Accounting deep dive · August 2026
The Useful-Life Lever: When One Accounting Estimate Moves Billions of AI Earnings
In 2025, META added $1.00 to diluted earnings per share without selling another advertisement. The reason was not a new product or a better quarter. It was a longer estimate for how long its servers would remain useful.
Longer useful lives are not automatically aggressive accounting. But when the estimate changes earnings by billions, I need to separate operating improvement from expense timing—and ask what physical evidence supports the new life.
The mechanism in one clean example
Assume a company places a $6 billion server fleet into service, expects no salvage value, and depreciates it in a straight line. Nothing about the hardware changes in the example below. Only the accounting estimate changes.
Annual depreciation
Annual depreciation — $200M less expense per year
Illustration, not a company forecast. The total $6 billion cost does not disappear. It reaches the income statement more slowly. Cash generally left when the equipment was purchased, so the estimate changes reported earnings—not the original cash outlay.
Four companies, four material effects
| Company | Change | Reported earnings effect |
|---|---|---|
| META | Most servers and network assets extended to 5.5 years in 2025 | +$2.59B net income +$1.00 diluted EPS |
| AMZN | Five to six years in 2024; a subset returned to five in 2025 | +$2.5B in 2024 −$1.0B in 2025 |
| ORCL | Five to six years beginning fiscal 2025 | +$573M net income +$0.20 diluted EPS |
| GOOGL | Servers moved from four to six years in 2023 | +$3.0B net income +$0.24 diluted EPS |
Why Meta's $1.00 matters
META disclosed that its 2025 useful-life change reduced depreciation by $2.92 billion and increased net income by $2.59 billion. Against reported 2025 operating income of $83.28 billion and net income of $60.46 billion, my calculation puts the benefit at 3.5% of operating income and 4.3% of net income. SEC filing ↗
That is not a footnote-sized effect. It also is not evidence that the underlying business weakened. Server and network depreciation still climbed from $2.63 billion in the first quarter of 2025 to $4.38 billion in the first quarter of 2026 as the infrastructure base expanded. The change softened the expense curve; it did not stop the curve. Q1 filing ↗
Amazon is the counterexample that keeps this honest
AMZN extended server lives from five to six years in 2024, reducing depreciation by $3.2 billion and increasing net income by $2.5 billion. One year later, it moved a subset back to five years because of the faster pace of development in AI and machine learning. The reversal increased 2025 depreciation by $1.4 billion and reduced net income by $1.0 billion. 2025 filing ↗
This is the part I would underline twice. Better software, component reuse, and redeployment can make servers productive for longer. Faster hardware cycles can also make yesterday's fleet obsolete sooner. Both can be true. The useful-life estimate is management's judgment about which force wins.
The accounting tail is getting larger
ORCL nearly doubled gross computer and network equipment between fiscal 2025 and fiscal 2026 while construction in progress more than doubled. GOOGL more than doubled first-half capital expenditures year over year, while depreciation grew 43%. That gap is partly timing: assets are not depreciated until they enter service, and then their cost is spread across years. Today's capex announcement can therefore become tomorrow's earnings pressure slowly—and then all at once as projects leave construction in progress.
What I would monitor each quarter
- Capex versus depreciationA widening gap shows how much future expense has not fully reached the income statement.
- Construction in progressThese assets are not yet depreciating. Watch when they enter service.
- Accelerated retirementsEarly write-offs are the clearest evidence that a prior life estimate ran too long.
- The operational explanationSoftware optimization and redeployment are evidence. “We reassessed” by itself is not much of one.
The fair counterargument
A longer life can be the right estimate. The accounting rules require management to revise estimates when experience changes, and better fleet management can genuinely keep hardware productive. My concern is not that a change occurred. It is whether I accidentally credit an accounting benefit to pricing power, demand, or efficiency that did not produce it.
When a useful-life change is material, I will show earnings both with and without the disclosed benefit. Changing the scorecard is not the same thing as improving the swing—even when the new scorecard is perfectly legitimate.
Primary-source ledger
All company figures are reported unless I label them as my calculation.