Earnings are an opinion, cash flow is a fact

Asymco analyst Horace Dediu argues that hyperscaler AI spending is not heading for a bust, comparing it to Amazon's early infrastructure investment. He notes that while profits are thin due to reinvestment, revenue exists and P/E ratios are modest, with debt being the real risk rather than equity. Dediu concludes that AI demand remains high and a crisis is unlikely unless usage collapses or debt mispricing occurs.

An Office Hours question asked by Dave Emery, June 9, 2026. Q: Are we heading for an AI bust for the large hyperscalers? Or will the spending eventually produce the revenue the advocates promise ? And if it’s a bust, how does Apple adjust? There is already revenue — there’s just not much profit, and not much free cash flow, because everything’s being reinvested. This feels like the early days of Amazon, another infrastructure play, when CapEx was huge, margins were thin, and the P/E looked absurd because earnings were almost nothing. Eventually they recorded earnings, and AWS became the savior. Remember: earnings are an opinion, cash flow is a fact. You can declare software development as R&D and manage the earnings opinion. Cash you cannot. Is there a bust coming? Look at the hyperscalers — their P/E ratios are fairly modest, lower than Apple’s, and their top lines are growing fast. They’re so hungry for cash that some are doing private placements, diluting investors. But if they raise equity, I don’t worry. Equity is non-contagious. If you get hurt, it only hurts you. Subscribe to Asymco One for full access to our live Q&As. Subscribe to Asymco One for full access to our research. https://asymco.com/one/ subscribe My concern is debt. Almost every crash has come from debt, not equity, being mispriced. The dot-com crash was an overvaluation with no earnings and no revenue behind it, and it was short-lived. A debt-fueled crisis takes a long time to unwind, with repercussions across the economy. So far, AI hasn’t become a debt story. The bubble only bursts if people stop using it, run out of money to pay for it, or decide it’s a productivity mirage. Every indicator I see is the opposite: we wish we had more compute. That doesn’t scream crisis. There is a subtler issue, though. The old machines of capital — looms, mills, lathes — compounded value and ran for years. These new machines are fragile. Unlike dark fiber that sits in the ground, this compute capital has an unpredictable lifespan and real externalities: cooling, energy, water, and popular opposition to data centers, an unrealized liability that planners are only now confronting. The economics of industrial capital took a century to understand. We’re trying to figure this one out in three years. But as far as Apple is concerned, all of this is an externality. Editor’s Note: This was one of the questions asked by participants in Asymco’s June 2026 Office Hours live Q&A session, open to Asymco One subscribers. Subscribe to Asymco One https://asymco.com/one for more in-depth analysis and commentary like this.