This is an AI generated image, but it may very well represent the to-date somewhat vague vision that is in the heads of some of our politicians and industry leadership: an Ontario that provides data centre services to the world; one that is, to paraphrase a certain provincial expression, open for data centre business. Rhetoric is ratcheting up, despite some of the clearest and deepest public opposition to any issue in recent memory. With the recent release of Ontario’s new data centre playbook, what more do we know?
This much is clear: Ontario has moved to end automatic grid connections and proposed full cost recovery for large data centres. The 10 gigawatt queue the province estimates still lacks a demand model, a capacity ceiling and a test separating domestic need from export hosting. The government appears to have answered one of the questions at the centre of the data-centre fight, with a tiny tip of the hat to public opposition: every project asking for electricity shouldn’t de facto receive it.
It hasn’t answered the other, more important one: how much data-centre capacity does the province actually need? And how is that broken out? Ontario went from an implicit “connect what asks” obligation to a ministerial gate, full-cost recovery for large loads, a preference for self-generation, and an assessment that at least names sovereignty, community impact, and electricity-system effects. That is more than most Canadian jurisdictions have done.
But a selection process without a capacity budget will still over-allocate. Each project can look individually attractive, jobs, local benefits, closed-loop cooling, paid-for upgrades, while the province accumulates several gigawatts that no one has shown Canadians actually need. That is the fourth pillar the draft policy does not have: demonstrated domestic requirement, broken out by use.
We’ve looked at this issue federally. On August 13, the Ontario government launched the framework for a new Data Centre Playbook. The draft creates a strategic assessment process for large data centres seeking connections to Ontario’s electricity grid. Projects would be evaluated for their economic contribution, data-security and sovereignty benefits, investments in host communities and impact on the electricity system.
The government has also proposed making new data centres pay the full cost of their electricity and grid impacts, potentially through a separate electricity rate for facilities larger than one megawatt. Projects that supply some of their own generation would be favoured. Ontario says it will offer non-financial assistance such as expedited service and permitting support rather than direct financial incentives.
These are meaningful changes. Once implemented through the final framework and regulations, they would replace a system in which utilities were generally required to connect projects with one in which large loads can be selected according to provincial priorities. They also recognize that electricity used by data centres has an opportunity cost and that households and other industries should not be required to subsidize it.
The Playbook provides much of the approval machinery that has been missing from Canadian data-centre policy. But it does not provide a number. It also looks at sovereignty as ownership and grid when it should contemplate a more expansive view including weights, foreign control and access to data, and other dependencies. These are substantial vulnerabilities and blind spots.
Ontario’s queue is larger than Canada’s estimated domestic requirement
Ontario reported in July 2025 that data centres seeking grid connections represented more than 10,000 megawatts of requested load, approximately 30 per cent of the province’s 2024 peak electricity demand and roughly the output of the Bruce Power nuclear generating station.
That queue should not be treated as demonstrated Ontario demand. It is a collection of requests from individual developers. It can include facilities intended to serve foreign customers, cloud workloads unrelated to AI, duplicated or speculative reservations and projects whose business cases may never survive financing or construction.
My previous analysis, “We Did the New Math. We Don’t Need More Data Centres,” estimated that Canada’s current daily text-prompt workload would require about 51 megawatts of provisioned capacity after peak demand and redundancy were included. Canada already operates approximately 337 megawatts of AI data-centre capacity.
The same analysis modelled the largest credible domestic-demand scenario: widespread use of compute-intensive agents by Canadian workers, combined with a sovereign frontier-training capability. That scenario reached approximately 1.5 gigawatts nationally.
Ontario’s connection queue alone is therefore approximately 127 times the estimated capacity required for current Canadian text inference and more than four times the maximal national scenario. These are not exact like-for-like comparisons. The Ontario queue includes cloud infrastructure and potentially export-oriented capacity, while my prompt calculation excluded image and video generation. The agentic multiplier could also move significantly in either direction.
Those differences strengthen the case for classifying the queue. They don’t justify treating 10 gigawatts of connection requests as 10 gigawatts of public need.
Three assessment pillars and a missing fourth
The draft Playbook is organized around three strategic pillars. Its stated vision begins from the position that data centres are critical infrastructure and Ontario should embrace their expansion as an economic opportunity. The assessment is designed to manage that expansion rather than determine its required aggregate scale.
The first is economic development. Projects should create high-quality jobs, generate tax revenue, use Canadian and Ontario suppliers and provide compute that makes Ontario businesses more productive.
The second is data security and digital sovereignty. Projects should protect high-sensitivity and critical data, support Canadian data residency and preferably be Canadian-owned and operated.
The third is community investment and public confidence. Developers should finance local infrastructure and training, secure local support and minimize water, noise, pollution and other environmental effects.
Each pillar asks whether a particular project is better than another project competing for electricity. None asks whether the province should allocate another block of electricity to data centres at all. That missing fourth pillar is demonstrated capacity need.
Without an aggregate electricity budget, Ontario could approve several gigawatts of individually attractive projects and reproduce the same overbuild one project at a time. A facility may create construction work, pay the full cost of a new transmission connection and use closed-loop cooling. Those features improve the project. They do not establish that its compute is required by Canadians or that electricity is being allocated to its highest-value public use.
The distinction is especially important because Ontario is also trying to electrify manufacturing, transportation, heating and other parts of its economy. The Playbook says data centres should not hinder strategic growth in sectors such as manufacturing. It offers no quantitative test for determining when that displacement begins.
The Playbook addresses some externalities
Ontario has responded to several of the strongest objections raised by communities.
The government proposes a higher electricity rate for new data centres above one megawatt and is considering restricting their participation in the Industrial Conservation Initiative. Developers would be assessed on plans to build their own generation, finance required grid reinforcement and recover the full cost of their system impact. The framework also favours closed-loop or waterless cooling, direct-to-chip liquid cooling, noise-reduction engineering and waste-heat recovery.
Host communities are expected to receive significant financial and non-financial benefits, potentially including roads, broadband, electricity upgrades, training programs, apprenticeships and community facilities. If implemented effectively, these measures could reduce the transfer of costs from developers to ratepayers and give municipalities more leverage over projects that consume local land, water and administrative capacity.
They operate on the environmental and economic-impact side of the debate. The capacity question survives every one of them. A fully self-financed, water-efficient and locally supported data centre still consumes power and grid capacity. The fact that a project can mitigate its harms does not establish a domestic requirement for the service it produces.
Data residency is only one layer of sovereignty
The Playbook’s sovereignty language is more developed than a claim that locating servers in Ontario automatically creates Canadian control. Its preference for Canadian ownership and operation is important, as is its focus on high-sensitivity data and critical infrastructure.
The framework still concentrates on where data is stored. It does not require Canadian possession of model weights, independence from foreign model-access controls or Canadian control over the full technical stack. And any American hosting is potentially subject to Cloud Act provisions and access.
A data centre in Ontario can run an API supplied by a foreign company. If that company or its government restricts model access, the Ontario facility can lose the underlying intelligence service while retaining the building, chips and stored data. Domestic data residency does not, by itself, remove a foreign model-access chokepoint.
A stronger sovereignty test would ask whether Canada can continue operating the workload if a foreign provider withdraws service. That requires attention to the operator, cloud layer, model licences, weights, training data, orchestration software and jurisdiction governing each component.
The phrase “preferably Canadian-owned and operated” also leaves discretion to approve foreign-controlled hyperscalers. The final Playbook should explain what compensating controls would be required when a project does not meet that preference.
AI’s economic benefit is not a data-centre demand forecast
Ontario describes the Playbook as an early component of its coming AI strategy and connects it to an estimated $122 billion in economic growth by 2035 and more than 17,000 new jobs per year from AI adoption.
Those figures describe the expected value of using AI across the economy. They don’t establish that the same value depends on constructing a corresponding volume of data centres in Ontario.
An Ontario manufacturer can become more productive using a model served from existing Canadian capacity or another jurisdiction. A local software company can build an export business without owning the facility that runs every inference. AI adoption may justify ensuring reliable access to compute; it does not convert every proposed megawatt into necessary provincial infrastructure.
The Playbook partly recognizes this distinction by asking developers to quantify the jobs, capital expenditure, tax revenue and local compute access created by their individual projects. The final assessment should go further by separating temporary construction employment from permanent operating jobs and distinguishing benefits produced by the facility from benefits attributed to AI adoption generally.
The forecasts have to reconcile
Ontario does have a forecast for data-centre electricity consumption, even though it has not converted that forecast into a capacity budget.
The IESO’s 2026 Annual Planning Outlook projects that Ontario data centres will consume 10.32 terawatt-hours in 2035 and 21.64 TWh in 2050 under its reference scenario. Its 2050 low-demand scenario is 14.12 TWh and its high-demand scenario is 28.38 TWh.
Expressed as continuous average load, those 2050 forecasts are approximately 1.61 GW, 2.47 GW and 3.24 GW respectively. Average load is not the same as provisioned capacity or the maximum connection approved for a facility, but the conversion makes the scale easier to compare.
The federal numbers are even broader. The Canada Energy Regulator’s 2026 scenarios add 3.5 GW of Canadian data-centre load by 2050 under its reference assumptions, 1.5 GW under the lower scenario and 12 GW under the higher scenario. The regulator explicitly says existing models cannot yet reliably estimate long-term data-centre demand.
Ontario’s official consumption forecast is therefore not larger than the federal high-demand forecast. Ontario’s proposed connection pipeline, now estimated at more than 10 GW, is larger than the federal reference forecast for the entire country. Ontario’s reference consumption forecast is also larger than the 1.5 GW maximal domestic AI scenario in my previous analysis.
That does not automatically make any of these calculations wrong. They measure different things. My calculation estimated the provisioned capacity required for specified Canadian AI workloads, including widespread agent use and sovereign frontier-model training. The IESO forecasts electricity consumption across the commercial data-centre sector based primarily on known projects. That sector also includes cloud computing, storage, enterprise software, streaming, colocation and potentially a large volume of foreign or export-oriented computing.
The difference is nevertheless too large to leave unexplained. A project-pipeline forecast estimates what developers may attempt to build. It does not establish what Ontario or Canada needs. Using proposed projects as the principal input can also make the forecast partly self-generating: more speculative connection requests produce a higher demand forecast, which is then used to justify building enough electricity infrastructure to serve those requests.
Ontario’s final framework should require a public, reproducible usage forecast containing at least:
- Existing consumption and incremental growth, reported separately.
- Connection capacity, expected coincident peak demand, annual energy consumption and average utilization, rather than moving interchangeably between MW and TWh.
- IT equipment load and total facility load, including cooling, power conversion and redundancy.
- Separate forecasts for cloud services, storage, enterprise computing, AI inference by modality, AI training, agentic workloads, cryptocurrency and other major uses.
- Canadian workloads, foreign workloads and export hosting, reported separately.
- Public-sector, consumer and commercial requirements, including the capacity reserved for critical infrastructure and sovereign services.
- The proportion of requested capacity associated with operating, financed, approved, preliminary and speculative projects, with probability adjustments for projects unlikely to materialize.
- Assumptions about chip efficiency, model efficiency, utilization, power usage effectiveness, workload growth and the possibility that efficiency improvements increase total use.
- Hourly and seasonal demand profiles, regional and transmission-zone impacts, ramping behaviour, on-site generation and storage, and the amount of load that can be shifted or curtailed.
- Low, reference and high scenarios that are reconciled annually against metered consumption and actual project completion.
Without these categories, Ontario cannot tell whether it is forecasting necessary domestic digital infrastructure, a commercial export-hosting industry or a speculative development queue.
BC started with an allocation
British Columbia has taken a structurally different approach. Its current competitive process places AI and data-centre projects inside an allocation of up to 400 megawatts, with a maximum project size of 145 megawatts. BC Hydro will determine how much of the allocation to award based on its supply-and-demand assessment.
That approach establishes an aggregate boundary before projects compete within it. Ontario has created the competition without setting the boundary. Ontario hasn’t, and does not necessarily need to, adopt British Columbia’s exact number. Our economy, electricity system and existing infrastructure are different. But does need a published capacity budget built from a reproducible demand model.
At minimum, the final Playbook should disclose:
1. Ontario’s projected domestic inference, agentic and training requirements for 2030 and 2035.
2. The additional reserve required for security, redundancy and sovereign capability.
3. How much proposed capacity is intended for Canadian workloads and how much is intended for export.
4. The amount of electricity Ontario is prepared to allocate to each category.
5. The conditions under which that allocation would be increased.
This would allow the province to support a hosting industry openly as industrial and export policy while reserving the capacity needed for Canadian sovereignty. It would also let municipalities and ratepayers evaluate the opportunity cost using the same numbers as developers and government.
The Ontario government is accepting public comments on the draft framework for 30 days beginning August 13, 2026. The proposed separate electricity rate, connection threshold and related measures remain subject to consultation or future regulation.
What Ontario should negotiate for every megawatt
A capacity budget should be accompanied by standard connection conditions. The province is negotiating access to a scarce public system, not merely approving another industrial building.
First, Ontario should establish service priorities before scarcity occurs. Critical Canadian workloads supporting health care, emergency response, public administration, financial infrastructure and security may require firm service. Batch AI training, cryptocurrency and export-oriented computing can often be delayed or moved. Those flexible workloads should receive interruptible or curtailable service and should be required to reduce demand first during system constraints.
This should be contractual, not improvised during an emergency. Operators claiming critical-service status should have to identify and verify the workloads that require it. A facility should not receive priority for its entire load because a small portion supports essential services.
Second, the proposed exclusion of new data centres from the Industrial Conservation Initiative should not eliminate the incentive to reduce peak demand. Ontario should replace it with a data-centre-specific flexibility requirement, including time-sensitive and location-sensitive pricing, minimum demand-response obligations, automated curtailment capability and higher charges for firm 24-hour service.
Third, full cost recovery should extend beyond the direct connection. Developers should pay for transmission and distribution reinforcement, additional generation and storage required by their load, congestion and system losses, operating-reserve requirements, power-quality mitigation and any additional emergency-service capacity required by the facility.
New generation commitments should be physically deliverable to the relevant part of the Ontario grid and operational when the data-centre load arrives. Renewable certificates purchased elsewhere should not substitute for deliverable supply.
Fourth, capacity reservations should carry deposits, construction milestones and use-it-or-lose-it deadlines. Developers should not be able to hold scarce grid capacity indefinitely or transfer a speculative reservation as an appreciating asset. Capacity that is not used within the agreed schedule should return automatically to the provincial allocation.
Fifth, community benefits should be binding and proportional to the scale of the project. Host agreements should include a transparent payment based on reserved capacity or actual electricity use, protection against undisclosed tax concessions, funding for roads, fire and emergency services, water and wastewater infrastructure where required, broadband, apprenticeships, local procurement and workforce training.
Commitments should be published, independently audited and subject to clawbacks if employment, investment or operating targets are missed. A project that reserves hundreds of megawatts while creating relatively few permanent jobs should not satisfy the community-benefit test through temporary construction employment alone.
Finally, operators should report actual electricity consumption, peak demand, water use, curtailment performance, permanent employment and the domestic or export share of their workloads. Commercial confidentiality may justify aggregating some data, but it should not prevent Ontario from showing whether the industry is developing within the capacity budget used to approve it.
The province’s plan commits to not promising every applicant uninterrupted electricity and then ask communities to absorb the resulting infrastructure plan. It should also decide which uses justify firm service, which uses can be flexible, who pays for the system built around them and what host communities receive in return.
