A debate is raging over the accounting treatment of Nvidia chips and other equipment that tech companies are splurging on. This time, some of the AI companies' critics may be overreaching.
Rarely have investors obsessed over a topic as seemingly banal as the proper depreciation schedule for fixed assets. But when spending on artificial-intelligence infrastructure is in the hundreds of billions of dollars at a handful of the world's biggest companies, the market pays attention.
Michael Burry, the famed investment manager played by Christian Bale in 2015's film "The Big Short," recently added fuel to the fire. "Extending useful life decreases depreciation expense and increases apparent profits," he wrote in an article last month. "It is one of the more common frauds of the modern era and results in overvalued assets and overstated profits."
Whatever the merits of that criticism, some perspective is in order.
This year, for instance, Meta Platforms increased the estimated useful lives for most of its servers and network assets to 5.5 years. It previously said it used a range of four to five years. As recently as 2020, Meta said it used as little as three years.
Meta said the latest extension reduced its depreciation expense by $2.3 billion for the first nine months of 2025. That isn't a small number. But to grasp the scale, its total depreciation expense was almost $13 billion, while pretax profit topped $60 billion.
Alphabet, Microsoft and Amazon.com also use longer useful lives for similar assets than they did five years ago. Alphabet and Microsoft are at six years, up from three in 2020. Amazon used a four-year period in 2020 and was up to six years by 2024, but this year cut the number to five years for some servers and networking equipment.
Companies include depreciation expenses in their earnings because eventually their fixed assets will wear out or become obsolete. The practice ensures that the costs for capital investments get recognized over time on the financial statements. Management is tasked with picking a certain number of years over which to write them down gradually.
Big tech companies for years have drawn investor scrutiny when they have extended the useful lives for the assets they are depreciating. That is understandable, because doing so can boost current earnings by shifting expenses further into the future. It looks cheeky, if not aggressive, for management to increase a year's earnings by billions of dollars with a keystroke, simply by changing an accounting estimate.
Ultimately, though, the better question may not be about the right number of years, but what the right depreciation method is. Straight-line is the prevailing method, where each year of depreciation expense for a given purchase is the same.
However, some types of assets fall in value more sharply early on, then stabilize and decline more gradually on a predictable curve. A case in point: According to Silicon Data, which tracks pricing for Nvidia chips, the average resale value for an H100 system in its third year of use recently was around 45% of the price for a new H100.
In a situation like that, so-called accelerated depreciation may have a better shot at reflecting economic reality than the straight-line method. Using a six-year period, the depreciation expense would be higher in the early years, when the economic benefit is consumed faster, and lower in later years.
Still, the difference in this case wouldn't be tremendous. Under the straight-line method, the accumulated depreciation during year three would be less than half of the purchase price. Using an accelerated timetable, it would be somewhat greater than half. And investors in the big AI hyperscalers are well aware that any expectations of big returns on today's investments are years down the line.
Fundamentally, most of the numbers in companies' financial statements are built on estimates, guesswork and assumptions. The expense known as depreciation is a construct, like so much else in accounting.
Precision is rare. Nobody will get a provably correct answer for how much a company's fixed assets diminish in value each year. The company's management may not know what a particular asset's useful life is, especially for high-demand tech equipment.
Management is supposed to record a larger write-down if asset values become severely impaired. But often the write-downs come only after the company's own stock price has collapsed, which obviously isn't the case for the Magnificent Seven.
If investors someday conclude that much AI investment is being wasted, it won't be due to the depreciation schedules companies have picked. While there are other good reasons to be skeptical of how technology giants are accounting for their AI adventures, the depreciation debate won't move the needle.
Comments