The Perils of Adjusted EBITDA
On Monday morning Michael Burry tweeted out more details about his AI Bubble thesis. His claim is that the AI hyperscalers are “understating depreciation” by extending the period over which they are depreciating the Nvidia chips and all the other capital equipment they are buying to build out AI. This morning Jim Chanos applied the same logic to CoreWeave (CRWV).
The first thing to understand is that Net Income is an accounting number. It is not the amount of cash that the company actually earns because it reflects certain approximations – one of the most important of which is depreciation.
Let’s say a company buys a factory for $50 billion. It doesn’t expense that entire amount in the year it spends the money. Instead it depreciates it over the useful life of the asset. But the useful life is an estimate. Nobody knows ahead of time how long the asset will be productive.
If you depreciate it over 10 years, that’s a $5 billion expense per year. If you depreciate it over 20 years, the annual depreciation expense goes down to $2.5 billion. Therefore, the depreciation expense used for capital expenditures is an approximation – but an extremely important one in determining the all important bottom line i.e. Net Income that investors fixate on.
Let’s take a look at CoreWeave (CRWV) which reported 3Q25 earnings after the close Monday. Adjusted EBITDA more than doubled to $838 million from $379 million a year ago. On the surface, then, they are a profitable company – at least on this metric. But they backed out $630 billion in Depreciation and Amortization from their GAAP Net Loss of $110 million to arrive at that number.
It’s only when we turn to the Cash Flow statement that we understand the how CoreWeave is financing their operations. Cash Flow from Operations in the first 9 months of 2025 was $1.5 billion. CapEx was $6.25 billion. Those are cash figures, not accounting ones. In other words, Free Cash Flow for the first nine months of the year was -$4.75 billion. That’s right: CRWV has burned up nearly $5 billion so far this year. How are they financing this? By selling debt. Essentially the whole difference is made up by their net debt issuance of ~$4.5 billion.
Where does this leave us? Clearly the hyperscalers are spending an enormous amount of money on Nvidia chips and the other capital equipment required to build out AI. Their thesis is that the ROI on these investments over the long term will be excellent. If AI is truly the game changer many say it is, the returns may well outweigh any current concerns about the accounting. If not, a lot of this CapEx may be malinvestment and have to be written down in the future.
The overall point is that we are operating in the dark here because we don’t know the future returns on investment and we don’t know the appropriate rate to depreciate these huge Capital Expenditures to get the right Net Income numbers in the present.
And so what happens is that people take sides based on their biases. The permabulls think this is a New Era while the permabears think it’s a Bubble. My own bias is that this is similar to the Dot Com build out of the internet in the late 1990s. Ultimately the internet revolutionized the economy and everyday life – but not before a nasty bust in the first years of the 2000s wiped out an enormous amount of malinvestment.
My best guess is that AI will have a similar effect on the economy and everyday life – but first there will be a reckoning in which a lot of malinvestment is revealed. The trillion dollar question is the timeframe over which this plays out…..
