Exploration · Ch 8 · Long-Term Operational Assets

The Depreciation Race.

You just spent $10.0B on AI infrastructure. Same building. Same servers. Two depreciation methods, one useful-life estimate — and the numbers on your income statement, balance sheet, and cash flow statement move very differently.

Your capex

Estimated useful life5 years

Salvage value fixed at 5% of cost ($500M). Revenue and operating expenses (before depreciation) are modeled as 100% and 40% of asset cost per year. Tax is deliberately omitted — MACRS (IRC §168) governs tax depreciation regardless of the GAAP method you pick here, so mixing them would confuse the story.

Depreciation expense by year

Pick a year to see the three-statement impact

Same asset, same operations. Different method. Watch where the choice shows up on each statement.

Straight-line · Year 1

Income Statement

Revenue$10.0B

Operating expenses−$4.00B

Depreciation expense−$1.90B

Net income$4.10B

Balance Sheet · end of Year 1

Asset (at cost)$10.0B

Accumulated depreciation−$1.90B

Net book value$8.10B

Statement of Cash Flows · Operating (indirect)

Net income$4.10B

+ Depreciation (non-cash)$1.90B

Cash from operations$6.00B

Double-declining balance · Year 1

Income Statement

Revenue$10.0B

Operating expenses−$4.00B

Depreciation expense−$4.00B

Net income$2.00B

Balance Sheet · end of Year 1

Asset (at cost)$10.0B

Accumulated depreciation−$4.00B

Net book value$6.00B

Statement of Cash Flows · Operating (indirect)

Net income$2.00B

+ Depreciation (non-cash)$4.00B

Cash from operations$6.00B

The cash-flow beat

DDB reports $2.10B LESS net income than straight-line — but cash from operations is identical at $6.00B under both methods.

This is the “non-cash” property of depreciation made visible. On the income statement, depreciation reduces net income. On the cash flow statement, we add it right back — because no cash left the business when we recorded it. The two effects cancel exactly, so the method choice affects REPORTED INCOME and BOOK VALUE, but not cash. (Tax depreciation lives on a separate schedule prescribed by MACRS — not affected by the GAAP method you pick here.)

Depreciation and book value — year by year

Year	SL depreciation	SL book value	DDB depreciation	DDB book value
1	$1.90B	$8.10B	$4.00B	$6.00B
2	$1.90B	$6.20B	$2.40B	$3.60B
3	$1.90B	$4.30B	$1.44B	$2.16B
4	$1.90B	$2.40B	$864M	$1.30B
5	$1.90B	$500M	$518M	$778M

Real world · What the hyperscalers actually did

Extending your useful-life estimate from 5 years to 6 years would report $317M less depreciation in Year 1 — same physical asset, one estimate changed.

Between 2022 and 2024, every US hyperscaler made this exact move on their server fleets:

Company	Server life	Estimated impact
Microsoft	4 → 6 years (FY23)	~$3.7B reduced depreciation → boosted FY23 income
Meta	4 → 5 years (2022)	~$3.4B benefit reported in 2022 filings
Google	to 6 years (2023)	~$3.9B reduction to full-year depreciation
Amazon (AWS)	3 → 5 years (2022), further changes in 2024	Billions in reduced depreciation each year

Same servers, same racks, same electricity. An accounting estimate change added tens of billions of dollars to reported aggregate income across the four companies. Whether the longer lives are realistic depends on how long GPUs stay useful before the next generation displaces them — a technical question that lands on the income statement as a very large number. Analysts flagged the pattern as aggressive at the time.

This exploration teaches the GAAP method choice in isolation. Tax depreciation is prescribed by MACRS (IRC §168) regardless of the GAAP method chosen for financial reporting, so companies keep two sets of books and the GAAP choice doesn’t affect cash tax. That’s a whole separate chapter — but leaving it out here lets the “same asset, same cash, different reported income” story land cleanly.