AI Has a Structural Problem

Download Full Report

This article was authored by: James Foord

The AI bull market has a structural problem, and it’s beginning to show. Microsoft, Amazon and Alphabet are guiding toward a combined $565 billion in capital expenditure this year. That number is up more than 60% from already-historic 2025 levels. But their cash flow is no longer covering it, and the market is punishing them for it. Microsoft, Amazon, and Alphabet have been high-margin cash flow machines over the last decade, but not anymore. Cash flow no longer covers their spending, and it’s even going negative in some cases. Estimates from Morgan Stanley suggest that hyperscalers will borrow around $400 billion in 2026 alone, pushing total issuance of high-grade US corporate bonds to a record $2.25 trillion for the year. A year ago, that might have been celebrated, but the price action over the last month speaks for itself. AMZN and MSFT are down 12% and GOOGL 8%. The main beneficiaries right now are the picks-and-shovels plays, building out AI. This is real spending and real earnings going from the hyperscalers to the likes of Micron (MU) or Advanced Micro Devices (AMD). But will the benefits revert back to the hyperscalers eventually? That’s something the market is not so sure about. We’re now moving into the stage where CAPEX is being funded by debt, and this can be sustained for some time. When risk appetite shifts, whether from a recession signal, a geopolitical shock, or simply a market that shifts its narrative, then credit spreads explode higher. This is where AI becomes a macro story. Roughly half of the Fed’s members are now arguing for at least one rate HIKE before year-end. And the stocks that make up most of the S&P 500 are now carrying double or even triple the debt they once held. Analysts expect earnings to compound at 24% over the next five years, and they have never been this optimistic before in history. Much of this optimism is grounded in the AI trade, but history shows that analysts are quite poor at predicting earnings over the medium and long-term. This isn’t a call to sell everything and hide in cash. The AI buildout is real, the technology is transformative, and the hyperscalers aren’t going bankrupt. But risk/reward has definitely shifted. The upside from here requires AI monetization to accelerate, credit conditions to stay loose, and the Fed to cooperate. Credit spreads are the primary signal to watch. As long as investment-grade spreads stay tight, the debt machine keeps running, and the cycle extends. For now, though, it makes sense to trim hyperscalers, while staying invested in the direct AI bottleneck and picks-and-shovels stocks. Investors have to stay disciplined and manage risk accordingly.