• Home
  • AI Infrastructure Spending and Real Yields Above 2% Force a New 60/40
AI Infrastructure Spending and Real Yields Above 2% Force a New 60/40
By Patrick Henneberry profile image Patrick Henneberry
3 min read

AI Infrastructure Spending and Real Yields Above 2% Force a New 60/40

The Magnificent Seven collectively plan to spend more than $280 billion this year on data centers and GPU servers. That figure, equivalent to the entire GDP of Finland, represents the single largest coordinated capital expenditure cycle in modern corporate history, and it is happening while ten-year real yields hover near 2%, a level not seen since before the 2008 financial crisis.

These two forces do not reinforce each other. They collide. AI-focused equities demand extraordinary valuations justified by extraordinary growth. Bonds yielding 2% above inflation provide genuine income for the first time in fifteen years. The friction between them is rewriting how institutions allocate capital.

The end of "there is no alternative"

From 2010 through 2021, real yields on government bonds were negative or near zero. A portfolio that held 40% in fixed income was accepting a guaranteed loss after inflation in exchange for stability. The equity allocation carried the entire return burden. Safe alternatives did not exist, so investors accepted higher concentration risk in growth stocks. From 2010 through 2021, this constraint was called TINA: there is no alternative to equities.

That era ended when central banks raised rates to combat inflation. By mid-2024, Canadian high-quality corporate bonds were yielding 4.5% to 5.5%, and ten-year US Treasuries were delivering real returns between 1.5% and 2.2%. Fixed income stopped being a drag. It became a viable return source, which changed the calculus for the equity side.

Capex as both signal and risk

Microsoft, Alphabet, and Meta have each committed over $40 billion annually to GPU servers and data center hardware. These are defensive moves in a winner-take-all race where the firms that own the compute layer will control the monetization layer. Capex at this scale creates a structural moat, only the largest balance sheets can compete.

But capital intensity of this magnitude also raises the hurdle rate for equity returns. When a company spends $40 billion a year on GPU servers and data centers that will not generate revenue for three to five years, the stock price embeds an assumption about future cash flows that must clear a higher bar when bonds offer 2% real. The gap between what AI stocks must deliver and what bonds guarantee has widened, and that gap is volatility.

The mechanics favor firms that can self-fund. Companies with fortress balance sheets, low debt, high cash flow, can build without issuing bonds at elevated rates. Smaller competitors cannot, which accelerates consolidation. The AI capex cycle is narrowing the field to the handful of firms with enough cash flow to fund multiyear buildouts.

The barbell takes shape

Institutional portfolios are responding with what amounts to a barbell strategy. One end holds concentrated positions in the handful of firms that can afford to compete in GPU and data center buildouts. The other end holds high-quality government and corporate bonds that finally provide income and diversification. The middle, broadly diversified equities with moderate growth, has less justification when the edges offer both extremes.

The equity 60 is hyper-focused on capex-intensive tech, not broad market exposure. The fixed income 40 is doing the work it was designed to do a generation ago: generating real returns and damping volatility. The structure looks familiar, but the risk profile has changed.

Canadian institutional allocators, including British Columbia Investment Management Corporation and the Canada Pension Plan Investment Board, have increased exposure to US firms spending heavily on GPUs and data centers while raising their fixed income quality standards. The TSX, historically weighted toward financials and energy, does not provide the AI exposure these portfolios require, which has driven capital toward US-listed hyperscalers.

The power constraint nobody prices

Data centers require stable, high-capacity power grids. Ontario's Independent Electricity System Operator projects 2.2% compound annual growth in peak demand through 2050, driven primarily by industrial computing loads. British Columbia's grid, while greener, faces similar capacity questions as GPU deployment scales. The physical limits of energy availability represent a constraint that most equity valuations do not reflect. Capital can be deployed faster than substations can be built.

Rising real yields and AI capex are not temporary dislocations. They define the investment environment. Portfolios built for the prior decade will not survive the next.