By 2030, the world’s data centres could consume slightly more electricity than Japan does today. That is quite the utility bill for technology supposedly floating around in a “cloud.” Investors have already begun bidding up power producers and electrical-equipment companies, although the physical artificial intelligence (AI) buildout still has years of substations, transformers, and generating capacity left to install.
AI can’t run on promises
The International Energy Agency expects global data-centre electricity consumption to roughly double to 950 terawatt-hours (TWh) between 2025 and 2030. Power use from AI-focused facilities could triple. Accelerated servers create the demand, while cooling, backup systems, grids, and power conversion keep those servers from becoming extremely expensive space heaters.
Power stocks can reprice before all that demand reaches earnings because new generating plants and grid connections take years to approve and build. Existing capacity, available transformer production, and long-term contracts suddenly become more valuable. The winners should be businesses that can convert scarcity into orders without borrowing themselves into a rather impressive crater.
That changes how investors can approach Canadian AI stocks. Chip designers compete with one another and today’s leading model may not remain fashionable. A transformer, meanwhile, doesn’t care which chatbot wins. It earns from the electricity every customer needs, which points toward two Canadian picks-and-shovels businesses.
HPS
Hammond Power Solutions (TSX:HPS.A) manufactures dry-type transformers, power-quality products, and conversion systems. These products change voltage and distribute electricity safely through data centres, factories, renewable projects, and transportation infrastructure. The AI might answer in milliseconds, but somebody still has to prevent the building from tripping a breaker.
Second-quarter sales surged 44.7% to a record $324.8 million, driven partly by higher U.S. data-centre shipments. Backlog remained 96.9% above last year, with large data-centre orders providing visibility through 2026. Newly acquired AEG Power Solutions also expands Hammond Power stock beyond transformers into power conversion and reliability systems, giving it more equipment to sell into each electrification project.
The opportunity has not gone unnoticed. As of writing, Hammond Power stock trades near 52 times trailing earnings. That valuation leaves little room for delayed projects, cancelled backlog, rising copper costs, tariffs, or a messy AEG integration. I would build a position gradually rather than plug the entire Tax-Free Savings Account (TFSA) into one socket.
BEP
Brookfield Renewable Corporation (TSX:BEP.UN) provides equivalent economic exposure to Brookfield Renewable’s enormous portfolio of hydroelectric, wind, solar, storage, and nuclear-related assets. Data centres need power around the clock, not only when the sun and wind have checked their calendars.
Brookfield signed a framework with Google to deliver up to 3,000 megawatts of U.S. hydroelectric capacity. The first 20-year contracts cover two Pennsylvania facilities, turning AI demand into unusually long revenue visibility. Second-quarter funds from operations (FFO) per unit rose 11%, showing the broader platform is already growing while the Google framework develops.
Brookfield traded near 16 times trailing FFO. Higher interest rates can make financing and distributions less attractive, while construction delays, weak hydrology, asset sales, and debt add risk. Still, long contracts and diversified generation make Brookfield one of the more practical Canadian renewable-energy stocks for the AI power shortage.
Buy the bottleneck
Hammond Power stock offers faster growth and a far more demanding price, while Brookfield supplies contracted power with a slower, steadier profile. Neither is undiscovered, and both could fall if AI construction cools.
Yet electricity demand is moving from forecast slides into orders, contracts, and cash flow. Investors arriving before the grid catches up may still be early to the part of AI that cannot be downloaded.