AI data centres are a dead end
by Zac Trolley
The hardcore science fiction junkies recognize today, Aug 29th, as Judgment Day. This is the day in the Terminator franchise where the Skynet AI becomes self-aware and launches a nuclear attack to remove humans from the Earth.
In 1984 when the movie came out, this was science fiction, but today we see AI infrastructure being built at an alarming pace.
Building AI for AI sake
The Carney Government is going all in with AI, creating an AI minister and announcing billions of dollars of investment. On the surface this looks like a forward thinking move, since AI is everywhere. The problem is that AI is everywhere because of hype, not substance, and I have a growing fear that our planned investment in AI is digging a hole. The billion dollar price tag for new data centres is not just a reckless use of taxpayer money; it’s a fundamental misunderstanding of what problem they are trying to solve.
I am bullish on AI.
I don’t think it’s a world-ending technology like Terminator would have us believe, but it’s not a solution to all our problems either. According to a paper published by MIT, 67% of AI success comes from solving specific problems, not from leveraging massive infrastructure projects. They found that a staggering 95% of AI deployments in large organizations fail. Not because the models are bad, but because they are so general they can’t adapt to specific issues. The companies that are succeeding in AI are choosing a single business pain point, and building tools to resolve it.
The cutting edge is no longer large models like Chat GPT-5, rather small precision solutions. A new paper released by NVIDIA in June is casting serious doubt on the “bigger is better” business case that has led to massive investment into AI in the first place. The researchers outline how Small Language Models (SML) outperform Large Language Models (LLM) when they have a specific task and work together. This means that a small business can deploy 10 or 20 resource light SMLs specific to their business needs on consumer hardware.
No big data centres needed.
Projects like O’Leary’s Wonder Valley here in Alberta are planning to build city block-sized data centres with a “build it and they will come” mentality. The absurdity of this situation is staggering. A gaming PC can run the AI required to solve small business tasks, yet we’re being told that we require billions of dollars worth of servers to achieve the same goal. It’s a solution in search of a problem.
The next generation of entrepreneurs isn’t waiting for Canadian data centres to be built; they’re already building solutions in their basements, using open source and off-the-shelf tools to launch successful startups. The MIT paper points out that these startups can achieve $20M in revenue without the need for data centres, yet our government remains committed to a plan that will divert funds to Big Tech vanity projects.
Coming Bubble
When pressed about the lacklustre launch of Chat GPT-5, Sam Altman himself says he is starting to worry about the over-hype on AI investment. A reporter asked if “investors as a whole are overexcited about AI” he answered “yes”. He then went on to warn reporters about a possible AI bubble, saying “If you look at most of the bubbles in history, like the tech bubble, there was a real thing. Tech was really important. The internet was a really big deal. People got overexcited.” If he thinks AI is overvalued, we should take a long hard look at risks of investing into AI mega projects.
I think Alberta should pursue AI as a strategy to bolster the economy. But they should do so strategically and focus on pain points where it can actually help businesses act more nibble. We need to automate the business basics so we can unleash creativity to solve our big problems.
And data centres are a distraction, not the solution.