AI on the Horizon: August 27, 2026

Data Centres and Neoclouds

Canada’s latest AI news on the economy, society, and policy. In this issue of AI on the Horizon, we look at the rules being drafted for domestic data centres and the new cloud computing mania.

Everyone wants a playbook

Ontario released a draft data centre playbook outlining a framework for how the province will decide which large data centres get to plug into the grid. It identifies three guiding principles (investment in Ontario and local communities; covering the full cost of electricity; accepting non-financial support alone) and rests on three strategic priority pillars (advancing economic development, protecting data security and digital sovereignty, and investing in communities and earning public confidence).

Projects must pay the full cost of their power, including the grid upgrades they trigger. Given that Ontario is not offering any cash incentives, only non-financial support, and it wants a separate, higher electricity rate for any data centre above 1 MW (Quebec applies its higher rate above 5 MW), the message to developers is clear – Ontario wants you to bring your own energy rather than lean on everyone else’s.

Ontario’s playbook is not alone in seeking to establish control over data centre construction and power supplies. Quebec introduced tightly controlled application process for large blocks of data centre power in 2025, and Hydro-Québec has proposed to roughly double the cost for new data centres. Alberta’s system operator capped large load connections at 1,200 MW through 2028, and that capacity is already fully allocated to two projects near Edmonton. British Columbia put 400 MW up for competitive bidding earlier this year in January, with specific caps of 100 MW for conventional facilities and 300 MW for AI facilities over the first two years.

At the same time, the buildout is being delayed by requests from city councillors and citizens to pause projects. Oakville has passed a one-year moratorium, Mississauga council votes on one in September, and Toronto residents are pushing for a pause on projects in Etobicoke and Scarborough.

The neocloud and its landlord

Despite delays and setbacks potentially putting data centre projects on hold, AI companies like Nvidia are proactively changing course. In fact, Nvidia is now several things at once: an AI chipmaker, most valuable company in 2026, and landlord of the “neocloud” landscape. Neoclouds are specialized AI-first cloud computing providers for AI applications and systems. No sprawling catalogue of managed services, no virtualization layer, just bare metal and access to the hardware AI needs, rented by the hour. While traditional cloud computing providers have historically reached 20 to 30 per cent operating margins, neoclouds will work for about 15 per cent.

Another difference is that neocloud firms don’t have a traditional customer base, which makes it challenging for investors to gauge their value. To mitigate this challenge, Nvidia sought lenders willing to finance neocloud firms, announcing partnerships to mobilize more than $500 billion of third-party capital. One of these partners is Canada’s Brookfield, and CEO Bruce Flatt is quoted in the release calling compute a core pillar of the firm’s AI strategy.

Which brings us back to open-source AI models. In July, Huang used his first ever post on X to publish an industry letter defending open weights, eventually signed by more than 150 organizations saying “The world needs both frontier closed models and frontier open models.” OpenAI, Anthropic and xAI were not among those signing.

In short, the logic is simple – models are the complement to chips, and Nvidia is the largest global AI hardware provider. If open models also make that compute interchangeable between both suppliers and consumers of AI models, the push for open models and the $500 billion infrastructure fund are the same bet. Nvidia takes a share of the cloud revenue those chips generate while holding equity in its own customers. Critics call the arrangement circular, but we will have to wait and see if being leveraged on both sides of the economic pie can be true at once.

On the Horizon: Neoclouds are coming to a place near you

Ottawa is trying to pull Canadian pension money into building domestic computing muscle. Back in 2024, the Liberal administration confirmed a $15 billion program of loans and equity meant to unlock up to $45 billion in total investment, with pension funds putting in $2 for every federal dollar.

But Canadian pensions are not new entrants. The Maple 8 hold nearly US$6.6 billion in US-listed data centre and digital infrastructure firms, up from US$2.3 billion in 2022. More recently, CPPIB and the federal Strategic Response Fund are supporting up to an estimated $465 million for a data centre buildout in Cambridge, Ontario, which CoreWeave now operates and Canadian Cohere uses.

Ontario’s final playbook lands this fall with the province’s AI strategy. If the draft holds, Ontario will be picking data centres with the electricity bill in mind, and the people who pay for it are watching.

Interested in what’s next for Canada’s AI future? Listen to this edition of Signal49’s Future Skills Podcast where Jeremy Strachan chats with experts to talk about Building Canada’s AI Advantage.