Compute infrastructure
This page maps major buildings and cloud regions that house servers for AI, cloud services, telecom networks, and high-compute workloads. Location matters because these facilities draw electricity, need cooling, and can affect local grid planning.
Evidence synthesis
The useful question is not whether data centres are simply good or bad. The evidence points to a narrower test: what each facility asks of the local grid, water system, public budget, and disclosure record.
The issue is not only national electricity totals. Large facilities create concentrated loads that can test local transmission, winter peaks, tariffs, and queue rules.
Who pays for connection, transmission, and any firm capacity the project needs?ISED's closed intake targeted large-scale AI data-centre proposals above 100 MW and asked proponents to show Canadian-client benefits, Indigenous participation, environmental impacts, supply-chain fit, readiness, and cost structure.
Which proposed projects move from intake to an MOU, and what public benefits are attached?Hydro- and nuclear-heavy grids can make operating emissions much lower, but scarce clean electricity still has an opportunity cost for homes, industry, exports, and electrification.
What public value is created in exchange for priority access to clean power?Cooling water cannot be read from a national average. Direct cooling, indirect electricity-related water, local basin stress, and drought protocols all change the answer.
What water source, cooling design, and drought rule applies to the specific site?A serious assessment separates location-based grid emissions, market-based renewable claims, construction, cooling plant, servers, chips, and replacement cycles.
Is the claim about annual power use, real-time grid impact, or full life-cycle impact?Construction work, tax revenue, and procurement can matter. Permanent operating jobs are usually smaller, and public subsidies need a clear benefit test.
How many permanent jobs, what local procurement, and what enforceable community benefit are attached?Facility-level MW, PUE, WUE, water source, backup generation, hourly load shape, and effective clean-energy matching are often missing from public records.
What is reported by facility, and what remains modelled or unknown?Policy watch
The clean-grid advantage is real, but the public question is shifting toward who gets scarce firm power, what is disclosed, and what local benefit is enforceable.
The table below is easier to read after the provincial concentration is visible.
Different sites create different planning questions for power, cooling, connectivity, and sovereignty.
The map uses these type labels for filtering and comparison.
The planning burden changes when proposed sites are read separately from facilities already operating.
81 operating facilities with estimate ranges.Impact helper model
6 planned or proposed facilities with estimate ranges.Impact helper model
Province-level estimates combine facility count with the local grid context.
21 operating facilities; grid intensity 1.9 gCO2e/kWh.
33 operating facilities; grid intensity 59 gCO2e/kWh.
13 operating facilities; grid intensity 18 gCO2e/kWh.
11 operating facilities; grid intensity 438 gCO2e/kWh.
2 operating facilities; grid intensity 234 gCO2e/kWh.
Electricity, emissions, and water ranges should be read as separate planning clues.
Annual electricity use is estimated from reported capacity where available and broad proxy assumptions elsewhere.
AI Canada Pulse impact helperLocation-based emissions depend heavily on each province electricity grid.
ECCC grid factorsWater values are broad because cooling systems and disclosure levels differ by facility.
LBNL/DOE assumptionsComparable impacts help readers understand scale, but only when the estimate range is narrow enough to avoid false precision.
The estimate range is wider than 10x; a single household count would imply false confidence.
Statistics Canada, Households and the Environment Survey: Energy use, 2021The estimate range is wider than 10x; a single vehicle count would imply false confidence.
U.S. EPA, typical passenger vehicle emissionsThe estimate range is wider than 10x; a single water-use count would imply false confidence.
Statistics Canada, residential water use, 2021Strategy context
The national AI strategy promises sovereign compute, a public supercomputer, and SME access. This map shows infrastructure context; the tracker separates those commitments from proof of delivered capacity.