When a company buys an AI processor for $30,000 or $40,000, it does not record the whole amount as an expense on the day of purchase. It 'depreciates' it, spreading the cost over the years it expects to use it. If the company estimates the processor will serve it for six years, a sixth of the cost is expensed each year. If only three, a third each year. That difference sounds technical, but it dramatically changes the profit the company reports.
In recent years, the cloud giants gradually extended their servers' depreciation periods, to five or even six years. Each extension cut the annual expense and raised reported profit by billions. The problem is that in AI the pace is the opposite: Nvidia has moved to a new generation almost every year, and each generation offers far more performance per dollar and per watt. Critics argue a three-year-old processor is no longer competitive, so long depreciation periods inflate profits.
The debate sharpened this week. CoreWeave CEO Michael Intrator argued the company has a revenue opportunity of nearly $10 billion from older processors, without further investment. That is an argument for the optimistic side: even when new chips exist, there are customers willing to pay for old ones, for example to run models rather than train them. On the other hand, demand for old chips depends on the whole market staying hungry for compute.
Why does this matter to everyone, not just accountants? Because cloud companies, especially neoclouds like CoreWeave, finance chip purchases with debt, and the collateral for the loans is in many cases the chips themselves. If chips turn out to lose value faster than expected, the collateral shrinks, book profits are wiped out and lenders start asking questions. That is the mechanism that could turn a slowdown in AI demand into a financial crisis.
How to follow it? In cloud companies' reports, look for changes in depreciation periods, the ratio between investment and revenue and the prices at which older processors are rented. As long as those prices are stable, the optimists are right. The day they start to plunge, this accounting question will become the most important question in the market.
