The Unreality of AI Compute Demand

We have spoken at length on the subject of AI demand in Canada. Despite our best efforts, it seems that Telus insists on ignoring us, and so we have yet to get any further answers to our most pressing question: who, exactly, is the company building its enormous data centres for?

Our question stems from a very reasonable premise: it seems as if the demand for AI compute in Canada is nebulous, non-existent, or entirely manufactured. In any case, the urgency with which Telus and the federal government seem to be moving ahead with building out large power/water-hungry computerized behemoths is troubling at best, and enraging at worst. We are trying, dear readers; oh how we are trying to get any sense of what exactly the plan is for these enormous projects, other than “build it and hope for the best”, but with so little to go on, we must continue our guessing game.

Today’s subject for examination is once again the Rimouski data centre, which Telus is so fond of wheeling out as a huge success for its AI compute business. (Once again, we remind you that selling out an AI data centre with only 500 last-gen GPUs at most is not impressive.) We have been trawling the public internet in the hopes of gaining some insight into who exactly is paying Telus’s overpriced GPU/hr rates for Rimouski, and we have not been particularly successful. But another clue has presented itself, and possibly provides some insight into why Telus is so enthusiastic about Rimouki in its marketing materials.

L-Spark is a Canadian startup incubator in the fashion of Ycombinator, which does interesting things to help new and emerging businesses in their search for market share, profits, and riches beyond the dreams of mortal man. They have been around since 2014 and have, of course, incubated quite the number of businesses in their time (none of which we have ever heard of, but we won’t hold that against them). Now, an interesting point about L-Spark is that one of its main programs is the Telus Sovereign AI Accelerator, which is as follows:

TELUS and L-SPARK announced a first-of-its-kind program designed to enable high-potential Canadian startups and scaleups to build, train and deploy advanced AI solutions on Canada’s fastest and most powerful sovereign AI supercomputer. The TELUS Sovereign AI Accelerator will usher in a new wave of Canadian innovation by accelerating the go-to-market strategies and investment readiness of select businesses.

Participating companies will receive compute credits from the TELUS AI Factory – powered by 99% renewable energy and NVIDIA platform – alongside one-on-one guidance from seasoned L-SPARK executive advisors. The engagement is designed to fast-track product development, unlock new customer relationships and build the investor networks critical to long-term success, all while maintaining complete control over data and intellectual property.

Well, now look at that. Participants in the program receive compute credits for Telus’s glorious AI factory! How very interesting. Free compute resources, you say? We must note that Telus announced this program, with the first participating companies, on April 28th 2026.

OTTAWA, ON, April 28, 2026 – TELUS and L-SPARK today announced a first-of-its-kind program designed to enable high-potential Canadian startups and scaleups to build, train and deploy advanced AI solutions on Canada’s fastest and most powerful sovereign AI supercomputer. The TELUS Sovereign AI Accelerator will usher in a new wave of Canadian innovation by accelerating the go-to-market strategies and investment readiness of select businesses.

How very kind of Telus to support new startups. Very generous, of course. But unfortunately this raises another, somewhat more dubious issue when taken in context with other data. On May 11th, barely two weeks later, Telus posted another press release containing a statement from Darren Entwhistle, the then-CEO, in which he states the following:

"We are incredibly proud to be working with the Government of Canada to help build Canada’s sovereign AI infrastructure,” said Darren Entwistle, President and CEO of TELUS. “The unprecedented demand that completely sold out our first AI Factory in Rimouski proves that Canadian innovators want cutting-edge AI built right here on Canadian soil.”

Stop right there, Mr. Entwhistle: what exactly do you mean, sold out?

How can it be “sold out” when Telus is giving the compute away for free to a bunch of AI startups?

How much of the compute is being given away, and how much of it is actually being paid for at the premium price Telus is asking for it?

If AI is the future of everything, why can’t the startups pay for their compute bills on their own merits?

This is a freakishly bad situation, dear readers. Getting up and running with a cloud provider like AWS, for example, is highly scalable for startups and should be well within the range of even the most basic seed funding. Buying AI compute on demand is trivial today, and it’s entirely possible to buy the exact same compute from other providers without having to pay the Telus premium, as we’ve said before. The only reason to use Rimouski (and therefore be locked into Telus’s ecosystem) is if data location is all-important due to client needs or provincial regulation.

The startups in question who are availing of the free compute are listed on the press release, and they are as follows (quoted from said release):

• Airy3D Airy3D's DepthIQ™ IP delivers simultaneous 2D images and 3D depth maps from a single passive image sensor – providing a compact, power-efficient, and cost-effective solution for use in robotics, automotive, industrial automation and consumer devices.
• Codalio is an AI-driven product and application development platform that empowers startups and companies to launch MVPs and build scalable, enterprise-grade applications faster and more affordably.
• Edge Signal helps retail and telco retail use physical AI to increase revenue and profitability, improve customer experience and optimize daily operations across every location.
• PataBid offers AI enterprise-grade construction bidding software designed for complex specialty trades, delivering standardization and risk reduction for commercial, industrial, and institutional projects across teams and regions.
• TopoLift transforms generic AI into a bespoke intelligence layer that learns the structure of the customer's business and grows smarter with their data - delivering clearer reasoning, fewer errors and highly accurate, context-aware decisions.

Notice that all of these companies are business-to-business operations, and none of them are specifically related to healthcare or government operations, which are the prime targets for data retention in-country. None of these products are in any way specific to Canada, nor do they suggest that the Canadian market for such would be significant in comparison to the US or European Union; there is no reason whatsoever to think that keeping the data in Canada is relevant to their operations. So why, we ask, would they use Rimouski, instead of going with a cheaper international compute provider?

We believe the answer is this: they were offered the compute for free.

There is no further reason. In our most enlightened opinion, Telus could not sell the compute at the premium that would make it worthwhile, hence the company went to L-Spark and created the program to offer it for free to startups as part of their incubator program. They then crowed about their AI factory being sold out, and pretended that the demand had existed all along.

We have long believed that the demand for AI compute is entirely manufactured across the world, and this is an example of it. Telus are, in our view, being misleading every time they state that their AI compute has sold out in Rimouski, because that compute is not actually being bought at all.

If they would care to correct the record on this point, then perhaps they should finally start replying to our emails.