The editorial reframes the story away from climate framing toward grid-bound compute, noting that every hyperscaler has signed a nuclear deal in the last 18 months (Amazon-Talen, Microsoft-Constellation/TMI). It cites Hydro-Quebec turning away crypto and AI training tenants because it physically cannot serve them, suggesting Canada's plan is positioning for the same demand wave.
Points to IESO revising its 2050 Ontario demand forecast upward by roughly 75% due to electrification, industrial reshoring, and data centers. Argues the federal plan formalizes what provinces were already doing piecemeal because the load curve no longer resembles the previous decade.
Surfaced the CBC story to a technical audience where it drew 467 points and 318 comments, framing Canada's 10-reactor plan as a serious national commitment. The editorial reinforces this by noting Darlington's BWRX-300 is on track to be the first grid-scale SMR in any G7 country, with the federal strategy treating it as a replicable template rather than a one-off.
The federal government has set out a national nuclear strategy that envisions building up to 10 new reactors over the next 15 years, spanning large units, refurbishments, and small modular reactors (SMRs). The plan, reported by CBC, formalizes what has been happening in piecemeal fashion across provinces — Ontario Power Generation's BWRX-300 at Darlington, Bruce Power's expansion studies, New Brunswick's SMR roadmap at Point Lepreau, Saskatchewan's siting work with SaskPower — into a coordinated federal posture with Atomic Energy of Canada Limited (AECL) re-cast as a more active anchor.
The timing is not subtle. Canadian grid operators have spent the last 18 months publishing load forecasts that look nothing like the previous decade. The Independent Electricity System Operator (IESO) in Ontario revised its 2050 demand forecast upward by roughly 75% in its most recent annual planning outlook, citing electrification, industrial reshoring, and — increasingly — data centers. Hydro-Quebec, historically a net exporter, has been turning away large industrial loads, including crypto miners and AI training tenants, because it physically cannot serve them.
The Darlington SMR, a GE Hitachi BWRX-300 with a target output of around 300 MW, broke ground in 2025 and remains on track to be the first grid-scale SMR in any G7 country. The federal strategy treats it as a template rather than a one-off.
The story most coverage tells is climate policy. The story senior infrastructure people should be reading is grid-bound compute.
In the last 18 months, every hyperscaler has signed a nuclear deal. Amazon agreed to co-locate data centers at Talen Energy's Susquehanna site (an initial behind-the-meter arrangement up to 960 MW, later trimmed by FERC, with subsequent re-filings stretching toward roughly 1.92 GW of contracted offtake across phases). Microsoft signed a 20-year PPA with Constellation to restart Three Mile Island Unit 1 — 835 MW, target online 2028, rebranded as the Crane Clean Energy Center. Google contracted with Kairos Power for SMR capacity. Meta has issued an RFP for up to 4 GW of nuclear. The implicit thesis across all of these deals is that wind and solar plus batteries cannot, on any reasonable build schedule, supply the round-the-clock baseload that frontier-model training and inference require.
Canada's plan is a national-scale version of the same bet, with one important asterisk: nation-states build nuclear on different cost curves than utilities chasing a single tenant. Vogtle Units 3 and 4 in Georgia came in roughly $17 billion over budget and seven years late. The Hinkley Point C project in the UK is tracking similarly. The BWRX-300 design is supposed to be the answer — fewer custom parts, factory-built modules, simpler passive safety — but no one has actually finished one yet. Until Darlington produces electrons on a real grid at something close to its quoted ~$5 billion budget, the entire SMR economic case is a slide deck.
There is also a refurbishment story buried in the announcement. The CANDU fleet — Bruce, Darlington, Pickering B — has been undergoing mid-life refurbs that have, by industry standards, gone surprisingly well. Bruce Power's units are coming back online roughly on time and on budget. That track record is part of why the federal government feels comfortable putting AECL back in the driver's seat for new builds. It's also part of why the practical near-term capacity adds will come from extending existing CANDUs and uprating Bruce, not from greenfield SMRs.
Community reaction on Hacker News (the thread cleared 460 points) split predictably between "finally" and "15 years is the cope timeline." Both are right. Nuclear projects of this scale routinely slip — but the alternative path, in a country with this load growth, is gas peakers and a carbon-intensity number that breaks every existing corporate Scope 2 commitment.
If you choose AWS regions, GCP zones, or Azure regions for a living, this is now a multi-year planning input. Canadian regions — `ca-central-1`, `ca-west-1`, `northcentralus` — are quietly becoming a hedge against US grid constraints in Virginia (PJM interconnection queues are years deep) and against the increasingly likely outcome that some US data center clusters will end up burning natural gas to meet AI demand. If your sustainability report has to show year-over-year carbon intensity improvement on inference workloads, where you place model serving matters more in 2027 than it did in 2024.
For anyone building latency-sensitive products into Canadian users, the corollary is that capacity in `ca-central-1` is likely to grow faster than it has historically. Quebec hydro is tapped out, but Ontario's nuclear-heavy mix has headroom and is about to get more. Expect new availability zones and specialty instance families (especially GPU) to land in Canadian regions over the back half of the decade.
For anyone running on-prem or hybrid in the US Northeast, the relevant signal is the opposite: PJM and NYISO interconnect queues for new large loads are now multi-year, and several utilities have moved to deny or defer new data center connections. The 18-month timeline you assumed for capacity expansion is now closer to four years, and your renewal negotiation with your colo provider should reflect that.
The honest read: this plan will probably deliver 4–6 reactors on roughly the announced timeline, not 10. The Darlington SMR is the keystone — if it lands on budget and on schedule, the strategy becomes credible and Ontario becomes one of the most interesting jurisdictions in the world to host AI infrastructure. If it slips Vogtle-style, Canada will quietly fall back on CANDU refurbishments and gas, and the AI-compute migration story moves to whoever finishes a reactor first. Watch the 2029-2030 commissioning window. That's when this stops being policy and starts being electrons.
OK, so when does the first one come online? "The strategy calls for construction to start on two new large-scale reactors by 2035, for five more to be planned or under development by 2040 and for at least one reactor to be under construction outside Ontario by 2035."That's not serious
Always thought it was weird that the Commonwealth Realm nations had never pooled resources to have standardised reactor designs and expertise. Canada and Australia have loads of uranium - seems like an obvious strategic move. Instead, the UK turns to China, lol.
To my surprise Canada are actually quite ahead with the Darlington New Nuclear Project. There is a construction site [0] with work taking place. Not sure how Kairos Power are progressing in the USA. Nice job, Canada.0 - https://www.neimagazine.com/news/darlington-smr-secures-fina
Interesting to see the general opinion on nuclear swing so far from environmental and safety concerns (whether warranted or not) to pretty broad support for energy independence.I can't help but think its a sign that those concerns were easy to hold when energy was cheap and you could actually t
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
Makes alot of sense. Canada has:- one of the largest uranium reserves- a well respected and safe nuclear design in CANDU- experience with building and refurbishing nuclear reactors(Darlington)and for Ontario itself A need for more baseload to work with the large amount of solar and wind that Ontario