CapEx & Depreciation: Michael Burry vs. AI and Big Tech

CapEx & Depreciation Definition: In accounting, capital expenditures (CapEx) represent a company’s spending on long-term assets that will last for multiple years; since they will be useful for many years, this CapEx spending must be allocated over time and recorded as “Depreciation” on the Income Statement. This Depreciation number reduces the company’s Pre-Tax Income and counts as a tax deduction in the current period, but the CapEx does not.

Michael Burry, famous from The Big Short, recently made waves on X when he claimed that many Big Tech companies, such as Oracle and Meta, have overstated their earnings by understating their Depreciation:

Between 2020 and 2025, these companies extended the useful lives of the GPUs and data centers they use for AI, thereby reducing depreciation on the Income Statement.

For example, if Microsoft buys new Nvidia chips for $90 million, it records Depreciation of $30 million per year if the useful life is 3 years, but $18 million per year if the useful life is 5 years.

Burry claims that since many of these chips and servers are on 2-3-year “product cycles,” these Big Tech companies have significantly overstated their earnings, implying that they are currently overvalued.

This is not quite correct because the proper way to value a company is with metrics that normalize a company’s capital structure and Depreciation policies – such as EBITDA – and with a DCF based on Unlevered Free Cash Flow, which does something similar.

Ironically, depreciating assets more quickly can boost a company’s valuation because of the tax savings that come from this.

That said, higher Depreciation also tends to imply that future CapEx will be higher because companies must spend more to replace and upgrade their assets.

Therefore, the key question is not whether Depreciation is “understated,” but the proper mix of Depreciation and CapEx (as percentages of revenue).

Also, we need to know something about the future revenue and cash flows that might be generated by this very high CapEx, which no company has yet been able to estimate.

To explain how this works, we’ll start with a quick review of the standard accounting for CapEx and Depreciation:

Files & Resources:

• CapEx and Depreciation – Accounting and Valuation Examples (XL)
• CapEx and Depreciation – Presentation Slides (PDF)

Video Table of Contents:

• 0:00: Introduction
• 1:41: The Short Answer
• 5:19: Part 1: CapEx and Depreciation Accounting
• 8:57: Part 2: CapEx and Depreciation in Valuation and the DCF
• 11:56: Part 3: Why No One is “Right” About This AI CapEx Debate
• 13:31: Recap and Summary

How Do CapEx & Depreciation Normally Work on the Financial Statements?

To illustrate the accounting, let’s say that a company spends $100 on CapEx on January 1 of the year, such as $100 for a new data center or factory.

On the financial statements, the initial CapEx shows up as a cash outflow on the Cash Flow Statement, reduces Cash on the Balance Sheet, and increases Net PP&E.

The Balance Sheet remains in balance because one Asset has decreased by $100, and another has increased by $100:

The company believes that the salvage value of this newly purchased asset is $0, so it won’t be able to sell it for anything at the end of its useful life.

Management also believes that this asset should be useful for about 10 years, so it depreciates the following amount each year in a straight-line pattern:

Depreciation = (Initial CapEx – Salvage Value) / Useful Life

Depreciation = ($100 – $0) / 10 years = $10 per year

Therefore, the company must record $10 in Depreciation for the entire year (i.e., from January 1 to December 31) on its financial statements.

This appears on the Income Statement, reduces Pre-Tax Income by $10, and reduces Net Income by $7.5 at a 25% tax rate.

On the Cash Flow Statement, Net Income is lower by $7.5, the $10 in Depreciation gets added back, the $100 of CapEx is still there, and Cash at the bottom is down by $97.5.

The intuition is that the CapEx reduced Cash by $100, but the Depreciation provided $2.5 in tax savings, so Cash is down by only $97.5:

On the Balance Sheet, Cash is down by $97.5, Net PP&E is up by $90 due to the $100 of CapEx minus the $10 of Depreciation, and so Total Assets are down by $7.5.

On the other side, Common Shareholders’ Equity is also down by $7.5 due to the reduced Net Income, so both sides of the Balance Sheet balance:

The key point here is that the CapEx reduces the company’s cash flows, but not its “Earnings” or Net Income – only the Depreciation after the fact affects that.

So, How Do CapEx & Depreciation Accounting Affect a Company’s Valuation?

If you use a valuation multiple that ignores or adds back Depreciation, such as TEV / EBITDA, the useful life of a company’s capital assets makes no direct impact.

EBITDA literally ignores Depreciation – it’s even in the name! (“Earnings Before Interest, Taxes, Depreciation & Amortization”).

To demonstrate, here’s what the EBITDA multiples look like for two companies that are the same in all ways except for their Depreciation figures:

The TEV / EBITDA multiples are the same, but the P / E multiples differ, and the company with higher Depreciation, indicating a shorter average useful life, “looks” more valuable.

This is why you should focus on metrics and multiples that better normalize for these issues, such as EBITDA and Unlevered FCF.

In a DCF, since Depreciation is deducted to calculate taxes and then added back as a non-cash expense, the exact number matters mostly for the tax savings it provides.

In some sense, the “spread” between Depreciation and CapEx also matters, but it depends heavily on how the model drivers work; the absolute CapEx level still matters most.

It’s difficult to show the exact impact of a longer or shorter useful life, but we can simulate it by changing the Depreciation number.

For example, if CapEx is currently 15% of Revenue, and Depreciation is 12% of Revenue, the output from a simple DCF looks like this:

But if the company extends the average useful life of its assets, so that its Depreciation falls to 10% of Revenue, the implied values from the DCF decrease:

This happens because when Depreciation is lower, the tax benefits are spread out over a longer period, reducing their Present Value.

So… on the surface, it seems like shorter useful lives improve valuations!

There is a caveat, though: These useful-life changes do not occur in a vacuum.

If a company reduces the average useful life of its assets, that tends to imply that it must spend more on CapEx to replace and upgrade these assets more frequently.

If both CapEx and Depreciation change, anything could happen.

For example, here’s the difference between a 15% CapEx / 12% Depreciation scenario and an 18% CapEx / 15% Depreciation scenario:

Michael Burry vs. AI and Big Tech: Why No One is “Right” About CapEx and Depreciation of Nvidia GPUs

As shown above, a company’s estimated useful life for its assets and its CapEx & Depreciation policies do impact the results of a DCF-based valuation.

But in the grand scheme of things, the effect size is quite small; no one buys one company and sells another because of a useful-life difference of 4 vs. 6 years.

The real question is the additional revenue and cash flows generated by all this AI CapEx.

Various people and firms have attempted to estimate this, but no one has a true answer because the Big Tech firms do not break out their results in a useful way.

Yes, many of them have recorded impressive growth rates, but it’s unclear whether that growth has been sustained by AI spending or would have happened without spending ten gazillion dollars on data centers and GPUs.

For a consumer/retail or restaurant business, we could estimate the impact of CapEx by determining the revenue per new store and the expenses associated with it, and weighing those against the construction time and budget:

But nothing like this exists for Big Tech or other AI-centric companies.

Until it does, these debates about the proper Depreciation policy or the useful life of GPUs are distractions from the main issue.

Michael Burry might be right, or he might be wrong, but it doesn’t matter until we have some way of measuring the long-term cash-flow impact of this very high AI CapEx.