian context: growth, incentives, and gaps in oversight​

Policy environment and economic pitch​

Canada has positioned itself as a desirable data‑centre market: relatively cool climate, abundant low‑carbon electricity in many provinces, and active federal and provincial incentives. Federal programs and provincial outreach have combined to encourage large cloud providers to acquire large tracts of land and pursue campus builds.
Local economic development messages emphasise:
  • Capital investment into previously underutilised land.
  • Construction employment and potential indirect jobs.
  • Property tax revenue and ancillary commercial activity.
These promises can be persuasive to municipal councils and may incentivise permissive zoning decisions in the absence of robust local technical assessment.

Regulatory patchwork​

But regulation governing water use for data centres in Canada is patchy:
  • Energy regulators often review grid‑connection and electrical load requirements, so power impacts get technical scrutiny.
  • Water, however, is largely a municipal responsibility. Many municipalities may not have metering infrastructure or the technical staff to scrutinise complex cooling system forecasts.
  • Where metering exists, it may be applied unevenly: some large commercial accounts are billed on flat or non‑volumetric rates, reducing visibility into true consumption.
This creates a governance gap: the political and planning arms that approve or zone these facilities may lack the technical capacity or the regulatory levers to require conservative, transparent water commitments.

Case examples and lessons​

Overpromising and surprise usage elsewhere​

International precedent provides cautionary tales. In multiple jurisdictions, initial public statements about planned water use for data centre campuses have been followed by much higher measured consumption once facilities ramped to full capacity or during hot years.
These hindsight examples share common features:
  • Initial forecasts were framed as “typical” or “current” usage and did not represent full‑capacity or worst‑case scenarios.
  • Construction and ramp‑up phases incurred additional water use that was not always clearly separated from operating consumption in public disclosures.
  • Local regulators and communities had limited ability to enforce conservative limits once permits were issued.
The lesson is not that data centres must be banned — it is that realistic scenario planning, conservative permitting, and binding monitoring are needed to avoid community harm.

Canadian projects under scrutiny​

In Canada, several large cloud operators have aggressively expanded land holdings and development plans. Some projects have received municipal approvals that include water allocations measured in litres per second or cubic metres per year. Where those allocations exist, they can equate to hundreds of millions or even over a billion litres a year for large campuses — volumes that matter when a community’s potable supply is constrained by drought or seasonal low flows.
At the same time, some Canadian facilities have been permitted without detailed metering or without clear stipulations tying water draw to declared cooling modes (for example, strictly permitting municipal potable water only when air‑cooling thresholds are exceeded). That combination — large allocations and limited ongoing transparency — is the core of community concern.

Environmental and social risks​

Competition for scarce water​

In water‑stressed regions, data centres can compete directly with agricultural users, small businesses, and residents. Even where total annual use seems small in the context of municipal supply, the timing of withdrawals during heatwaves or droughts can create critical shortages. Municipal systems are sized for residential and traditional industrial loads; sudden spikes for bio‑industrial cooling can exceed routine provisioning.

Lock‑in of high‑water infrastructure​

If a region permits multiple water‑intensive data centres early in the development cycle, the cumulative burden may push municipalities into hard choices later: either invest in new, expensive water infrastructure; restrict other local water uses; or face public backlash and litigation.

Reputational and legal consequences​

Operators that understate or obscure their water use risk reputational damage. Where local communities perceive that promises were made and then broken, opposition can grow rapidly — leading to delays, lawsuits, and cancellations that damage both corporate and municipal interests.

Industry responses and technical alternatives​

Cooling design choices​

Data centre design can dramatically alter water intensity. Cooling options include:
  • Air cooling (free‑air economisation): uses outside air when ambient conditions permit and can eliminate water use most of the year in colder climates.
  • Closed‑loop water systems and liquid immersion cooling: highly efficient but still may require some water for heat rejection, depending on the heat sink.
  • Use of non‑potable water (reclaimed or industrial): avoids stressing potable supplies but requires treatment and local approvals.
  • Direct evaporative cooling: effective but water‑consumptive.
  • Hybrid systems: default to air cooling and only use water during extreme ambient conditions.
Investments in alternative cooling architectures and reuse systems reduce potable water use but may raise costs or require more on‑site land for heat rejection.

Corporate commitments and their limitations​

Some operators publish high‑level sustainability commitments, including pledges to deploy air cooling where feasible or to use recycled water. However, independent verification is uneven. Where operators commit to drawing municipal water only above certain temperature thresholds, critics have shown that hot years, construction periods, or misclassification of operational modes can still drive significant potable withdrawals.

Practical recommendations — technical, regulatory, and municipal​

For municipalities evaluating proposals​

  • Require third‑party, scenario‑based water demand modelling: permit applications must include full‑capacity worst‑case estimates, seasonal and daily peak demand curves, and separate line items for construction vs operations.
  • Mandate volumetric metering and public reporting: facilities drawing municipal water should be metered with real‑time or monthly public reporting to a municipal portal, and billing should reflect measured volumes.
  • Permit conditional allocations tied to cooling technology: approvals should specify allowable cooling modes and require binding transitions to lower‑water designs over time.
  • Insist on non‑potable sources where feasible: require proof that alternative water (recycled, reclaimed, or treated industrial) has been considered and that potable draw is minimized.
  • Require water‑risk offsets only as a last resort: market tools like water credits can be useful but should not substitute for reducing on‑site potable use.

For provincial and federal policymakers​

  • Establish standardized reporting requirements for water withdrawals and consumption at data centre facilities, with harmonized definitions of “withdrawal” vs “consumption.”
  • Tie public incentives to demonstrated low water intensity per unit of compute, not merely to job creation metrics.
  • Fund technical assistance to municipalities so planning departments can properly evaluate highly technical cooling and water modelling.

For operators and cloud providers​

  • Publish facility‑level water metrics under a standardised framework, including total withdrawals, net consumption (evaporative loss), reuse rates, and source type (potable vs non‑potable).
  • Design future campuses with water‑minimal cooling as a default, with air‑first strategies and reserved water only for emergency contingencies.
  • Build community water advisory committees for major campus projects that meet regularly and receive measured water use data.

Balancing the books: jobs, growth, and the public trust​

The economic argument for data centres — large capital inflows, construction activity, and potential local spend — is real and can be beneficial. But the jobs picture is often misread: operating hyperscale campuses tends to produce a relatively small number of permanent on‑site technical roles compared to the headline construction jobs, and many hosting profits return to multinational firms rather than local economies.
Public acceptance depends on trust. Trust is built when companies are candid about resource use, when municipal approvals are conservative and enforceable, and when communities are compensated fairly for any direct burdens. Where that trust breaks down, opposition has shown it can halt even billion‑dollar projects.

What measurable progress looks like​

A credible, forward‑looking approach would include measurable, verifiable milestones:
  • Full‑scope water audits for each data centre campus, independently verified and updated annually.
  • Mandatory real‑time metering (or at least monthly reporting) of potable and non‑potable draws published by municipalities.
  • Cooling technology transition plans embedded in permits, with sunset clauses for high‑water approaches.
  • Financial surety (bonds) that can be drawn on if water usage surpasses agreed thresholds in ways that harm municipal supply.
These mechanisms allow both parties — municipalities and operators — to manage risk without stopping innovation.

Risks to watch​

  • Cumulative impact: One facility alone may be manageable, but multiple projects in a watershed can produce systemic stress.
  • Climate volatility: Historical water availability is a poor predictor of future supply under climate change; planning must assume hotter, drier summers and more intense demand peaks.
  • Transparency gap: Without consistent disclosures, community monitoring is impossible; lack of data is itself a major risk factor.
  • Technological complacency: Overreliance on promises of future water‑efficient designs without enforceable commitments creates regulatory and reputational exposure.
When these risks compound, the result can be either under‑resourced municipalities, curtailed agricultural or domestic water use, or costly retrofits and legal challenges that erode the supposed economic benefits.

Conclusions​

Canada stands at a crossroads between capturing economic value from the AI infrastructure boom and protecting its most precious public good: water. The physical reality is simple — high‑performance AI infrastructure requires both power and cooling. The policy question is not whether data centres should exist in Canada, but how they should be regulated and designed so their presence does not impair human and ecological water needs.
Practical steps are available and proven: conservative permitting, mandatory metering and public reporting, conditional approvals tied to cooling technology, and state/provincial support to strengthen municipal technical capacity. When municipalities insist on rigorous, enforceable water plans and when operators invest in low‑water cooling and full transparency, it is possible to reconcile local prosperity with water stewardship.
Failing to put those guardrails in place risks a future in which short‑term economic incentives give way to long legal battles, community backlash, and avoidable pressure on drinking water systems. The technology that powers modern life should not be built on hidden consumption; it must be engineered and governed to preserve the basic resources that sustain communities.

Source: CBC https://www.cbc.ca/news/ai-data-centre-canada-water-use-9.6939684
 
On a sun‑baked summer day in Nanaimo, B.C., neighbours rallied not against a new mall or highway but against a proposed 200,000‑square‑foot data centre whose projected thirst for municipal drinking water has crystallized a growing national debate: Canada is courting hyperscale AI infrastructure with generous incentives and a cool‑climate sales pitch, yet scrutiny of water impacts, metering and enforceable oversight remains thin.

Background / Overview​

Canada’s appeal to cloud and AI builders is straightforward: abundant low‑carbon electricity in many provinces, relatively cool ambient conditions that aid air‑cooling, and active federal and provincial outreach to attract investment. That economic case, championed by municipal leaders eager for capital projects and perceived “jobs of the future,” sits uneasily beside a growing body of reporting and research showing large AI‑optimised data centres can place material demands on local water systems — sometimes measured in millions or even hundreds of millions of litres per year.
At the same time, the architecture of modern AI workloads — dense GPU clusters running continuous training and large‑scale inference — changes the cooling calculus. Where earlier web workloads could rely on intermittent free‑air economization, generative AI’s power density and duty cycle often push operators toward water‑based cooling or closed‑loop liquid systems that still require water for heat rejection or for the wider electricity supply chain. Independent technical assessments and industry reports show that while estimates vary widely, the aggregate water footprint of the data‑centre + electricity system is not trivial and is poised to grow with the AI build‑out.

How much water do AI data centres actually use?​

The numbers are variable — and opaque​

Quantifying water use for a single facility or an entire sector is complicated by inconsistent reporting, varying definitions (withdrawal vs consumption vs evaporative loss), and the mix of direct on‑site uses versus indirect water tied to electricity generation. Some credible international compilations estimate very large-scale figures: one authoritative energy sector dataset maps electricity‑related water consumption across fuels and regions, underscoring how much of the data‑centre water footprint is embedded in power generation rather than on‑site cooling alone.
Independent technical writers and industry analysts also highlight the problem of opacity: operators publish sustainability claims and high‑level targets, but few provide per‑facility, audited water‑use numbers accessible to municipal regulators or the public. That gap leaves communities to rely on developer forecasts, sometimes optimistic vendor assertions, and patchwork municipal permitting documents.

Representative figures and common benchmarks​

  • A 2023 sector estimate compiled by energy and environmental analysts put global data‑centre water consumption for cooling into the tens to low hundreds of billions of litres annually, but these figures vary by methodology. The IEA and related energy datasets show the electricity sector’s water footprint is a major driver of that total.
  • Technical investigations have produced a range of illustrative numbers: one study cited in mainstream coverage estimated that 20–50 ChatGPT prompts could correspond to roughly 500 millilitres of water when accounting for both the power generation and cooling footprints — a headline‑friendly figure that captured public attention but masks substantial geographic and methodological variation. Other rigorous papers find far smaller per‑query water figures for inference on modern equipment, again underscoring volatility in the estimates. Interpretation matters.
  • Municipal permitting documents for some Canadian projects quantify allocated cooling water in volumetric flow terms; one Ontario project was authorized planning documents citing up to 39.75 litres per second — a rate that converts to roughly 1.2 billion litres per year if operated continuously at that maximum. Those permit figures have become a focal point for local concern.

Why the range is wide​

  • Cooling architecture: ambient air economization, closed‑loop liquid cooling, evaporative cooling towers, and immersion systems have very different consumptive profiles.
  • Climate and seasonality: in cold northern climates, free‑air cooling is feasible much of the year; in hot, arid regions, water‑intensive cooling is the default.
  • Grid mix and indirect water: the water required to generate a kilowatt‑hour depends on the local mix of hydro, nuclear, gas, coal and renewables; this drives a substantial share of the “embedded” water footprint.
  • Ramp‑up and construction phases: initial estimates may exclude construction‑phase usage and operational ramping to full load, which has happened in other jurisdictions where actual use exceeded initial forecasts.

What’s happening in Canada: projects, promises and local fights​

Local flashpoints: Nanaimo, Etobicoke, Varennes​

  • In Nanaimo, residents including retired academic Kathryn Barnwell rallied municipal officials over a rezoned wooded lot proposed for a large data centre. Their central worry: potable municipal water would be used for evaporative cooling during hot spells, at a time when the region faces drought stress and competing needs. The Nanaimo story is not an isolated NIMBY episode; it reflects a broader question about whether municipalities have the capacity to evaluate the water and infrastructure implications of hyperscale builds.
  • In Etobicoke, planning documents for a Microsoft facility — referenced in media reporting and municipal filings — included an allowed draw of up to 39.75 litres per second for cooling. That allocation, when annualized at continuous operation, equates to well over a billion litres of potable water a year. Microsoft counters that new Canadian facilities are engineered to rely primarily on free‑air cooling and captured rainwater, and that municipal water will be used only under very high temperature or extremely low humidity conditions. Critics point to past cases elsewhere where projected water use proved optimistic.
  • In Varennes, Quebec, an Amazon Web Services facility has been flagged in reporting for lacking a dedicated potable water meter, leaving the municipality without granular visibility into its water draw. The owner currently pays a flat commercial rate rather than volumetric charges based on consumption; that arrangement has alarmed local advocates and regulators who argue a meter is essential to measure and manage risk.

Federal incentives and industry appetite​

Ottawa has signalled a desire to capture the economic upside of cloud and AI investment — including a federal programs skewed toward data‑centre attraction — and some provinces have actively courted hyperscalers. Those policy signals, combined with the supply of land and transmission corridors, have encouraged multiple hyperscale projects to acquire tracts of land across Canada. But the attention to energy interconnection often outpaces formal consideration of water impacts: energy regulators typically examine grid impacts and firming needs, while water permitting and metering remain largely municipal concerns with inconsistent capacities.

Technical responses: cooling architectures and real options to reduce potable water use​

Cooling strategies (and their water implications)​

  • Free‑air economization (air cooling): Uses low ambient temperatures to cool servers without water. Highly effective in cold climates and can eliminate on‑site potable use for much of the year if humidities and temperatures permit.
  • Closed‑loop liquid cooling and immersion: Moves heat efficiently from chips to a closed system; reduces evaporative losses but still requires a heat sink and may involve some water for the heat rejection loop or for ancillary systems.
  • Evaporative cooling (cooling towers): Very water‑intensive and common in hot climates due to efficiency and cost; uses potable or non‑potable water that is partly lost to evaporation.
  • Hybrid systems: Default to air cooling and engage water only under extreme ambient conditions; policy and permit language can require such modes as enforceable conditions.

Industry claims and real‑world deviations​

Major operators routinely assert that modern data centres use far less potable water than older facilities. Microsoft, for instance, has publicly emphasized design choices intended to minimize municipal potable draw — including air economization thresholds and rainwater reuse — and has stated that municipal water would be used only when outside temperatures exceed specified thresholds. But international precedents show these promises are not always borne out in practice: in one Dutch case, a facility initially promised modest annual potable consumption but later appeared to use multiples of that estimate during hotter periods, prompting local pushback and scrutiny. That history underscores the importance of enforceable, auditable reporting rather than declarative corporate commitments.

Governance gaps: why municipal metering and conditional permits matter​

The current oversight landscape​

  • Energy interconnection processes and provincial scaling often get rigorous scrutiny; water oversight is frequently devolved to municipalities that may lack the technical expertise or legal levers to impose conservative, enforceable conditions on data‑centre operators.
  • Metering is inconsistent. Where volumetric metering exists, municipalities can monitor and bill based on consumption; where flat fees or no meters exist, actual draw may be invisible to local regulators, eroding accountability. The Varennes example — a large cloud facility without a dedicated water meter — illustrates this gap.

Practical governance failures and consequences​

  • Overpromising at approval, followed by higher usage in hotter years or during full‑capacity ramps, has produced community backlash in multiple jurisdictions. That backlash can culminate in delays, legal challenges, reputational damage, or project cancellations — outcomes that could have been avoided with conservative permitting and binding monitoring requirements.
  • Without transparent, regular reporting (audited WUE/WUE metrics, peak day draws, and meter data), municipalities and residents are unable to verify corporate assertions about low water use or rainwater reliance.

What municipalities, provinces and Ottawa can — and should — do​

Immediate, practical steps for local governments​

  • Require dedicated volumetric metering and real‑time reporting for any large water allocation tied to a data‑centre permit. Metering is the baseline for enforceable management.
  • Make approvals conditional: binding clauses that force the operator to default to air‑cooling first strategies, restrict potable water use to narrowly‑defined emergency thresholds, and require alternate non‑potable sources or treated reclaimed water where feasible.
  • Mandate audited, public annual reports on water withdrawal and consumptive use with independent third‑party verification.
  • Require contingency plans for drought scenarios and explicit community benefit agreements that tie water usage and infrastructure costs to local investments.

Provincial and federal roles​

  • Provinces can harmonize permitting standards and set sectoral thresholds for water use that recognize watershed capacities and seasonal constraints.
  • The federal government, which is actively promoting data‑centre investment, can require grant or incentive recipients to meet minimum transparency and non‑potable water commitments as a condition of support. That would align public money with public accountability.

Technical and procurement levers​

  • Encourage or require use of non‑potable sources, treated wastewater, or industrial recycling for heat rejection where available and safe.
  • Fund municipal technical capacity building so smaller towns can independently assess proposed resource demands and negotiate enforceable conditions.
  • Condition public procurement and sovereign cloud strategies on independent, auditable WUE and PUE reporting for infrastructure located on Canadian soil.

Strengths, limitations and critical trade‑offs​

Strengths of the emerging industry approach​

  • Operators are investing in closed‑loop cooling, immersion and air economization technologies that materially reduce potable water dependence compared with traditional evaporative towers.
  • Large hyperscale builds can bring local economic activity, property tax bases, and indirect employment during construction. Those benefits are part of the economic pitch many municipalities find persuasive.

Key limitations and risks​

  • Transparency deficits reduce community trust. When meter data is lacking or corporate forecasts are optimistic, the social licence to operate erodes quickly.
  • Conditional designs and operational thresholds can be gamed if not paired with independent verification and robust permit enforcement.
  • The embedded water footprint of electricity generation remains a systemic constraint: even if a Canadian data centre uses little potable water on‑site, its electricity supply may rely on water‑intensive power plants in the broader energy system. That reality complicates simple narratives that “Canada is low water because it has hydro.”

What the research and reporting tells us — and what it does not​

  • Independent studies and energy agency compilations show data‑centre water impacts are real and can be large in aggregate, particularly when indirect electricity‑related water is included. The IEA’s sector data and technical reviews make clear the energy sector’s water footprint is a major driver of the total numbers.
  • Research on per‑query water use for generative models is highly sensitive to assumptions: some widely circulated figures (e.g., “500 ml per short ChatGPT session”) were useful as communication devices but vary greatly by model, location, and the baseline electricity mix. Robust policymaking should not rely on single headline numbers but on audited, location‑specific facility data.
  • Journalistic investigations in other jurisdictions have shown facilities sometimes consume far more water than early project documents suggested. Those precedents motivate conservative permitting and mandatory metering in Canada.

Actionable checklist for municipal councils and community advocates​

  • Demand a dedicated potable‑water meter and public disclosure of hourly/daily withdrawal data.
  • Require legally binding operational limits (e.g., air‑cooling as the default, potable draw only above specific, verifiable thresholds).
  • Insist on third‑party audits of water use and climate/energy modelling during the first two years of operation.
  • Negotiate community benefit agreements that internalize water‑infrastructure costs if the facility causes the town to invest in new supply capacity.
  • Coordinate regionally: neighbouring municipalities should share information and avoid siloed approvals that cumulatively stress shared watershed resources.

Conclusion: reconcile economic opportunity with custodial duty​

Canada has a legitimate strategic interest in hosting the infrastructure that will underpin AI services for businesses, governments and consumers. The country’s climate, electricity mix and public policy orientation make it attractive to hyperscalers seeking new capacity. But building economic opportunity on top of public goods — especially potable water — without robust, consistent oversight, metering, and enforceable conditions is a short‑term strategy that risks long‑term cost and reputational damage.
The path forward is practical and achievable: require metering and transparency, condition approvals on low‑water cooling defaults and third‑party audits, and align public incentive programs with enforceable environmental safeguards. Those steps preserve the benefits of investment while protecting the water systems that sustain communities.
The Nanaimo protest, Etobicoke’s permit allocations, and the Varennes meter gap are early indicators of a broader governance question: as AI infrastructure becomes more concentrated and more water‑sensitive, will Canada choose to compete on openness and accountability — or on permissiveness and short‑term revenue? The technical solutions exist to reduce potable water demand; the pressing question is whether public policy and municipal practice will insist on their enforcement.

Bold, enforceable transparency rules and a commitment to auditable metrics are not anti‑investment; they are the precondition for durable investment that communities will accept and regulators can defend.

Source: CBC https://www.cbc.ca/news/ai-data-centre-canada-water-use-9.6939684?cmp=rss