Bell Canada Announces $52.5 Billion AI Hub in Saskatchewan, Targeting 1.2 Gigawatts Capacity
Canada commits to sovereign AI compute with a major new hub, while global data center development pivots to "speed-to-power" amid surging demand.
The story
The global race for AI compute capacity saw a significant commitment this week as the Government of Canada announced support for Bell Canada's expansion of its Bell AI Fabric in Saskatchewan. This ambitious project, valued at up to $52.5 billion in capital investment, aims to establish a 1.2-gigawatt AI infrastructure hub in the province, with 900 megawatts of new capacity planned.
This move underscores Canada's push for sovereign computing capabilities, ensuring critical infrastructure operates under domestic control. Meanwhile, in Jordan, Mindstream Energy is advancing its Al-Risha Sovereign AI and Digital Infrastructure Campus.
Announced on September 15, this project is designed to scale to approximately 400 megawatts, with an initial 70 MW deployment targeted for Q2 2027, leveraging on-site natural gas generation. These large-scale infrastructure developments highlight the ongoing shift in data center strategy, where power availability and rapid deployment are becoming paramount. Industry analysts note that "speed-to-power" is now a key factor in site selection, as world data center power demand is projected to reach 132 GW globally in 2026.
Silicon
MTIA 450 (Arke)
Maker: Meta
What: Third-generation custom AI processor; prototype testing completed with performance within 2-3% of simulations.
For Whom: Meta's own data centers for AI inference workloads.
The build-out
| Project | Who | Scale | Where |
|---|---|---|---|
| Bell AI Fabric expansion | Bell Canada, Government of Canada | Up to 1.2 GW AI infrastructure hub; $52.5 billion capital investment | Saskatchewan, Canada |
| Al-Risha Sovereign AI and Digital Infrastructure Campus | Mindstream Energy | Designed to scale to ~400 MW; initial 70 MW deployment Q2 2027 | Jordan |
Supply & policy signals
AI leaders (Anthropic, OpenAI, xAI) call for slower advanced model development.
Implication: Expected near-term pressure on chip stocks, but limited long-term impact on computing infrastructure demand due to persistent shortages.
Global data center power demand projected to reach 132 GW in 2026, with the US facing a 49 GW shortfall by 2028.
Implication: Shift in data center development focus to "speed-to-power" delivery, influencing site selection and procurement.
AI chip production identified as the primary constraint on AI compute buildout in 2026, with Google reportedly missing 2026 production targets due to insufficient manufacturing capacity.
Implication: The bottleneck for AI scaling has shifted from data center power to chip production itself, impacting growth timelines.
What we'll be watching
- Microsoft's official unveiling of its Maia 300 AI chip.
- Customer sampling of Intel's Crescent Island data center GPU in the second half of 2026.
- SpaceX's $13.3 billion annual AI compute deal commencing revenue generation in December 2026.
- Meta's installation of its MTIA 450 (Arke) custom AI processors in data centers in H1 2027.
- Further details on Nvidia's Rubin Ultra and Feynman chip roadmap.
Reporting + analyst voices: grounded via Google Search at publish time.