Alphabet lost about 12% on Thursday, trading near $285 against Wednesday’s $324.89 close, and roughly $490 billion of market value went with it in one session. The headline was diluted EPS of $2.12 against a $3.02 consensus. Most of that gap is SpaceX. Alphabet holds about 551 million SpaceX shares, SpaceX fell from $170.86 to $150.86 over the quarter, and the net loss on equity securities came to $8.7 billion, or $0.55 a share after tax. Add it back and EPS is $2.67. That is still $0.35 short of consensus, so the miss is real, but the mark itself cuts both ways: the same portfolio booked $99 billion of gains in the second quarter, which is why the trailing P/E near 15 makes the stock look cheaper than it is. On forward earnings it trades at about 20 times.
The number that matters is 12%.
Google Search & other revenue grew 12% to $63.3 billion. In the first quarter it grew 19%. In the second it grew about 17%, to $63.27 billion. That is seven points of deceleration in two quarters, and sequentially the line barely moved from Q2 to Q3. A year earlier the same step added about $2.3 billion. Over the same period Pichai said query growth hit its fastest pace in five years. More queries, flat revenue. Revenue per query is falling, and the market priced that, not the SpaceX mark.
Philipp Schindler gave three reasons on the call, in this order. Foreign exchange was the largest. Then vertical effects: APAC retail advertisers hit by the EU’s removal of its de minimis exemption, and media and entertainment spend normalising after a second quarter inflated by the World Cup. Then AI Mode, where growth in queries weighed on revenue because those queries monetise less. The first two reverse on their own. The third compounds. AI Mode queries run three times longer than classic searches, one in five uses voice, images or documents, and most of them are incremental, meaning they would not have happened as ordinary searches at all. Google declined to say how much less they earn. When a company reports its fastest usage growth in five years and its slowest search revenue growth in years in the same quarter, the gap between the two is the price of the transition, and management chose not to size it.
The search moat in its classic form, default distribution plus auction density, is intact on usage. What is in question is price per unit. Schindler’s best counterpoint was that click-through rates in AI Mode are already similar for sponsored and organic product listings, which means an ad placed inside an AI answer gets clicked about as often as the answer itself. Google is now scaling ads in AI Overviews and AI Mode to more users and queries, taking AI Mode ads international and adding multi-turn ads in the US. A three-times-longer query is a richer targeting signal than a four-word one. The open question is ad load, how many paid slots an AI answer can carry before it stops being trusted, and nobody outside Mountain View has that number yet.
Cloud carried the quarter. Revenue rose 103% to $30.7 billion against a $28.7 billion estimate, and operating income more than tripled to $11.1 billion, a 36% segment margin. Backlog reached $561 billion, up from $514 billion in Q2 and just over $460 billion in Q1. Management expects just over 55% of it to convert to revenue within 24 months, which is roughly $309 billion, or about $154 billion a year, against a current annualised run rate near $123 billion. Eighteen product lines now generate more than $1 billion a year each. Microsoft reported Azure at $33.2 billion the same evening, so Google Cloud is now within striking distance of Azure on quarterly revenue while growing much faster.
This is where the moat has moved. TPU system sales contributed meaningfully to Cloud growth, and TPUs are now going into Oracle’s cloud. Alphabet is the only hyperscaler that designs its own accelerator at scale and also sells it to others as a full system, with compute, storage, networking and software attached. Every TPU workload is a workload where Nvidia’s gross margin stays out of Alphabet’s cost base. The catch is cash timing. CFO Anat Ashkenazi explained that a system deal lands in backlog first, inventory gets built next, and cash arrives last. Inventory rose $10.96 billion in the first nine months. On top of that, supply is tight enough that Google started renting third-party compute at the end of Q3 and will ramp it through Q4, which Ashkenazi flagged as a near-term Cloud margin headwind.
Then the bill. Capex was $51.4 billion, up 114%, and the five-quarter run reads $24 billion, $28 billion, $36 billion, $45 billion, $51 billion. Free cash flow was negative $9.1 billion after negative $5.9 billion in Q2. Buybacks went to zero. The 2026 capex guide stays at $195 to $205 billion and 2027 is guided “significantly” higher, with the street already modelling above $300 billion. Funding comes from the $80 billion equity programme announced in June, including $10 billion from Berkshire Hathaway, $12.4 billion of at-the-market proceeds raised in the quarter, and a debt balance above $100 billion. Alphabet now issues stock to buy servers. For a company that retired shares every quarter for a decade, that changes what the multiple is paying for.
Depreciation is the line that will absorb it. Q3 depreciation was $8.4 billion against capex six times larger. About 65% of the quarter’s technical infrastructure spend went to servers and networking with six-year useful lives, so each quarter at the current pace adds roughly $5.6 billion a year of server depreciation before counting data centres. Four more quarters like this one add something in the region of $22 billion of annual depreciation to a base that is running near $33 billion. The 31.8% consolidated operating margin and the 40.7% Google Services margin were earned before that wave hits the income statement.
The street cut and stayed bullish. Of 21 analyst updates on Thursday, 18 lowered targets. TD Cowen went to $460 from $485, BMO to $425 from $480, Evercore to $435, Truist to $410, Wedbush to $387 from $445, KeyCorp to $385 from $410, BofA and Rosenblatt to $380, Scotiabank to $375 from $450, Cantor and Raymond James to $350, Piper Sandler to $345 and UBS to $332. DA Davidson, the one neutral in the group, went to $300 from $350. Guggenheim held $450. Consensus still sits around $415 to $420, near 45% above the tape, which says the targets have not caught up with the price.
The stock went into the print up about 3% for the year and comes out of it down roughly 9%, about 30% below the $408.61 high of May 18 and only some 6% above the $267.67 52-week low. The earnings reactions this year tell their own story: up 10% after Q1, down 7% after Q2, down 12% after Q3. Each quarter the capex number rose, and each quarter the market paid less for the growth that came with it.
The base case is a range of $270 to $330, with the 52-week low acting as a floor until the Q4 call puts a number on 2027 capex. The bull case is a retest of $380 to $410, which needs Search back to mid-teens growth once FX is stripped out, visible ad load inside AI Mode, and a 2027 capex figure arriving alongside evidence that backlog is converting at the $150 billion-plus annual pace management implied. A recovery in SpaceX would help the reported numbers, at roughly $5.5 billion of pre-tax value for every $10 move in its shares. The bear case is a cohort derating to $215 to $245, about 15 to 17 times forward earnings, if the market starts valuing the hyperscalers as a group on free cash flow rather than on revenue growth, with Alphabet’s negative FCF making it the easiest name in that group to mark down.
One disclosure settles which story is true. The Q3 10-Q will break Search growth into paid clicks and cost per click. Paid clicks grew 13% in each of the first two quarters. If clicks hold near that pace and cost per click turns negative, AI Mode dilution is measurable and FX was the smaller factor. If cost per click holds, Schindler’s ordering was right and 12% is a trough.
