The Two Numbers Dominating Tech
In the world of Big Tech right now, two terms are unavoidable: capital expenditures (capex) and depreciation. Capex is the eye-watering amount of money companies like Google (Alphabet) are spending on the physical gear needed for the AI revolution—think
servers packed with powerful GPUs, sprawling data centers, and high-speed networking equipment. This spending is immense, with Alphabet projected to spend nearly double in 2026 what it did the previous year. But that cash outflow is only the first part of the story. The second part, depreciation, is how the cost of those assets hits the income statement. Instead of taking a one-time hit, the cost is spread out over the asset's 'useful life.' This isn't just boring accounting; it's a strategic lever that directly impacts the quarterly profits that investors watch so closely.
The 'Useful Life' Dilemma
Here’s where the critical detail emerges. The 'useful life' of an asset is an estimate of how long it will be economically valuable. For a building, that might be 30 years. For a traditional computer server, it used to be around three to four years. However, as AI competition has intensified, a fascinating tug-of-war has emerged. On one hand, AI hardware like GPUs can become obsolete incredibly quickly as newer, more powerful chips are released annually. This would logically argue for a shorter useful life. A shorter depreciation period means higher annual expenses, which would eat into profits more quickly. On the other hand, some companies have extended the useful life of their servers to five or even six years. A longer useful life reduces the annual depreciation expense, making current earnings look much healthier. The difference is massive: depreciating $660 billion in assets over three years creates a $220 billion annual expense, while a six-year schedule cuts that in half to $110 billion.
What Google's Moves Signal
Google has been right in the middle of this debate. In early 2023, the company extended the useful life of its servers from four to six years, a move that significantly boosted its earnings that year. However, in a telling adjustment in 2025, it shortened the life of some servers back to five years, explicitly citing the faster pace of AI development. This isn't a contradiction; it’s a signal. By fine-tuning these estimates, Google is communicating its view on the longevity of its specific hardware. Extending the life might suggest confidence that its custom-designed TPUs (Tensor Processing Units) have a longer shelf life than off-the-shelf GPUs. Shortening it acknowledges the relentless pace of innovation and the need to stay on the cutting edge. This detail reveals a company balancing the pressure for short-term profits against the reality of a rapid, expensive technology cycle.
Why Investors Are Watching Closely
This accounting choice has become a central point of debate for Wall Street. Critics, including famed investor Michael Burry, argue that extending the useful life of rapidly advancing tech is aggressive accounting that artificially inflates earnings and masks the true, capital-intensive nature of the AI business. If depreciation schedules are too long, companies risk overstating their profitability today, only to face a larger financial cliff when hardware needs replacing sooner than planned. For Google, the metric to watch is whether the massive revenue growth from AI-powered services, particularly in Google Cloud, can outpace the tsunami of depreciation expenses that will inevitably follow its capex surge. The company is currently spending far more on new infrastructure than it is depreciating, creating a gap that represents a future drag on earnings. Depreciation, therefore, has transformed from a footnote into a key indicator of a company’s operational honesty and its ability to generate a real return on its staggering AI investments.
