Two Canadian data center stories hit on the same morning — and one of them probably won’t be the one everybody leads with. This is the AI Daily Briefing. Meta plants C$13 billion in Alberta, DeepInfra quietly stands up inference capacity in Toronto — so what, exactly, is Canada selling? Same maple leaf, two completely different layers of the stack. Let's start with why the compute's suddenly heading north. This one's from The Hindu:
Tech giant Meta announced Wednesday it will build a massive data centre in central Alberta, the company’s first in Canada, as it rapidly builds out computing capacity to support the global AI boom. The 1-gigawatt data centre, which will be built with the ability to scale up to 1.8 gigawatts, will be located in Sturgeon County and represents a total investment of C$13 billion, or $9.17 billion, Meta said.
Meta's putting C$13 billion into Sturgeon County, Alberta — a 1-gigawatt build that can scale to 1.8. First facility they've ever had in Canada. Alberta makes sense here. Cheap power, probably a lot of it natural gas, and a provincial government that's been openly courting this exact kind of capital. The energy sourcing is half the deal here, even if the press release doesn't spell it out. Yeah — and notice what's missing from thirteen billion dollars of press release. Nobody's telling me what fraction of that 1.8 gigawatts is training Llama versus serving inference versus just sitting there as optionality. Why Canada, why now? No federal AI regulation on the books yet, and power's cheap. It looks like energy and regulatory arbitrage stacked on top of each other. And the map keeps spreading out. We started the week in Kentucky and Korea, ran through Australian tenders, and now two Canadian builds land in a single day. The map's expanding faster than the financing structures can keep up. This one's from Let's Data Science:
DeepInfra opened a 1.7 MW Toronto AI inference cluster on July 8, 2026, saying it is the company's ninth data center and first site outside the United States. The facility will host more than 1,000 NVIDIA Blackwell B300 GPUs for production inference workloads. For teams building agents, model APIs, search, and voice applications, the useful signal is geographic: inference capacity is spreading closer to users and data rather than staying concentrated in a few U.S. regions.
So DeepInfra's Toronto site: 1.7 megawatts, ninth data center, first one outside the US. And unlike the Alberta story we just hit, this one comes with a business model attached. Right. Meta's building a hyperscaler campus. DeepInfra's renting you inference by the token. Same country, same day, completely different layer of the stack. More than a thousand Blackwell B300s, all pointed at production inference. No training hall, no anchor tenant to chase — they're the tenant. They stand up capacity and sell access. Which is refreshing, honestly. And they run open-weight models at scale. So now a third-party inference provider is putting its own iron in a new geography, and the debate over who controls the inference stack is showing up in Canada — not just Virginia. What I actually want, though, is day-one utilization. A thousand B300s is a real capital commitment. Selling tokens is a real model — but only if the boxes stay warm. Got a question, a story idea, or a correction for us? Send it to aidailybriefing at lantern podcasts dot com. We read every note, and your feedback helps shape the show.
We’ve put links to every story from today’s briefing in the show notes, so if something caught your ear, you can dig into the original reporting there. That’s AI Daily Briefing for today. This is a Lantern Podcast.