Gray Sees Data Centres, Griffin Sees Prices: Where AI's Capital Risk Sits
Utility planners in Alberta are working out how much electricity the provincial grid can hand to data centres, the buildings that run AI models around the clock. Our cool climate and open electricity market make the province an attractive place to put them. That planning work is the most down-to-earth version of a debate that two of the biggest names in finance have been having in public.
Jon Gray, president of Blackstone, looks at AI and sees buildings, land and power that will earn rent for decades. Ken Griffin, who runs Citadel, sees a speculative cycle that may only be in its middle innings. Both are describing the same boom, and they can both be right. Lined up together, they show where the real capital risk sits: in how much future spending your purchase price quietly assumes. If you hold US index funds in an RRSP or TSFA, or you've looked at industrial land around Edmonton, you already have money on one side of this.
Who pays for the build matters more than how big it is
Gray's case starts with the buyers. In the dot-com era, much of the money went to young companies with little or no revenue. The companies buying AI chips and renting data centre space today are large, high-margin businesses with deep cash reserves. Blackstone has made data centres a core part of its strategy, treating them as long-term bricks and mortar.
When the customer pays out of existing profits, the spending can run longer than skeptics expect. A tenant with a strong balance sheet keeps paying rent through a rough year.
A price can assume the orders never slow
Griffin's worry sits one step further out. Each new generation of AI model costs more to build. If the jump from one version to the next stops delivering much improvement, companies may cut back on hardware, and demand for chips could taper off.
Here is clearly illustrative arithmetic. Say a chipmaker earns $10 a share and the stock trades at $300, or 30 times earnings. If earnings grow 20% a year for three years, they reach about $17.28. Even if investors then pay only 20 times earnings, the stock is worth about $346, a modest gain. Now suppose orders level off and earnings stay at $10. At 20 times, the stock is worth $200, a loss of a third. The business is still profitable in both versions. The price is what gets hurt.
Both bets carry friction
The building and the stock have different kinds of trouble. A data centre is slow to build, hard to sell quickly, and depends on power the grid may not be able to supply, which is exactly what Alberta's utilities are studying. A tech stock you can sell in seconds, but you pay for that convenience by taking whatever price the market sets on the worst day.
Most people never choose either one directly. A small group of US tech giants, often called the Magnificent Seven, has driven a large share of market returns into 2026. A broad US index fund in a TSFA puts a heavy weight on Griffin's side of the bet without asking. If prices do fall, Canada's tax rules let you carry a capital loss back three years against earlier gains in a non-registered account, though losses inside an RRSP or TSFA give you no such relief.
The question that decides it
The Bank of Canada sees AI as a possible productivity boost, with early adoption concentrated in finance and insurance, while noting it will take time before significant gains show up in overall productivity data. That uncertainty is where you make your call.
For any AI-linked holding, ask one thing: is the income I'm paying for being paid today, by customers with profits, or does my price only work if spending keeps climbing? If it's the first, you're standing with Gray, and time is on your side. If it's the second, you're carrying Griffin's risk, and you should size it like a bet you can afford to see shrink.
Utility planners in Alberta are working out how much electricity the provincial grid can hand to data centres, the buildings that run AI models around the clock. Our cool climate and open electricity market make the province an attractive place to put them. That planning work is the most down-to-earth version of a debate that two of the biggest names in finance have been having in public.
Jon Gray, president of Blackstone, looks at AI and sees buildings, land and power that will earn rent for decades. Ken Griffin, who runs Citadel, sees a speculative cycle that may only be in its middle innings. Both are describing the same boom, and they can both be right. Lined up together, they show where the real capital risk sits: in how much future spending your purchase price quietly assumes. If you hold US index funds in an RRSP or TSFA, or you've looked at industrial land around Edmonton, you already have money on one side of this.
Who pays for the build matters more than how big it is
Gray's case starts with the buyers. In the dot-com era, much of the money went to young companies with little or no revenue. The companies buying AI chips and renting data centre space today are large, high-margin businesses with deep cash reserves. Blackstone has made data centres a core part of its strategy, treating them as long-term bricks and mortar.
When the customer pays out of existing profits, the spending can run longer than skeptics expect. A tenant with a strong balance sheet keeps paying rent through a rough year.
A price can assume the orders never slow
Griffin's worry sits one step further out. Each new generation of AI model costs more to build. If the jump from one version to the next stops delivering much improvement, companies may cut back on hardware, and demand for chips could taper off.
Here is clearly illustrative arithmetic. Say a chipmaker earns $10 a share and the stock trades at $300, or 30 times earnings. If earnings grow 20% a year for three years, they reach about $17.28. Even if investors then pay only 20 times earnings, the stock is worth about $346, a modest gain. Now suppose orders level off and earnings stay at $10. At 20 times, the stock is worth $200, a loss of a third. The business is still profitable in both versions. The price is what gets hurt.
Both bets carry friction
The building and the stock have different kinds of trouble. A data centre is slow to build, hard to sell quickly, and depends on power the grid may not be able to supply, which is exactly what Alberta's utilities are studying. A tech stock you can sell in seconds, but you pay for that convenience by taking whatever price the market sets on the worst day.
Most people never choose either one directly. A small group of US tech giants, often called the Magnificent Seven, has driven a large share of market returns into 2026. A broad US index fund in a TSFA puts a heavy weight on Griffin's side of the bet without asking. If prices do fall, Canada's tax rules let you carry a capital loss back three years against earlier gains in a non-registered account, though losses inside an RRSP or TSFA give you no such relief.
The question that decides it
The Bank of Canada sees AI as a possible productivity boost, with early adoption concentrated in finance and insurance, while noting it will take time before significant gains show up in overall productivity data. That uncertainty is where you make your call.
For any AI-linked holding, ask one thing: is the income I'm paying for being paid today, by customers with profits, or does my price only work if spending keeps climbing? If it's the first, you're standing with Gray, and time is on your side. If it's the second, you're carrying Griffin's risk, and you should size it like a bet you can afford to see shrink.
Sources
Read Next
A 7-Step Bank Fee Audit to Find the Charges Your Statement Buries
What a General Inspection Misses: 7 Furnace Checks Before Your Fall Offer
Lenders Want Growth in 2027, but Funding Costs and AI Will Decide Who Gets It
Buying Tylenol on Credit: Kimberly-Clark Tests Its Investment Grade Cushion