The AI Trade Was Priced for Endless Demand. Memory Chips Disagree.
Memory chips are the least glamorous part of an AI data centre, and they are usually the first part to tell the truth. Every AI server needs huge amounts of fast memory. When the companies building data centres slow their orders, memory makers feel it before anyone else. This fall they are feeling it. Memory and chip stocks have slid hard enough that investors around the world are asking whether the whole AI trade got ahead of itself.
The common view among careful investors is that AI is a decade-long shift, so dips are noise and the smart move is to hold. Much of that holds up. Businesses really are adopting AI, and plenty of selloffs turn out to be healthy corrections after too much excitement. The weak spot is how the stocks were priced. Prices assumed the flood of hardware orders would keep growing, as if every server bought proved AI was already paying for itself. If you hold index funds in an RRSP or TFSA, or have a down payment invested, that pricing assumption affects you more than the question of whether AI succeeds.
A chip sale is someone else's expense
Follow the money. Chip makers sell mostly to a handful of very large tech companies that have committed hundreds of billions of US dollars to data centres. On those companies' books, that money is a cost. It becomes lasting demand only when customers pay for AI products in amounts that cover it. Investors are now asking to see that return, and so far the answer is mostly "later."
Memory also has a long habit of booms and gluts. It trades a lot like a commodity: buyers rush to lock in supply, makers expand, then inventory piles up and prices fall. Rising chip inventories are the warning sign today. When buyers stock up out of fear of a shortage, their orders look like endless demand on the way up and disappear together on the way down.
The lesson works for any stock: when one company's sales are another company's spending, ask what the buyer earns from it. If nobody can answer yet, you own a bet on how patient those buyers will be.
"Diversified" can hide one big bet
Many Canadians own an S&P 500 fund and think of it as 500 companies. In practice, a large share of that index sits in a few AI-linked giants and their suppliers, and they tend to rise and fall together. The TSX leans more on banks, energy and materials, and if money rotates out of tech into steadier sectors, Alberta staples like energy and utilities may hold up better. US funds and ETFs in your accounts still carry the full tech exposure. Open your top ten holdings. If most of them depend on the same AI spending, treat them as one position.
Money with a closing date shouldn't ride a cycle
For Edmonton buyers, this is where it gets practical. A move-up buyer planning to sell and buy within a year or two may have part of the down payment in a tech-heavy fund. As an illustration: a $100,000 down payment that drops 20% becomes $80,000, so you'd need to borrow $20,000 more for the same house. You'd still have to qualify at the stress test rate, the higher of your contract rate plus 2 percentage points or 5.25%. A bigger loan tested at that higher rate can shrink your budget or push back your timing.
Long-term retirement money can wait out a correction, and history gives it time to. Money you need on a closing date has no such luxury. Ask yourself one question: if that fund fell by a fifth next month, could you still buy the home you're planning for? If the honest answer is no, that money is sitting in the wrong place.
Memory chips are the least glamorous part of an AI data centre, and they are usually the first part to tell the truth. Every AI server needs huge amounts of fast memory. When the companies building data centres slow their orders, memory makers feel it before anyone else. This fall they are feeling it. Memory and chip stocks have slid hard enough that investors around the world are asking whether the whole AI trade got ahead of itself.
The common view among careful investors is that AI is a decade-long shift, so dips are noise and the smart move is to hold. Much of that holds up. Businesses really are adopting AI, and plenty of selloffs turn out to be healthy corrections after too much excitement. The weak spot is how the stocks were priced. Prices assumed the flood of hardware orders would keep growing, as if every server bought proved AI was already paying for itself. If you hold index funds in an RRSP or TFSA, or have a down payment invested, that pricing assumption affects you more than the question of whether AI succeeds.
A chip sale is someone else's expense
Follow the money. Chip makers sell mostly to a handful of very large tech companies that have committed hundreds of billions of US dollars to data centres. On those companies' books, that money is a cost. It becomes lasting demand only when customers pay for AI products in amounts that cover it. Investors are now asking to see that return, and so far the answer is mostly "later."
Memory also has a long habit of booms and gluts. It trades a lot like a commodity: buyers rush to lock in supply, makers expand, then inventory piles up and prices fall. Rising chip inventories are the warning sign today. When buyers stock up out of fear of a shortage, their orders look like endless demand on the way up and disappear together on the way down.
The lesson works for any stock: when one company's sales are another company's spending, ask what the buyer earns from it. If nobody can answer yet, you own a bet on how patient those buyers will be.
"Diversified" can hide one big bet
Many Canadians own an S&P 500 fund and think of it as 500 companies. In practice, a large share of that index sits in a few AI-linked giants and their suppliers, and they tend to rise and fall together. The TSX leans more on banks, energy and materials, and if money rotates out of tech into steadier sectors, Alberta staples like energy and utilities may hold up better. US funds and ETFs in your accounts still carry the full tech exposure. Open your top ten holdings. If most of them depend on the same AI spending, treat them as one position.
Money with a closing date shouldn't ride a cycle
For Edmonton buyers, this is where it gets practical. A move-up buyer planning to sell and buy within a year or two may have part of the down payment in a tech-heavy fund. As an illustration: a $100,000 down payment that drops 20% becomes $80,000, so you'd need to borrow $20,000 more for the same house. You'd still have to qualify at the stress test rate, the higher of your contract rate plus 2 percentage points or 5.25%. A bigger loan tested at that higher rate can shrink your budget or push back your timing.
Long-term retirement money can wait out a correction, and history gives it time to. Money you need on a closing date has no such luxury. Ask yourself one question: if that fund fell by a fifth next month, could you still buy the home you're planning for? If the honest answer is no, that money is sitting in the wrong place.
Read Next
AI Stocks Are Priced for Productivity Gains That Haven't Arrived Yet
What Fairfax Financial's India Deal Must Deliver to Justify Nearly 32% Upside
Why the TSX's Leading Sectors Look So Different From the U.S. Market's
Who Really Gains When Canada's Banks and Telecoms Join Forces Against Fraud?