Alphabet made more from its investments than from Google. It sold almost none of them.

Alphabet reported pretax income of $138.8 billion for the second quarter of 2026, and net income of $112.2 billion after tax. Operating income for the same three months was $40.8 billion. The difference is mostly a single line in other income: a $99.0 billion net gain on equity securities, which is roughly 2.4 times what the core business earned and about 71 percent of pretax income.

The breakdown is the part worth reading. Raising the value of its stakes in private companies added $77.4 billion, more than the entire business earned in the same three months. Shares it actually sold produced $278 million. The rest is Alphabet marking up investments it still owns.

Where the $99.0 billion came from

The filing breaks the figure into three parts. Unrealized gains on non-marketable equity securities held under the measurement alternative came to $77.4 billion. Unrealized gains on marketable and other equity securities came to $21.4 billion. Realized gains on securities sold during the quarter came to $278 million, which is less than three tenths of one percent of the total.

For scale, the same line was $1.3 billion in the second quarter of 2025. Over the first half of 2026 it reached $135.9 billion, of which $836 million was realized.

What the measurement alternative does

Most of the gain arises from an election in ASC 321. A company holding equity in a business with no readily determinable fair value, meaning a private company, may carry that stake at cost and adjust it only when an observable transaction in the same or a similar security establishes a new price. The election is available when the holder does not have significant influence over the investee, which is the position Alphabet takes for these holdings.

Two consequences follow. The stake is written up when somebody else buys into the company at a higher price, and the investee's own results, profit or loss, are not reflected at all. The number moves on the last price paid by a third party rather than on how the business is performing.

The effect runs one way. If one of these companies loses money, none of that loss reaches Alphabet's income statement. If an investor buys into the same company at a higher price, that gain does reach it, and it is counted as profit. Alphabet reports the price going up. The investee's own losses never appear at all.

Eli Bartov, a professor of accounting at New York University's Stern School of Business, has advised money managers on financial reporting and valuation since 1996 and has served as an expert witness in securities fraud litigation. His 1993 paper in The Accounting Review examined how companies time asset sales to manage reported earnings. Asked what he tells investors to do with a net income figure composed largely of unrealized markups on private stakes, he identified the discretion first.

Unrealized markups may or may not involve significant management discretion. When discretion is involved, as with the measurement of gains or losses on non-marketable securities, there is a risk that management could use that discretion to influence reported gains or losses and achieve its desired reporting outcomes, potentially resulting in misleading information for investors.

He also drew a line between the two kinds of income on the statement. Alphabet, he noted, is a technology company operating in information technology, internet content and online advertising, and it is not a financial institution. On that basis he argued that investment gains, particularly unrealized ones, should be evaluated separately from operating income because they may be neither recurring nor indicative of how the underlying business is performing.

The balance sheet shows the same thing

The carrying value of the non-marketable holdings under the measurement alternative was $124.3 billion at the end of June, against $64.1 billion six months earlier. The filing separates that figure into what was invested and what has been added by revaluation.

Alphabet has put $47.6 billion into these positions. Cumulative upward adjustments stand at $85.7 billion. On those figures, about 69 percent of the carrying value is markup rather than money the company committed, which is our arithmetic from the disclosed totals rather than a figure Alphabet publishes.

The filing adds one more figure. Of the $124.3 billion carrying value, $87.9 billion was remeasured at fair value during the three months ended 30 June, and Alphabet states that it was primarily classified within Level 2 of the fair value hierarchy at the time. Level 2 means the price was not quoted in an active market but was derived from other observable inputs. About seven tenths of the portfolio was repriced in a single quarter on that basis.

Adjustments have run almost entirely in one direction

The same table shows the asymmetry over the first half of the year. Cumulative upward adjustments rose by $41.2 billion between December and June. Cumulative downward adjustments, which include impairments, rose by $295 million over the same period. Upward revaluation therefore ran at roughly 140 times downward revaluation in six months.

Some of that gap is genuine. Private technology valuations did rise over the period. But the accounting itself contributes, because the two directions are not detected the same way. An increase requires an observable transaction at a higher price, and a funding round is announced. A decline in a private company's prospects produces no such event, and a down round is negotiated quietly and disclosed late if at all. Asked whether the write-down arrives as quickly as the write-up, Bartov distinguished between the two kinds of holding.

For non-marketable securities, valuations often rely on models, comparable companies or transactions, and other inputs rather than readily observable market prices. This mark-to-model approach gives management and valuation specialists greater judgment over both the timing and magnitude of changes in reported fair value. As a result, there can be a lag between a deterioration in the underlying investment's economic value and the recognition of a corresponding write-down.

The same asset class, accounted for differently

Microsoft holds a comparable position in OpenAI and reports it under the equity method, which applies when an investor has significant influence over the investee. Under that method the investor recognises its share of the investee's results, including losses. Microsoft disclosed net losses on the OpenAI investment of $1.5 billion in fiscal 2024 and $4.8 billion in fiscal 2025, followed by a net gain of $6.5 billion in fiscal 2026.

The comparison is not that one treatment is right and the other wrong. Both follow the rules that apply to the relationship each company has with its investee. The point is that two of the largest positions in private artificial intelligence companies sit on two balance sheets under methods that behave in opposite ways, and only one of them can report a loss when the investee loses money.

What it means for the earnings figure

The gain is about 71 percent of pretax income. Tax applies to the gain and to the total alike, so the proportion is roughly the same measured after tax, which is where the figure Bartov used comes from. He was asked what happens if investors read the headline number without making that adjustment.

If investors focus primarily on net income without recognizing that approximately 70% of Google's Q2 2026 net income was non-recurring, they could overestimate the company's sustainable earnings power and bid the stock price above its fundamental value. If the market subsequently recognizes this valuation error, the resulting price correction could lead to investment losses.

The argument for the treatment

There is a reasonable case that this is the right accounting, and Bartov made it when asked. GAAP exists to provide a standardized and consistently applied framework, and a framework that applies uniformly across thousands of companies will not perfectly capture the economics of any one of them. Certain rules therefore create differences between reported earnings and the economics investors care about, and that is a known cost of standardization rather than a defect specific to Alphabet.

His conclusion was that the adjustment belongs to the reader. Investors should understand the limitations and, where appropriate, adjust reported GAAP figures to better reflect sustainable economic earnings and underlying performance, then use the adjusted figures alongside the reported ones rather than relying on GAAP net income alone.

What to watch

The next test is a quarter in which private valuations stop rising. The measurement alternative has no mechanism that forces a stake down in the absence of an observable transaction, so the first sign of a turn is more likely to appear as an impairment charge, or as cumulative downward adjustments moving by something other than a rounding error, than as an absence of gains.

The second is the gap between the two figures Alphabet already publishes. Operating income of $40.8 billion describes the advertising and cloud business. Pretax income of $138.8 billion describes that business plus the revaluation of a private portfolio. Both are correct, and they answer different questions.

Alphabet's 10-Q is here (opens in a new tab), and Microsoft's 10-K is here (opens in a new tab).

Cover photograph: Google signage on Charleston Road, Mountain View, by Dietmar Rabich (opens in a new tab), via Wikimedia Commons, licensed CC BY-SA 4.0 (opens in a new tab). Cropped and resized from the original.