@EvanLSolomon We can, and need, to be building data centers here in Canada. But "sovereign" became word that means nothing. Cohere just merged into a $20B transatlantic . A data center can sit on Canadian soil and still be, in every way that counts, an American asset.
https://t.co/2NFm4t3s7i
@build_canada "Sovereign AI" means ten things to ten people, and a data center built in Canada, can, for all intents and purposes, be an American asset. Build the compute here, own the control layer. Soil's only where it starts.
https://t.co/2NFm4t3s7i
@MickeyDjuric We should build data centers in Canada β we need to. But "sovereign AI" means ten things to ten people. A data center on Canadian soil can be the answer, or it can be an American asset in every way that counts. Soil isn't sovereignty.
https://t.co/hK9eQEiC1F
@politico We should build data centers in Canada β we need to. But "sovereign AI" means ten things to ten people. A data center on Canadian soil can be the answer, or it can be an American asset in every way that counts. Soil isn't sovereignty.
https://t.co/hK9eQEiC1F
@globeandmail Bullish on Canadian data centers β the opportunity is real and it's right now. "Sovereign AI" means ten things to ten people, so let's go in eyes open: build the compute here, own the control layer. Soil's where it starts.
https://t.co/2NFm4t3s7i
@EvanLSolomon We should build data centers in Canada β we need to. But "sovereign AI" means ten things to ten people. A data center on Canadian soil can be the answer, or it can be an American asset in every way that counts. Soil isn't sovereignty.
https://t.co/2NFm4t3s7i
We can, and need, to be building data centers here in Canada. That said β "sovereign" has quietly become a word that means whatever you need it to.
Cohere, our designated national champion, just merged into a $20B transatlantic entity. A data center can sit on Canadian soil and still be, in every way that counts, an American asset.
https://t.co/0X5RMCqXCO
@CTVNews We should build data centers here in Canada and we need to. But sovereign AI means 10 different things to 10 different people and a data center built on Canada soil can be the solution to this, or it could be in all aspects an American asset.
https://t.co/hK9eQEiC1F
I don't think so.
That facility is rated A because Microsoft is the A-credit sitting at the end of the contract. Strip Microsoft out and there's no investment-grade anything. The rating isn't IREN's. It's borrowed.
Which is exactly the point. Hand my openclaw a Microsoft offtake and he'll print the same paper at the same spread.
The one piece that isn't pass-through is the part IREN itself led with β they own the data centers the GPUs sit in. That's real, and it's the whole move: stop being a conduit, become the developer that owns the site, the power, the cost basis. So this doesn't prove the GPU-landlord model has a durable edge. It proves that owning infrastructure while renting someone else's balance sheet is the only version that finances at all.
https://t.co/5dAmO0M4db
That's partially my point.
Powered land is a real moat today β but it's a power moat, and power normalizes. Queues clear, the 2022-vintage IPP requests come online, renewables-plus-LDES start showing up as data center supply. Whatever scarcity rent IREN and others paid extraordinary sums to capture compresses on a 2-3 year horizon.
The deeper issue isn't supply timing, though. A neocloud with no owned infrastructure is a conduit, not a counterparty. Strip away the powered site and the GPUs and what's left is a spread on someone else's credit β the hyperscaler or chipmaker actually standing behind the contract. As the credit-wrap gap closes, that spread compresses to zero. Nothing durable to own.
Which is why "unless vertically integrated" is the whole game. The neoclouds that survive stop being conduits and become developers β they own the land, the interconnection, the generation, the cost basis. Everyone else is a bridge financing structure wearing a compute company's clothes.
https://t.co/5dAmO0M4db
The neocloud boom case feels too supply-side to me.
Yes, compute demand is enormous. But Iβm not convinced the natural owner is an independent GPU landlord. The AI labs and hyperscalers are already building internal power, real estate, procurement, and development teams. For strategic capacity, they can usually finance cheaper than a neocloud because the credit is ultimately tied to the customer, not the intermediary.
So the question is less βwho will build all this?β and more βwhy does this balance sheet deserve to exist in the middle?β
Neoclouds make sense for overflow, speed, niche customers, or temporary scarcity. But at scale, credit quality becomes the issue. If your assets are GPUs with fast depreciation, your customers are concentrated, and your leases depend on continued model demand, that is not the same as an IPP with contracted power infrastructure.
The boom may happen. Iβm just not sure the terminal structure is a long tail of valuable neoclouds rather than labs/hyperscalers internalizing the best sites and leaving intermediaries with financing risk.
@globepolitics "Domestic Firms" is so nebulus - we've seen Cohere receive federal funding for Sovereign AI and then merge with a German AI leader; the same way a data center can be located in Canada and for all intents and purposes, be an American asset.
https://t.co/hK9eQEiC1F
@build_canada@UofT@professor_ajay Really looking forward to their sovereign data center strategy. Sovereignty has many different meanings.
https://t.co/hK9eQEiC1F
@TheGeorgePu The issue with "Sovereign" is there isn't any definition; Globe & Mail is reporting that there will be funding for "Domestic Firms" but Canadian AI has many different flavours.
https://t.co/hK9eQEiC1F
Canada's national AI strategy lands this week. One of its six pillars is building the Canadian sovereign AI foundation β sovereign compute, under Canadian governance.
Watch that last phrase. A data centre on Canadian soil is real and worth having: construction jobs, power demand, a domestic backbone. But location is the easiest layer to deliver, and it quietly stands in for the harder ones. Governance is a property of control, not coordinates β who can switch the system off, whose courts reach inside it, who owns the model and the data it produces.
So the test is not whether the strategy funds buildings. It is whether it names a mechanism β lease, offtake, equity, legal structure β for the layers that actually decide control. Soil is the foundation. It is not the whole claim.
I think βclear the runwayβ has to mean something concrete.
The bottlenecks are power, permitting, interconnection, and demand certainty.
TELUS is a useful example: the project has an MOU, but the government release says no funding has been committed, and the harder question is still who fills the capacity.
Private companies still need to take development risk. Governmentβs role is to make that risk financeable: identify where 100MW+ loads can connect, accelerate grid upgrades, standardize approvals, define sovereignty requirements, and create credible anchor demand where public workloads actually need Canadian control.
Pension and infra capital can fund this. But they need sites, timelines, power, counterparties, and rules they can underwrite.
The real opportunity is ownership, not just siting.
If First Nations are treated as a path around suburban NIMBYism, it will fail. If they are long-term infrastructure partners through powered land, equity, revenue sharing, training, and governance rights, data centres can become durable economic development.
Many First Nations are already central to Canadaβs next generation of energy and transmission projects. The better frame is not βhost community.β It is infrastructure partner: land, power, equity, revenues, and governance.