The Depreciation Panic Around AI Stocks

Every few weeks, someone builds an entire investment thesis around how much AI companies are depreciating their infrastructure, as if that single line item is the smoking gun that proves the whole sector is overvalued. It isn’t. It’s a distraction, and the fact that it gets so much airtime says something uncomfortable about how much of “analysis” in this space is really just spreadsheet reflex.

Here’s the mechanism, stripped down. Depreciation reduces the earnings figure on the income statement. It does not reduce cash. It gets reversed in any cash flow calculation worth doing. So the only people for whom this is a genuine emergency are the ones whose entire methodology begins and ends with a P/E ratio or an EBITDA multiple, frameworks built for a completely different kind of business, applied here out of habit rather than judgment.

That’s the real issue. Multiple-based investing is a convenience, not a philosophy. It works reasonably well when a business is mature, its capital cycle is behind it, and earnings are a decent proxy for cash generation. Point that same toolkit at a company still in its build-out phase and you’ll get an answer, sure, just not a correct one. If that’s as deep as your process goes, the honest move isn’t to keep forcing it onto companies it wasn’t designed for. It’s to admit the limits of the tool and put your money in something passive instead.

There’s actually a category of business where the depreciation conversation is worth having: mature, capital-heavy infrastructure, think utilities, pipelines, toll roads. Those businesses spend enormous sums up front, then spend decades collecting revenue against an asset that’s already built. Depreciation there isn’t noise; it’s a structural feature that shapes reported earnings for years and genuinely affects how you think about cash flow. Fair enough, worry about it there.

An AI company doesn’t fit that shape at all. Nothing about it resembles a finished asset quietly generating tax shields while the capital spending winds down. These businesses are still mid-buildout, spending continuously and aggressively on compute, talent, and infrastructure just to stay in contention. There’s no “after the capex” phase yet. The spending is the strategy, ongoing and escalating, not a sunk cost being amortized off in the background.

So when the loudest objection to an AI investment is “look at what depreciation is doing to earnings,” that’s a signal to stop listening, not to start worrying. The real questions sit somewhere else entirely. Is there durable technical differentiation, or is this a thin layer over someone else’s model? Is the customer relationship sticky, or one product launch away from churn? Is the growth real demand, or subsidized adoption that evaporates when the discounts end? Does the path to sustainable margins actually exist, or is capital intensity permanent? Is leadership deploying capital with discipline, or spending because the rest of the industry is spending?

Answer those, and you’re doing real work. Fixate on the depreciation line, and you’ve mistaken an accounting artifact for insight, which is really just a more sophisticated way of not doing the analysis at all.

The lesson generalises past AI: match your framework to the business you’re actually looking at, not the business your framework was built for.