Why Big Tech’s AI Spending Is $3 Trillion Higher Than It Seems - WSJ

Big Tech companies—primarily Microsoft, Google, Amazon, and Meta—are investing hundreds of billions of dollars annually in artificial intelligence infrastructure. This means buying physical servers, graphics processing units (GPUs, the specialized chips that power AI training), cooling systems, and data-center real estate. Under U.S. accounting rules (GAAP), companies face a critical decision: capitalize this spending (record it as a long-term asset that appears on the balance sheet and gets depreciated—spread across its useful life) or expense it (record it immediately as a cost that reduces profit this quarter). The article argues that if these companies expensed the full $3 trillion instead of capitalizing it, their reported earnings would look far weaker. This reveals a tension between how financial statements portray profitability and the true cash outflow investors are funding.

• Big Tech capitalizes AI server and GPU purchases as long-term assets, spreading their cost over years via depreciation rather than expensing them all at once.
• Capitalizing $3T in hardware makes current-year earnings appear $3T higher than if the same spending were expensed, a choice allowed by accounting rules.
• The capitalization decision affects both the income statement (lower expenses = higher profit) and balance sheet (higher assets), influencing valuation ratios investors rely on.
• Depreciation expense kicks in over time as capitalized assets are used up, so full profit impact is delayed, not eliminated—but timing matters to stock price.
• Identical spending can produce vastly different reported earnings depending on the asset-versus-expense classification, raising questions about comparability across firms.

capitalize

Record a purchase as a long-term asset on the balance sheet rather than as an immediate expense; the cost is spread (depreciated) over the asset's useful life.

depreciation

The process of allocating the cost of a long-term asset (like equipment) as an expense over its useful life, matching the asset's wear or obsolescence to the periods it generates revenue.

long-term operational assets

Physical items or rights (buildings, machinery, equipment, vehicles, patents) that a company expects to own or use for more than one year and that help generate revenue.

GAAP

Generally Accepted Accounting Principles — the standard rules U.S. public companies must follow when preparing financial statements.

balance sheet

A financial statement that lists what a company owns (assets), what it owes (liabilities), and the owners' stake (equity) at a specific point in time.

income statement

A financial statement that shows a company's revenue, expenses, and profit (or loss) over a period, typically a quarter or year.

GPU

Graphics processing unit — a specialized computer chip originally designed for graphics but now essential for training artificial intelligence models very quickly.

useful life

The expected number of years or units a long-term asset will be productive and generate value before it becomes obsolete or worn out.

1. 01

   What makes a $1 billion server purchase a 'long-term operational asset' rather than just an expense, and what does the company expect it to do over that time?

   Show answer

   Answer: A server purchase is a long-term operational asset because the company plans to own and use the hardware for multiple years (often 5–10 years) to generate revenue from AI services. Unlike a one-time consultant fee, the hardware retains value and utility across many reporting periods, so matching principle requires spreading its cost across those periods via depreciation.
2. 02

   If Microsoft capitalizes $50 billion in AI hardware this year, when and how does that spending finally hit the income statement as an expense?

   Show answer

   Answer: The $50 billion is capitalized (recorded as an asset) when purchased. Each year, a portion is depreciated—expensed on the income statement—based on useful life (e.g., $5B/year over 10 years). So the full $50B profit impact is realized over the asset's life, not immediately. Investors should note that the income statement profit looks artificially high in year one because depreciation expense is still small relative to the capital spend.
3. 03

   Does capitalizing versus expensing the same $3 trillion change how much cash Big Tech actually spent, or only how the company reports the timing of that cost?

   Show answer

   Answer: Capitalization versus expensing does NOT change how much cash left the company—cash outflow is identical either way. The choice only affects WHEN the cost appears on the income statement and WHERE it sits on the balance sheet. This is why some critics argue capitalization hides the true cash drain: profit looks healthier, but the underlying cash burn is just as real.

Open by sketching two columns on the board: 'Expense $3T Now' and 'Capitalize $3T Over 10 Years.' Show the income-statement difference: expensing kills earnings in year one; capitalizing spreads the hit, leaving year-one profit much higher. Then highlight a single company (e.g., Microsoft) and ask: 'If they told you their earnings are $50 billion, how much of that is real profit versus just delayed accounting of hardware they already paid cash for?' Use the depreciation table analogy—like a lease on a car you own outright, the benefit is real over time, but the expense doesn't all hit on day one. Stress that both methods are legal under GAAP; the rule lets management choose based on the asset's expected life. Close by pointing out: this choice matters because Wall Street and investors judge companies on profit ratios, so capitalizing inflates those ratios relative to expensing, even though cash spent is the same.