What bubble? JPMorgan says the $5.5 trillion AI capex explosion is profitable—for now - Fortune

JPMorgan's analysis of the $5.5 trillion AI infrastructure spend illustrates how companies capitalize large outlays for equipment and data centers as long-term assets rather than immediate expenses, and why sustained profitability determines whether that massive capex bet pays off.

Read article

Teaching notes are auto-generated. Worth a fact-check before class.

Why this matters

JPMorgan Chase, a major investment bank and financial advisor, recently published research arguing that the massive spending by tech giants (and other companies) on artificial intelligence infrastructure—totaling $5.5 trillion collectively—is justified by the profits those investments are generating. When a company like Microsoft or NVIDIA builds a data center, buys servers, or develops AI chips, it doesn't expense the entire cost in one year. Instead, it records the outlay as a long-term operational asset (like "property, plant, and equipment") on the balance sheet and expenses it gradually over many years through depreciation. JPMorgan's message: this huge capex wave makes sense because the AI systems are producing revenue and earnings that exceed the annual depreciation cost. The concern some critics raise—that companies are throwing money at AI hype—is real, but JPMorgan's data suggest, at least so far, the capex is paying for itself.

Key points

Tech companies spent $5.5 trillion on AI infrastructure (servers, data centers, chips), recorded as long-term assets rather than immediate expenses, which is the capitalization principle in action.
JPMorgan's test: if annual profits from AI exceed the annual depreciation (the portion of capex expensed each year), the capex is economically justified, tying asset cost allocation to profitability.
If AI capex later fails to generate promised returns, companies may write down the asset value, erasing the balance-sheet benefit and reducing reported earnings—a real consequence of overinvestment.

Key terms

Capex (capital expenditure): Money spent to buy or build long-term physical assets (equipment, buildings, vehicles) that will generate revenue over multiple years, not expensed immediately.
Long-term operational asset: Property, equipment, or infrastructure expected to help a company generate revenue for more than one year; appears on the balance sheet and is expensed gradually via depreciation.
Depreciation: The annual expense recorded to reflect the wear and tear or gradual use of a long-term asset; spreads the asset's cost over its useful life instead of expensing it all at once.
Capitalization: The accounting practice of recording a large expenditure as an asset on the balance sheet rather than expensing it immediately; used for purchases that will benefit the company over multiple years.
Write-down (or impairment): An adjustment that reduces the recorded value of an asset on the balance sheet because it is no longer worth what the company paid for it; reduces reported earnings in the year the write-down occurs.
Balance sheet: A financial statement showing what a company owns (assets), owes (liabilities), and the shareholders' stake (equity) at a specific point in time.
Data center: A large facility housing computer servers and networking equipment; for AI companies, it's the physical infrastructure that runs machine-learning models and stores data.

Discussion prompts

01
Why does GAAP accounting require a company to capitalize (record as an asset) a $500 million data center instead of expensing it all in year one?
Answer: Students should recognize that capitalizing a data center matches the asset's cost to the periods that benefit from it (matching principle). Expensing $500M immediately would understate profit year one and overstate it in years 2–10 when the asset is generating revenue. The asset's benefit spans multiple years, so its cost should too.
02
If a chip manufacturer spends $10 billion on a new fab (factory) and depreciates it over 10 years, but the fab only generates $800 million in profit per year, what might that signal about the investment decision?
Answer: The key point is that annual depreciation (~$1B/year) exceeds annual profit ($800M), meaning the company is not recovering its investment cost through earnings. This suggests either the fab is underutilized, prices are too low, or the capex was a poor decision. Students should see that profitability is the true test of whether capex made sense.
03
JPMorgan's argument hinges on profitability 'for now'—what would cause a company to write down $1 trillion in AI assets, and how would that affect the income statement?
Answer: A write-down occurs when the asset's fair value falls below its book value (original cost minus accumulated depreciation). If AI demand collapses or competing technology emerges, companies must record an impairment charge, reducing the asset's balance-sheet value and reporting a loss in that year. This is a painful but honest correction.

Bringing it to class

Start by drawing a simple two-column chart: 'Year 1' (capex $1B, expense $0) vs. 'Year 10' (capex $0, expense ~$100M depreciation). Ask students which year 'looks better' on the income statement and why that's misleading. Then introduce JPMorgan's insight: if annual profit from the asset exceeds depreciation, capex created value; if not, it destroyed it. Use a real analogy—buying a delivery truck for $100k and depreciating it over 5 years makes sense only if the truck earns more than $20k/year in net revenue. Conclude by asking: if we don't know yet whether AI will be profitable, why is capitalization the right choice? (Answer: we assume the benefit will materialize; if it doesn't, we write it down.)