Early last year, two of the biggest technology companies in the world looked at the same kind of equipment and came to opposite conclusions. Meta extended the useful life of its servers and network equipment to five and a half years and cut about $2.9 billion from its 2025 depreciation. Amazon shortened the life of a portion of its servers from six years to five, citing how quickly AI hardware is improving. Same racks, same chips, opposite directions.

That has become a very public argument about GPU and server depreciation, with investors accusing the cloud giants of stretching asset lives to pad earnings and the companies responding that older hardware keeps working long after the newest workloads move on. I don’t think anyone is being dishonest. What the fight proves is something every fixed asset person already knows. Useful life is judgment, not fact, and someone made a call and wrote it down.

The difference is that those companies know they made a call. They can tell you when they last reviewed it and what the change was worth, because a change in useful life is a change in accounting estimate and it gets disclosed. In more than a hundred implementations, I can count on one hand the clients who could answer that question about their own fixed asset register without someone going off to find the asset class table from the original conversion.

That table always has a backstory. The life came from the old system, or a peer company, or a consultant’s template, or the tax life because it was right there. It made sense at the time. Then the business kept moving and the table stayed put. IT went to a three-year laptop refresh while the books kept depreciating over five, so every year a batch of perfectly good laptops is retired with value still on them and the disposal loss gets explained as “just how it works.” Servers get handed down from production to test and run happily for years at zero book value. Software is capitalized over five years and replaced in three when the vendor sunsets the version. Nobody did anything wrong. Accounting entered what the policy said, IT refreshed on the cycle the budget allowed, and the gap between the estimate and reality lived quietly in the space between departments.

Software is about to get its own version of this. FASB’s update to internal-use software capitalization drops the old project stages and starts capitalizing when a project is authorized, funded, and probable to finish. Reasonable, and also one more judgment for finance to own and explain.

Which brings me to leadership. Reviewing a useful life isn’t a system setting. It’s a habit, and habits belong to people.

The controller who asks once a year whether the server life still matches how IT actually runs the data center is protecting the balance sheet more than any downstream reconciliation will. The CFO who asks how many assets were retired with book value left, and why, is asking the one question that tells you whether the estimate is still true. Those conversations are a little awkward because they mean admitting a number everyone relied on may have drifted. They’re still far more comfortable than defending that number to an auditor, or, if you’re a hyperscaler, to the entire internet.

Most of us will never have analysts picking apart our depreciation schedules, which is exactly why the review has to come from inside. Useful life is judgment, and judgment belongs to people. Your servers, laptops, and software have a life on your books because somebody decided they did. It’s worth knowing who, and when they last checked.

If you are not sure who set the lives in your own register, or when anyone last looked, that is exactly the kind of question a Fixed Asset Health Assessment starts with. Start a conversation →

References