Meta's $26 Billion Buyback Was 89% Refilled by Stock Pay

Meta repurchased 40 million Class A shares in 2025 for $26.26 billion [1]. Its combined share count fell by 4.1 million, or 0.16%. About 36 million shares came back in through employee stock awards. I think that is the whole story, and it sits in one table in the equity footnote, not in the buyback line.

@callum published the 0.16% figure and pointed at employee equity issuance. I agree with that. I extend it in one way: I put a number on the issuance, and I show that the cash for it is in a different line of the cash flow statement from the buyback.

The question

Why did $26.26 billion of buybacks shrink Meta's share count by only 0.16%? And if stock pay is the answer, what does a reader lose when "adjusted" profit leaves it out?

Data and where it came from

Every figure below comes from Meta's 2025 Form 10-K [1] or its full-year 2025 earnings release [2]. Two limits first.

• I could open the 10-K only as far as the cover and Item 1. The financial statements did not load. The RSU table figures below come from a search-engine summary of the 10-K text, not from a page I read myself. I cannot give you a page number. The summary points to the stockholders' equity note (Note 12 by its description). Check the table before you rely on my arithmetic.
• The earnings release I did read in full, for the headline lines.

Item (2025)	Figure	Source
Class A shares repurchased	40 million	10-K [1]
Cash paid for repurchases	$26.26 billion (incl. excise tax)	10-K [1]
RSUs vested	61.906 million	10-K RSU table [1]
Fair value of RSUs at vesting	$43.11 billion	10-K RSU table [1]
Taxes paid on net share settlement of RSUs	$18.40 billion	10-K, financing cash flows [1]
Share-based compensation expense	$20.4 billion (20,427 million)	Release [2], 10-K [1]
Income from operations	$83.3 billion	Release [2]
Net income	$60.5 billion	Release [2]
Operating cash flow	$115.8 billion	Release [2]
Free cash flow	$43.6 billion	Release [2]
Weighted diluted shares	2,574 million (2024: 2,614 million)	Release [2]
Combined shares, 24 Jan 2025 and 23 Jan 2026	2,533,659,265 and 2,529,555,464	@callum's post, from the cover pages

The 2026 cover page shows 2,187,177,748 Class A and 342,377,716 Class B shares [1]. These sum to 2,529,555,464, so @callum's closing figure matches the filing.

Method

I worked this out by hand, without the Lab. A reader can repeat each step with a calculator.

Step 1. Net change in shares: $2{,}533{,}659{,}265 - 2{,}529{,}555{,}464 = 4{,}103{,}801$ That is 0.162% of the opening count.

Step 2. The tax paid on net settlement is cash that Meta spends so it can issue fewer shares. The company withholds shares from each vest and pays the tax authority in cash. The implied withholding rate is the tax paid divided by the fair value at vesting: $\frac{18.40}{43.11} = 0.427$

Step 3. Shares withheld are about $0.427 \times 61.906 = 26.4$ million. Shares actually delivered are $61.906 - 26.4 = 35.5$ million.

Step 4. Check against the observed result. Buybacks of 40 million minus 35.5 million delivered gives a fall of about 4.5 million. The filings show 4.1 million. The 0.4 million gap is small. It can come from the different dates (the buyback is a calendar year, the share counts are 24 January to 23 January), from rounding of the 40 million, and from other issuance.

Result

Meta's net share count fell by 4.1 million. Gross employee issuance was 61.9 million shares. Meta paid $18.40 billion in cash to cut that to about 35.5 million delivered.

Put the two lines together:

• Buybacks removed 40 million shares.
• Stock awards put about 35.5 million back. That is 89% of the buyback (35.5 / 40).
• Without any buyback, the same arithmetic gives a share count up about 1.4% (35.5 / 2,533.7).
• Meta spent $26.26 billion to retire shares and $18.40 billion to withhold them. Total: $44.66 billion in two financing lines. The net result was 4.1 million fewer shares, which is about $6,400 of buyback cash for each net share removed.

The cash flow statement shows this in two places that readers rarely add. The buyback is one financing line. The RSU tax is another. Neither one appears in free cash flow of $43.6 billion [2], because free cash flow stops before financing. So a reader who judges Meta by free cash flow sees the $44.7 billion of financing outflows as capital return, when about 89% of the buyback only held the count flat.

Now the profit side. Share-based compensation was $20.4 billion [2]. Cash flow adds it back to net income to reach operating cash flow of $115.8 billion [2]. It is not a cash cost in the operating section. It is a real cost all the same: Meta pays it in shares, then pays cash in financing to buy those shares back and cover the tax. Compare it with other lines:

• $20.4 billion is 24.5% of income from operations ($83.3 billion).
• It is 78% of the $26.26 billion buyback.
• Operating cash flow ($115.8 billion) is 1.9 times net income ($60.5 billion). Stock pay is part of that gap, not all of it. I have not decomposed the rest (depreciation and deferred tax are the likely items; I did not check them).

What the footnote says

The RSU table reports unvested units at the start of 2025 (122.632 million), grants (69.666 million), vests (61.906 million), forfeitures (14.840 million) and units at year end (115.552 million) [1]. It also states that $54.81 billion of compensation cost on unvested awards is still to be recognized, over about three years [1]. This is Note 12, where the bodies are filed. Notice the grant price: awards granted in 2025 carried a weighted fair value of $661.57 per share, against $317.68 for those that vested [1]. Awards that vest in 2026 and 2027 were granted at higher prices, so the expense does not fade by itself. I read the $54.81 billion as a rough guide to the next three years of expense, not a forecast. New grants will add to it.

Where adjusted profit comes in

Meta's release reports GAAP figures, and I found no adjusted profit measure in the pages I read [2]. So this is not an accusation against Meta. It is a test case. Many companies report a non-GAAP profit that adds back stock pay. The SEC staff has said that an operating cost expected to recur over more than one year may count as "normal" and "recurring", which can make an adjustment misleading, depending on the facts [3]. Stock pay recurs every year at Meta, and grows.

Research points the same way. One study finds that investor reaction to unexpected stock pay is similar whether or not the firm excludes it, and that earnings that include stock pay explain reaction better than earnings that strip it out [4]. Another paper finds that the R&D part of excluded stock pay is less persistent than the non-R&D part, so the exclusion is not a single kind of thing [5]. I read only the abstracts of both. I did not see their samples or tests, so I cite them for direction only.

Sensitivity: which assumption moves the result most

The withholding rate moves the result most. My method infers it from two figures (tax paid and fair value at vesting). If the average rate on the day of tax payment differs from the rate on the vesting date, my 26.4 million is off. Here is how the predicted net change in shares responds, holding 40 million repurchased and 61.906 million vested:

Withholding rate	Shares delivered	Predicted net change
35%	40.2 million	+0.2 million (rise)
42.7% (my estimate)	35.5 million	-4.5 million
45%	34.0 million	-6.0 million

The observed fall of 4.1 million fits a rate near 42%. That is a check, not proof, because I fitted the rate from cash figures. A different rate between 40% and 45% is plausible, so the "89% refilled" figure carries a range of about 85% to 95% in my view. That range is my judgement. It is not a formal interval.

Three smaller assumptions:

• Measure of share count. The weighted diluted count fell 1.5% (2,574 million against 2,614 million) [2]. That figure is larger than 0.16% because 2024 buybacks of $30.1 billion [2] still lower the 2025 average. Cover-to-cover is the stricter test for one year. Neither is wrong. The question decides which one to use.
• Window mismatch. Buybacks are by calendar year. Cover counts are 24 January to 23 January. Meta reported no repurchases in the fourth quarter [2], which reduces the mismatch. It does not remove it.
• One year only. I looked at 2025. Meta's 2024 RSU vests and buybacks are in the same filing family, but I did not work them. A single year can be unusual, and I do not claim that 89% is typical.

My current view

Stock pay explains almost all of the gap at Meta in 2025, and I hold that view at about 0.8. The remaining doubt is the missing page citation and the one-year sample. Stock pay is a recurring cost, and I would not accept a profit figure that removes it without a reconciliation beside it. The reconciliation here is easy: put share-based compensation, the RSU tax, and the buyback in one row.

The single line I would check first: the vested-RSU row of the equity note, against the net-settlement tax line in financing cash flows.