Alphabet's Strongest Quarter on Paper Was Also Its First Cash-Flow-Negative One

本文另有中文版：Alphabet 帳面最強的一季，也是現金流第一次轉負的一季

The story already has a settled shape. Alphabet beat on revenue, raised its capital expenditure guidance again, and the stock fell: as much as roughly 5% after hours on July 22, with GOOGL closing the next session down 7.13%. Reporting attributed the move mainly to the guidance raise, secondarily to free cash flow going negative. That account is accurate, and it is the least interesting thing this filing does.

What survives once the price move is explained is harder. Three sentences, all from the same earnings release, all true:

Net income rose 298% and the headline numbers set records. Remove this quarter’s equity-securities gain and the derived figure lands roughly where the market expected. Free cash flow was negative for the first time since the 2004 IPO.

Nobody miscounted. Each of those sentences describes a different moment. One is a mark on what Alphabet’s shareholdings were worth on the last day of June; another tracks money that physically left the building; the third measures what the business itself earned while both of those were going on. Every one of those treatments is standard, and independent of the others. Put them in the same three months and they pull apart.

The part that actually collects money is caught in between. That is where this piece is headed.

One reported EPS, one derived from it

The totals, first. Revenue was $119.8 billion ($119,796M), up 24% year over year and 23% in constant currency; the release calls it the “12th consecutive quarter of double-digit revenue growth.” Net income available to common shareholders was $112.1 billion, up 298%. Diluted earnings per share came in at $9.11, up 294%.

Revenue up 24%, net income up 298%. The gap is not a mystery: the filing explains it. Alphabet recognized a $99.0 billion gain on equity securities ($99,031M). A footnote states that after $21.9 billion of tax, the gain increased net income by $77.1 billion and diluted EPS by $6.26.

Subtract: $9.11 minus $6.26 is $2.85.

So the quarter yields two per-share numbers, but they are not the same kind of thing. $9.11 is the GAAP diluted EPS printed on page one, the figure Alphabet formally reports. $2.85 is what you get after removing this one gain: the arithmetic uses the company’s own disclosure and is sound, but it is an outside-derived measure, not a second EPS the company reports. It also strips out only this item; other non-recurring items remain inside it, so it is not a clean read on the operating business either.

This is where a lot of commentary goes wrong, so be precise about it. Under current US accounting rules, equity investments are remeasured at fair value at period end and the change flows through the income statement. Recognizing the $99.0 billion was not a choice Alphabet made; it is what the rules require, the treatment is fully compliant, and there is no discretion here to be suspicious of. What matters is the character of the number. No cash came in, and not a single share was sold. The line records how much more those holdings were worth on June 30 than on March 31.

Character cuts both ways. The mechanism that produced a $99.0 billion gain on markups produces losses of the same kind on markdowns. This is a line that swings, not a win banked.

On that basis, Alphabet’s profit was roughly in line with what the market expected, and not as spectacular as the headline implies. One caution on that comparison: firms differ in what they strip out, and some remove only the non-marketable portion, which yields a different “adjusted” EPS. When you meet a version you don’t recognize, ask what it removed before asking how big it is.

Why this particular markup is hard to price

There is nothing exotic about a gain on equity holdings. What makes this one worth a second look is how little of it came from anywhere but a single position, and what that position happens to spend its own money on.

Most of the $99.0 billion traces to one company. Alphabet holds roughly 14% of Anthropic, carried at about $124.3 billion as of June 30, with an unrealized gain of roughly $77 billion in the quarter. Anthropic’s valuation was marked up sharply in a recent funding round, and Alphabet’s carrying value moved with it. (Anthropic’s unusual governance structure, which writes the mission into the share structure itself, is something we took apart separately: Ben Bernanke’s seat on Anthropic’s Long-Term Benefit Trust.)

Anthropic is also a large Google Cloud customer. The two announced in October 2025 that Anthropic would use up to one million TPUs and more than a gigawatt of capacity in 2026, in a deal described as worth tens of billions of dollars — up to roughly $40 billion. We’ve already worked through the design logic of arrangements where capital and compute circulate between the same two parties, in the OpenAI–Nvidia–AMD deal structures, and there’s no reason to repeat it. The question here is the next one down: once a structure like that starts generating accounting profit, what can the income statement still tell you?

Less than it looks like. With a gain on a listed holding, you can pull up the quote and form a view on whether the price is sensible. A private holding doesn’t offer that. The valuation comes from the most recent round, and the details that determine what that price means — who participated, on what terms, with what liquidation preference — are not fully visible from outside. When the company being marked up is also committing to buy from the company doing the marking, the assessment gets harder rather than easier. None of that is an accusation. Pricing a funding round and signing a supply contract are separate commercial acts, each standing on its own, and the accounting is beyond reproach. The narrow claim is that judging this gain’s quality from outside is harder than judging a gain on a listed stock.

The second source is SpaceX. Google took part in a $1 billion round in January 2015, roughly $900 million of it Google’s own. SpaceX listed on June 12 of this year, making this the first quarter Alphabet has marked the position to a public price: $94.1 billion as of June 30, split in the 10-Q between $80.0 billion under short-term sale restrictions and $14.1 billion restricted into the third quarter of 2027. Eleven and a half years, roughly 105 times; we’ve looked at how that valuation is built. Keep the as-of date attached, though — a newly listed stock moves, and $94.1 billion is a June 30 snapshot rather than a settled value. On this quarter’s contribution, Anthropic is the principal driver and SpaceX the secondary one.

The spending that hasn’t reached the income statement yet

The cash flow statement runs the other way.

Capital expenditure was $44.9 billion ($44,924M), exactly double the $22.4 billion of a year ago and 26% above last quarter’s $35.7 billion. Full-year guidance was raised the same day for the second time this year, from $180–190 billion to $195–205 billion, with the CFO saying capex will increase significantly again in 2027. Where it goes is disclosed rather than inferred: management put it at roughly 60% servers and 40% data centers and networking equipment.

Operating cash flow was $39.1 billion. Subtract capex and free cash flow is negative $5.9 billion ($(5,855)M), the first negative quarter since Alphabet went public in 2004.

Alphabet also raised money in the same quarter: $49.6 billion net through equity, including mandatory convertible preferred, plus $20.3 billion net through senior unsecured notes, for $69.9 billion in total. Worth pausing here. A cash flow statement does not assign a particular financing to a particular outlay, so the two facts sitting side by side do not establish that the second was raised to cover the first; corporate financing can also serve liquidity, refinancing of existing debt, or capital-structure decisions. What can be said is narrower: negative free cash flow and large external fundraising occurred in the same quarter at a company that has long funded both investment and buybacks from operating cash flow, and that combination has been rare. Whether it is a timing artifact or a new normal is a question for the next several quarters.

Accounting for capex runs in the opposite direction from accounting for the gain. The $44.9 billion left the building this quarter, but it does not hit this quarter’s income statement as a lump. It is capitalized first, and once assets qualify and are placed in service, the cost returns as depreciation across each asset class’s useful life. Alphabet’s 10-K sets out the schedule: servers and network equipment generally over six years, data center and office buildings over seven to 40 years. So only the portion already in service begins depreciating now, and the rest is spread over a long and uneven horizon. The mirror of that: this quarter’s income statement already carries depreciation, and it comes from capital committed in earlier years, not from this quarter’s spending.

What eventually pays for all of it is the open question. J.P. Morgan estimates that a 10% internal rate of return on AI infrastructure investment through 2030 would require roughly $650 billion a year in new, sustained revenue — about 0.58% of global GDP, or every iPhone user paying an extra $34.72 a month. For scale, pure-play AI companies currently generate somewhere in the range of $25–35 billion a year combined. That gap doesn’t mean the investment is destined to fail; most AI-driven revenue is not collected by pure-play AI companies. It does define what the argument is about: not whether demand exists, but how long it takes, and at what scale, to catch up with the capital already committed.

The business in the middle

Put both of those aside for a moment. What is actually being collected right now?

Google Cloud revenue was $24.8 billion ($24,768M), up 82%. Operating income went from $2.8 billion a year ago to $8.8 billion, lifting the segment margin from 20.7% to 35.6%. For a business long described as large but unprofitable, that margin is now in software-company territory.

Backlog reached $514 billion, up more than $50 billion in the quarter. On the call, management said it expects to recognize just over 50% of it as revenue within 24 months. Read that carefully: it is contracted and undelivered, not banked, and the remaining half converts over a longer horizon. It shows demand is real. It doesn’t show the revenue has arrived.

Advertising has not stalled either. Google Services revenue was $94.5 billion ($94,540M), up 15%, with Search and other at $63.3 billion, up 17%, and YouTube ads at $11.1 billion, up 13%. Group operating income was $40.8 billion ($40,770M), up 30%, and the operating margin moved from 32% to 34%. Usage kept expanding too: Gemini is processing 22 billion tokens per minute, up from 16 billion in the prior quarter, the Gemini app has 950 million monthly active users, more than 9 million developers are building on it, and Gemini Enterprise is now in nearly 90% of the Fortune 100.

Two numbers, finally, worth putting next to each other. Capex went from $22.4 billion a year ago to $44.9 billion, exactly double. Google Cloud revenue grew 82% over the same span.

Those two growth rates cannot be compared directly. Group-wide capital spending also supports Search, YouTube, DeepMind and internal model training, so it is not spent on Cloud alone; Cloud revenue is revenue, not a return of the capital; and spending committed today shows up on both the revenue and cost sides only over many subsequent quarters, so the timing does not line up either. Divide this quarter’s capex by this quarter’s Cloud revenue and you get about 1.81, which is a comparison of scale in exactly the same way — not unit economics, and not a measure of how fast anything is being recovered.

Answering how investment relates to return would take, at minimum, capex, placed-in-service assets and depreciation allocated to the same business scope across multiple quarters, set against that same scope’s revenue, gross margin or cash flow — and if the example is Cloud, it would first take knowing which assets and which depreciation can reasonably be attributed to Cloud. Public filings do not break those items out on that same business-scope basis. This quarter supplies a few points. It does not draw the line.

What to track instead of EPS

Say plainly what this is not. Nothing in the quarter is improper or even unusual. Fair-value recognition is required by the accounting rules, and capitalizing capex then depreciating it is equally standard. None of this is a new accounting phenomenon either. What makes this quarter worth stopping on is magnitude: a $99.0 billion fair-value swing is more than double the quarter’s operating income. When a non-cash line that moves in both directions gets that large, readers anchored on any single measure come away with very different impressions of the same company.

So the number to follow is not this quarter’s EPS.

Start with the sign on next quarter’s equity securities line, which by itself decides whether $9.11-scale earnings were a run rate or an event. Then free cash flow: whether it stays negative, and whether raising money outside the business becomes routine. One negative quarter can be timing. A run of them redefines the funding model.

The Cloud backlog matters less as a headline than as a conversion rate. If “just over 50% within 24 months” gets revised, the direction of that revision will say more than the headline backlog figure.

Operating margin is the slow one to watch. As newly placed-in-service assets push depreciation higher, the margin that emerges then — not this quarter’s — is the scorecard for this round of investment.

And there is the guidance itself, which has already moved twice this year. A third raise would signal demand. It would also lift the bar for what eventually has to come back.

The judgment in front of investors is not whether this was a good quarter. On the operating growth disclosed, it was a strong one. It’s whether a statement whose largest figures describe different moments can still answer, directly, the question you brought to it.

Frequently asked questions

Q: Is Alphabet’s Q2 EPS $9.11 or $2.85?

Both numbers can be computed, but they are not the same kind of thing. $9.11 is the GAAP diluted EPS on page one of the release, the figure Alphabet formally reports; under current US accounting rules, fair-value changes on equity holdings must run through the income statement. $2.85 is what’s left after removing the quarter’s $99.0 billion gain on equity securities, which the release says added $6.26 to diluted EPS after $21.9 billion of tax. The subtraction uses Alphabet’s own disclosure and is sound, but the result is an outside-derived measure rather than a second EPS the company reports, and it isn’t a clean read on the operating business: it removes only this item, leaving other non-recurring items inside. Firms differ in what they strip out when publishing an adjusted figure, so other versions circulate — ask what a given number removed before asking how big it is.

Q: Is the $99 billion gain real money?

It is a real accounting gain, but no cash came in and no shares were sold. It reflects the change in the June 30 carrying value of equity Alphabet holds, driven mostly by a large markup on its Anthropic stake. Recognition is required under current US accounting rules and the treatment is fully compliant. The same mechanism runs in reverse: if those valuations are marked down next quarter, the same line reappears with the opposite sign.

Q: Does negative free cash flow mean Alphabet is short of money?

No. Operating cash flow was $39.1 billion against $44.9 billion of capital expenditure, leaving free cash flow at negative $5.9 billion — the first negative quarter since the 2004 IPO. Alphabet also raised $69.9 billion net through equity and senior unsecured notes in the same quarter. A cash flow statement does not assign a particular financing to a particular outlay, so the two facts together do not establish that the money was raised to cover the shortfall, and one quarter is not enough to conclude the funding model has changed. What’s worth tracking is whether the combination persists.

Q: Can Cloud’s $514 billion backlog be treated as future revenue?

Not directly. Backlog is contracted work not yet recognized, and it grew by more than $50 billion in the quarter. Management said on the call that it expects to recognize just over 50% of it as revenue within 24 months. The rest converts over a longer horizon, and contract terms and customer circumstances can change. It is evidence of demand, not revenue in hand.