AI Capex Depreciation Risk: The Hidden Margin Threat

The Accounting Lag: Construction vs. Energization

When a hyperscaler spends on land, facilities, and hardware, cash is used immediately. The income statement does not necessarily absorb that cost at the same time.

Instead, qualifying costs are capitalized on the balance sheet. Alphabet’s FY2024 Form 10-K reported $52.5 billion in capital expenditures, driven by technical infrastructure. The filing also states that assets not yet in service are classified as under construction or assembly, and that straight-line depreciation begins when assets are deployed and ready for their intended use (Alphabet SEC Form 10-K).

The accounting sequence is straightforward:

• Capital outlay: Cash is spent on land, facilities, and hardware.

• Construction in progress: Assets are capitalized but do not yet produce depreciation expense.

• Energization: The data center or equipment is deployed and ready for intended use.

• Depreciation begins: Straight-line depreciation starts reducing GAAP operating profit.

This creates a timing mismatch. A company can spend heavily in one period while the margin impact arrives later. As assets under construction move into service, the income statement begins to reflect costs that were already visible in cash flow and on the balance sheet.

Useful Life Extensions Can Spread the Expense

Accounting assumptions can also shape the pace of recognition. When the estimated useful life of server or network equipment is extended, the same capitalized cost is depreciated over a longer period, reducing annual depreciation expense compared with a shorter life.

Microsoft disclosed in its FY2024 Form 10-K that it extended the estimated useful life of server and network equipment from four years to six years (Microsoft SEC Form 10-K).

That change may reflect how long the equipment is expected to remain useful. It also means near-term depreciation expense is lower than it would have been under the prior estimate. In a period of heavy AI infrastructure buildout, that distinction matters for margin analysis. If the economic usefulness of AI hardware proves shorter than its accounting life, companies could face pressure through faster depreciation or impairments later. That is the core earnings risk discussed in Big Tech AI Capex Depreciation: The Earnings Risk Investors Miss.

Why ROIC Is Vulnerable

ROIC is generally measured as Net Operating Profit After Tax divided by Invested Capital. The filings alone do not provide a complete company-specific ROIC bridge for AI infrastructure, and the companies do not isolate every AI-related asset in a directly comparable way. Still, the mechanism is clear: invested capital expands as infrastructure spending is capitalized, while operating profit later absorbs depreciation as assets enter service.

If revenue from newly deployed AI capacity grows fast enough, margins can remain resilient. If utilization or pricing fails to keep pace with the depreciation burden, operating margins can compress. In that case, ROIC can decline because the capital base has grown faster than operating profit.

This is why the depreciation issue is different from the free-cash-flow debate. Capex pressures cash flow immediately. Depreciation pressures operating income later. Both matter, but they affect financial statements at different times.

What Remains Unclear

The filings document the accounting mechanics, but they do not answer the central forecasting question: when, and how much, will the depreciation burden rise as current infrastructure projects enter service?

The specific energization dates and capacity ramps for assets in construction-in-progress are not disclosed in a way that allows a clean quarter-by-quarter schedule. Future revenue yields on those assets are also unknown. Strong AI compute demand could offset the depreciation drag. Slower utilization would make the margin impact more visible.

There is another limitation: not every disclosure category maps cleanly to AI alone. Technical infrastructure, cloud capacity, data centers, and equipment spending can include more than AI workloads. Any analysis that treats all infrastructure capex as AI capex risks overstating the precision of the conclusion.

Watchpoints In Upcoming Filings

The next useful signals are likely to come from depreciation expense, property and equipment additions, construction-in-progress disclosures, and any further useful-life changes in company filings.

The key question is whether depreciation begins rising faster than the revenue and operating profit tied to the new capacity. If it does, the accounting lag will have moved from balance-sheet footnote to margin pressure. If revenue scales faster than depreciation, the same capital cycle may look far less threatening.

For now, the risk is best understood as a delayed earnings-sensitivity issue, not a verdict on AI demand. The spending is already visible. The full income-statement effect may still be arriving.

***

*Disclaimer: This article is for informational purposes only and does not constitute financial, investment, tax, or legal advice. The analysis of accounting mechanics and potential margin risks is not a prediction of future stock performance or a recommendation to buy, sell, or hold any security. Readers should consult with a licensed financial professional before making any investment decisions.*