AI's Power Problem: The Buildout Is Hitting Limits
For all the talk of AI as pure software, it runs on a very physical foundation, power-hungry data centres wired into the electrical grid. And that foundation is starting to strain. Reports now say the wait for critical grid equipment like large transformers has ballooned from months to several years, putting much of the planned new AI capacity at risk of delay, while local communities increasingly resist new data centres in their backyards. The demand for AI compute is outrunning the world's ability to power and site it. That does not spell doom, but it is a useful reality check for how you plan.
Software with a very physical bottleneck
It is easy to imagine AI capacity as something that simply scales up in the cloud on demand. In reality, every new burst of capacity needs buildings, chips, and, above all, electricity delivered through grid infrastructure that takes years to expand. When the equipment that connects a data centre to the grid has a multi-year lead time, you cannot just conjure more capacity by writing a bigger cheque. Add growing public opposition to data centres nearby, and you get a genuine constraint: the appetite for AI is sprinting while the physical world it depends on can only walk.
Two things are true at once
This does not cancel the "AI is getting cheaper" story, it complicates it. Hold both ideas together.
| The long-term trend | The near-term friction |
|---|---|
| Better chips, more competition | Power and equipment shortages |
| Cheaper, more abundant AI over time | Firmer prices, tighter capacity now |
| Capabilities keep trickling down | Newest, compute-heavy features may lag |
We made the optimistic case in the AI chip boom; this is the sober counterweight. The destination is cheaper AI, but the road there has traffic.
The Canadian angle
This is not just an overseas story. Canada is attracting major AI data-centre investment, including large new facilities, thanks to its climate, power, and stability. That brings real opportunity: jobs, investment, a growing domestic AI-infrastructure sector. It also imports the same tensions seen elsewhere, pressure on local power and infrastructure and community debate over whether the trade-offs are worth it. For Canadian businesses, both sides are worth watching: the upside of a homegrown AI-infrastructure boom, and the reality that power and siting limits are a live issue here too.
The takeaway
None of this should stop you from using AI, it should just make you a little wiser about how. Treat AI capacity as valuable rather than infinite: use right-sized models for each task, avoid waste, and don't architect critical processes on the assumption that unlimited, ever-cheaper compute is guaranteed on your schedule. Keep some flexibility across providers so a capacity crunch at one does not strand you. And take the long view, plan for AI to keep getting cheaper and better, while budgeting for near-term bumps. AI stands on physical infrastructure, and infrastructure has limits. Build habits that stay resilient either way.
Frequently Asked Questions
What is AI’s "power problem"?
AI runs in data centres that consume enormous amounts of electricity, and the buildout is now colliding with physical reality. Reports say the wait for critical power equipment (like the large transformers that connect data centres to the grid) has stretched from months to several years, putting a big share of planned new capacity at risk of delay. On top of that, local communities are increasingly pushing back on new data centres nearby, with surveys showing most people oppose them. In short: the demand for AI compute is racing ahead of the world’s ability to power and site it.
Does this contradict the idea that AI is getting cheaper?
Not exactly, it adds nuance. Over the long run, better chips and more competition do push the cost of AI down, and that trend is real. But in the near term, physical bottlenecks (power, grid connections, equipment, and siting) can slow new capacity and keep prices firmer than the "everything gets cheaper" story suggests. Both things are true: the long-term direction is cheaper and more abundant AI, while the short-term path has real friction. Smart planning accounts for both, expecting cheaper AI eventually without assuming unlimited, ever-falling-price capacity right now.
How could this affect my business directly?
Mostly in subtle ways. If AI capacity is constrained, you might see it show up as firmer pricing, occasional rate limits or slower access to the newest, most compute-hungry features, or providers prioritizing their largest customers when things are tight. It is unlikely to stop you from using AI, but it is a reason not to assume infinite, dirt-cheap capacity is guaranteed on your timeline. For most small businesses the effect is indirect, but it reinforces a sensible habit: use AI efficiently and do not architect your operations around the assumption that unlimited compute will always be cheap and instantly available.
Is there a Canadian angle to this?
Yes. Canada is becoming a location for major AI data-centre investment, drawn by cooler climates, power availability, and stable conditions, including large new facilities being built here. That can mean local economic opportunity, but it also brings the same tensions seen elsewhere: pressure on local power and infrastructure, and community debate about whether the benefits are worth it. For Canadian businesses, it is worth watching both sides: the opportunity of a growing domestic AI-infrastructure sector, and the reality that power and siting constraints are a live issue here too.
What should a Canadian business actually do about it?
Do not panic or overreact, but plan realistically. Treat AI capacity as valuable rather than infinite: use it efficiently (right-sized models for each task, no waste), and avoid designing critical processes that assume unlimited, always-cheap compute. Keep some flexibility in your AI providers so you are not stranded if one gets capacity-constrained. And take the long view: expect AI to get cheaper and more capable over time, while budgeting for near-term friction. Mostly, this is a reminder that AI rests on physical infrastructure, and infrastructure has limits, so build habits that stay resilient either way.
Build an AI plan that's resilient either way
We help Canadian businesses use AI efficiently and stay flexible across providers, so your operations hold up whether compute is cheap and plentiful or tight and pricey.
Related Articles
Why Your Next Laptop Costs More: The AI Buildout Is Hitting Hardware Prices
When AI Usage Limits Don’t Match the Marketing: How to Vet AI Vendor Claims
When AI Runs the Company: The Accountability Test
AI consultants with 100+ custom GPT builds and automation projects for 50+ Canadian businesses across 20+ industries. Based in Markham, Ontario. PIPEDA-compliant solutions.