AI Giants Are Borrowing $350 Billion. The Fed May Make It Hurt.
Here’s the “huh” moment of the week: AI companies are collectively piling on $350 billion in debt to fund a $725 billion spending surge on data centers—all while the bond market is flashing a warning that the very rate cuts that made this borrowing cheap might be reversed. One CEO calls it a buildout “since the Great Wall of China.” Another analyst says “the demand is booked.” But the Federal Reserve is now hinting it may undo the 2025 “insurance cuts” that stabilized the economy, according to RBC Wealth Management.
The investing question isn’t whether AI is real. It’s whether the *financing* of AI is sustainable when the cost of capital rises.
Let’s be clear: the scale is staggering. Google, Amazon, and Meta are driving a 4% increase in North American data-center project pipelines in just one month. Meta alone just committed C$13 billion ($9.17 billion) for its first Canadian AI data center in Alberta. Cerebras CEO Andrew Feldman captured the mood on the All-In Podcast: “We’ve never seen a buildout like this since the Great Wall of China,” pointing to a $25 billion backlog where “the demand is booked.”
But here’s the tension that should make any investor pause. The 10-year Treasury yield sits at 4.56% as of July 10—up from 4.48% just four days prior. The Fed’s own calendar shows a critical inflation reading on July 14, with core CPI expected to remain sticky at 2.9%. Multiple Fed speakers are scheduled for the same week. The market is beginning to price in the possibility that the rate cuts that juiced risk appetite in 2025 were a temporary sugar high.
This isn’t a macro lecture. It’s a lens to separate the *cash-flow funders* from the *debt-dependent builders*.
Consider the distinction: A company like Meta, which saw its stock jump 12.5% in the past week and 17.2% in the past month, is funding its Alberta data center from a balance sheet that generates enormous free cash flow from its core advertising business. Its borrowing is strategic leverage, not survival financing. On the other hand, the broader $350 billion debt pile across AI giants includes companies where the return on that capital is still hypothetical—where data centers are built on the *promise* of future AI workloads, not pre-sold contracts.
The analyst who said “when we build a data center, it’s already pre-sold” was referring to Amazon Web Services—a business with proven revenue streams. But not every AI spender has that luxury. The market is already voting: Google (GOOG) is down 0.88% this week, while AMD—a supplier of the picks-and-shovels—is up 4.21% this week and 23.32% over the past month. The market is rewarding the *enablers* of the buildout more than some of the *builders* themselves.
This is where the curiosity sharpens into an investing frame. If the Fed does reverse course—if the 4.56% 10-year becomes 5% or higher—the math changes for every project that depends on cheap debt to achieve its internal rate of return. The “demand is booked” narrative works when the cost of financing that demand is low. When it rises, the question becomes: who has the pricing power to pass on higher costs, and who is stuck with a half-empty data center and a floating-rate note?
The answer isn’t binary. AI is real. The buildout is happening. But expensive money changes who captures value. The cash-flow-rich hyperscalers will likely survive and even thrive—they can self-fund or borrow at favorable rates. The debt-dependent players—the smaller cloud providers, the speculative AI startups leasing capacity on margin—will face a squeeze. And the suppliers (chipmakers, construction firms, energy providers) may see a short-term boom followed by a shakeout if projects get delayed or canceled.
So the reframed question for this week isn’t “Will AI change the world?” It’s “Can the AI capex cycle withstand a 5% risk-free rate?” Because if the Fed’s next move is a reversal, the winners won’t be the ones who borrowed the most—they’ll be the ones who didn’t *need* to.