Why We Own the Backbones of AI
In 2025, the world’s AI computers drew about three New York Cities worth of power. By 2027, U.S. data centres alone will draw more than six. By 2030, global demand is on track for nearly thirty. We own what plugs it all in.
Is it too late to own AI?
An app can be built in a weekend. A power plant cannot. That one idea explains where we invest in AI: not in guessing which application wins, but in the physical systems every application needs to run. AI feels weightless, yet behind every prompt sits a heavy physical system: data centres the size of arenas, transmission lines, cooling, fuel, and metals. The case was never about catching a stock. It is a decade-long buildout of power, compute, and materials that is still gathering speed.
Infrastructure cannot be built on demand
For more than a decade, markets rewarded businesses that could scale without building anything. Software compounded. Physical capacity did not. Energy systems, grid capacity, and strategic materials saw years of underinvestment, and AI has now arrived on top of that gap. When demand rises faster than supply can respond, scarcity develops, and scarcity increases the value of existing capacity. This is not a narrative. It is a supply problem, and supply problems take years to fix.
Where we stand
The world’s largest technology companies plan to spend roughly US$725 billion on AI infrastructure in 2026, up 77% from last year. OpenAI says the AI sector alone needs 50 GW of new capacity by 2028, twice New York City’s peak demand. U.S. data centre capacity is set to roughly double in three years, from about 80 GW in 2025 to about 150 GW in 2028.
Put it in scale terms. In 2023, data centres used a little over 4% of all U.S. electricity. By 2028, that share is on track to double, and in some estimates nearly triple. The global picture points the same direction: the power these facilities consume worldwide is set to more than double by the end of the decade. This is not a marginal shift in demand. It is a new heavy industry being built on top of the existing grid, and it is arriving faster than utilities can comfortably add supply.
Supply, not demand. New nuclear and the first small modular reactors do not arrive until 2028 or later, which is why Meta has already locked up more than 6 GW of nuclear power in advance. Power availability is now the limiting factor on AI itself. When demand runs years ahead of supply, the pricing power sits with whoever provides the supply.
Energy generation and transmission, because every gigawatt of compute needs a gigawatt of power and a grid to move it. Copper, because it is the conductive backbone of the entire buildout, and each new data centre uses five to ten times more copper than a typical commercial building. Lithium, because grid-scale storage is what makes 24/7 AI loads workable on a grid leaning on renewables. Cooling, because it consumes nearly 40% of a data centre’s power and gets more critical as racks get denser. Own the inputs the boom cannot function without.
Owning the shovels, not the gold
In the gold rush, the durable fortunes were rarely made by the prospectors. They were made by the people selling picks, shovels, and blue jeans. We apply the same logic to AI: rather than guessing which platform wins, we own the shared dependencies every winner will need. These are capital-intensive businesses with high barriers to entry, and bottlenecks do not stop adoption. They extend the cycle and shift value toward the systems that relieve them.
How we hold it
We hold this exposure as a satellite around a diversified core, spread across four pillars: compute, power and utilities, energy infrastructure, and critical materials. Infrastructure themes are not linear, and this one will have drawdowns. That is exactly why sizing and discipline matter more than conviction. The theme is the easy part. The real work is in balancing the pillars and knowing when to trim what has run.
This material is provided for discussion purposes only and does not constitute investment advice or a recommendation to buy or sell any security. Views reflect the perspective of MLD Wealth as of August 2026 and are subject to change without notice. Figures referenced are drawn from public, third-party, and company sources as of the dates noted, include projections that may not materialize, and should be verified at the time of reading. Past performance is not indicative of future results. Any positioning described is general in nature and should be assessed within the context of each investor’s full financial situation, objectives, and risk tolerance. Wealth management services are provided through a CIRO member firm. Member of the Canadian Investor Protection Fund (CIPF).