The accelerated depreciation of AI infrastructure is becoming a hidden pressure source on the profits of tech giants.

Written by: Li Jia

Source: Wall Street Insight

As the market immerses itself in the grand narrative of AI-driven growth, a long-underestimated variable is emerging: the accelerated depreciation of AI infrastructure may be becoming the most hidden yet far-reaching pressure source on the profitability of tech giants.

The core contradiction lies in the fact that the technological iteration of AI servers and GPUs is significantly shortening the economic lifespan of hardware, while the scale of capital expenditure matching this is expanding at an unprecedented pace, thereby driving up depreciation costs and continuously eroding profit margins.

It is estimated that in 2023, the combined capital expenditures of Google, Microsoft, Amazon, and Meta will reach about $750 billion, nearly half of the UK's annual fiscal expenditure; meanwhile, the total annual depreciation expenses of these four companies have nearly doubled from two years ago, rising to about $116 billion.

As the computing power infrastructure deployed intensively over the past 18 months gradually enters the depreciation cycle, this pressure continues to accumulate. Amazon has taken the lead in reducing the useful life of its data center assets from six years to five years, citing the significant acceleration of AI and machine learning technology iterations. This adjustment is not an isolated case but more like a preliminary signal of industry cyclical changes.

Currently, Meta, Microsoft, and Alphabet still maintain a six-year depreciation period, but the market generally expects that subsequent follow-up adjustments are only a matter of time. Once the industry universally reduces the useful life, depreciation expenses will further rise, transforming from a "buffer item at the accounting level" into a key variable that directly affects the profit and loss statement, thereby reshaping the market's pricing framework for the profitability quality of tech giants.

Unprecedented expansion of capital expenditures, but return realization still lags

Since 2023, the average stock price of the four tech giants has doubled, but the pace of capital expenditure expansion is even more aggressive, with quarterly capital expenditure budgets increasing approximately fourfold during the same period, clearly outpacing stock price performance.

On the financing side, Alphabet has raised about $85 billion through debt financing in the past year and plans to further raise about $80 billion through equity financing, an unprecedented scale. However, this financing path does not possess sustainable and linear replication capability.

Meanwhile, the expansion of AI infrastructure is gradually approaching real constraints. Bottlenecks in chip supply, power systems, and water resource infrastructure are beginning to emerge, with substantial resource constraint signals appearing in some developed regions. On the commercialization front, most AI projects are still in the investment phase and have not yet formed stable cash flow recovery capabilities, further amplifying the reliance on external financing.

Maintenance and upgrade costs are underestimated, depreciation pressure may still be rising

The market has long focused on new data center investments, but the maintenance and upgrade costs of existing assets are also accumulating.

From industry experience, the economic lifespan of data center servers is typically three to six years, but under the dual effects of high-intensity AI computing power consumption and rapid technology iteration, the actual asset cycle of ultra-large-scale cloud service providers is converging towards the range of three to five years.

More critically, about two-thirds of the costs in AI data centers are concentrated on the equipment itself. Once the future equipment replacement cycle is incorporated into a unified depreciation framework, the overall "real pressure" of capital expenditures will be significantly higher than the current market's implicit assumptions about the average depreciation path.

As large-scale equipment enters the peak of depreciation, if technological iteration further accelerates and forces companies to increase replacement frequency, the profit and loss statements of tech giants will continue to be under pressure. The long-term return logic of AI now depends not only on the speed of demand expansion but also on whether companies can maintain the sustainability of their financial structure amid high-intensity capital consumption and the upward cycle of depreciation.