AI Data Centers Face Higher Capital Costs

Investors should read that as a secular supply problem for duration. Data centers do not just need financing once; they need continuous reinvestment in servers, switches, cooling, power and land, which creates a persistent claim on capital even before the next wave of demand arrives. The SEC filings from CoreWeave underscore the point: depreciation and amortization have been rising sharply as infrastructure is placed in service, which means the business is building at speed while the carrying cost of that build is already showing up in the income statement.

That matters because the higher the capex-to-depreciation ratio stays above 2x, the more the business is effectively financing growth through outside capital rather than internal cash generation. In a world where Treasury yields are anchored close to 5%, that is not a trivial footnote. It is a pricing mechanism. The bond market is saying capital is scarce, and every incremental dollar poured into AI infrastructure has an opportunity cost versus government debt, corporate credit and equity buybacks.

The equity tape reflects that tension. Equinix has held well above its 200-day moving average, but recent momentum has cooled, with RSI readings slipping into the low-40s even as the stock remains a premium compounder. Digital Realty has also given back some of its earlier strength after a spring breakout, while CoreWeave has been far more volatile, a reminder that investors still reward the AI buildout but punish the financing risk when spending outruns monetization. These are not broken stocks. They are capital intensity stories trading in a market that is no longer willing to pay endlessly for growth without visible cash conversion.

The mispricing, in my view, is that many investors are still treating AI infrastructure as a pure demand story. It is not. It is also a funding story. Bell’s plan to quadruple its AI-focused data center in Saskatchewan fits the same pattern: the next phase of AI growth is becoming physically bigger, more energy-intensive and more capital-absorbing. That is good for the toll roads of the AI economy — REITs, power, networking, equipment and financing providers — but it raises the hurdle for returns everywhere else.

The winners are the picks-and-shovels businesses that get paid every time the capex cycle turns: utilities, grid equipment makers, cooling, fiber and select infrastructure landlords. The losers are the bondholders and equity holders of the most levered builders if rates stay high and depreciation keeps lagging the investment bill. If the 10-year remains near 5%, the market will keep forcing a simple question on every data-center dollar spent: does this build cash flow, or does it just absorb it?

For investors, that is the setup. Own the infrastructure that rents out scarcity, not the businesses that must keep buying it. The capex cycle is no longer a story in slides — it is in the cash flow statement, and the bond market is already voting on who pays for it.