The AI boom is powered by mountains of GPUs, with tech giants spending billions on these chips. But behind the headline purchases lies an accounting detail—depreciation—that can paint a misleading picture of a company's true financial health.
The Shopping Spree: It’s Not an Expense, It’s an Investment
When a company
spends billions on GPUs, that cash outlay doesn't immediately appear as a giant expense on its income statement. Instead, this spending is treated as a Capital Expenditure, or CapEx. Think of it like buying a factory or a fleet of delivery trucks. The company is acquiring a long-term asset that will generate value over several years. This purchase hits the cash flow statement as a major cash outflow, but it doesn't directly reduce the company's reported profit for the quarter. The GPUs are added to the balance sheet as property, plant, and equipment (PP&E), representing a tangible investment in the company's future growth.
The Slow Burn: What Depreciation Really Means
Since the GPU hardware won't last forever, its cost has to be accounted for over its 'useful life'. This is where depreciation comes in. Depreciation is an accounting method used to gradually allocate the cost of a tangible asset over time. Each year, a portion of the GPU's original cost is recorded as an expense on the income statement, which reduces the company's taxable income. For tax purposes in the U.S., computers and servers typically fall into a 5-year recovery period under the MACRS system. However, for financial reporting, companies estimate the asset's 'useful life' themselves—and this is where things get interesting.
The Critical Mismatch: Accounting Life vs. Economic Life
Here is the detail investors often miss: a GPU’s 'accounting life' can be very different from its 'economic life'. A company might decide to depreciate its new GPU servers over five or six years, an estimate that some major cloud providers use. This longer schedule results in a lower annual depreciation expense, which in turn boosts reported net income and earnings per share. However, in the fast-paced world of AI, a GPU’s peak competitiveness for cutting-edge tasks might only be two or three years before a new, more powerful generation arrives. This creates a mismatch: the books suggest the asset has years of high-value life left, while its real-world economic power may be declining much faster. This is the heart of a debate that has drawn in prominent investors, who argue that some companies are flattering their profits by using overly optimistic depreciation schedules.
How This Skews the Picture for Investors
An aggressive depreciation schedule—say, six years instead of four—can make a company appear more profitable than it truly is. For example, one analysis noted that Microsoft's move to a six-year life for servers added $3.7 billion to its fiscal 2023 operating income. While a longer useful life may be defensible if the GPUs can be repurposed for less demanding tasks like inference, it's a critical assumption for investors to question. To get a clearer picture, investors should compare a company's capital expenditures to its depreciation expense over time. If a company is consistently spending far more on new hardware (CapEx) than it's depreciating, it suggests a costly replacement cycle that might not be fully reflected in the rosier earnings-per-share number. The real cost is visible in the cash flow statement, not just the income statement.
