Cloud Doubles to $30.7B While Search Slows to 12%: The Sell-Off Prices the Cash Burn, Not the Franchise

Results vs. Consensus

Q3 2026 Scorecard

Year-over-Year Comparison

Quarter-over-Quarter Comparison

Revenue Assessment

Revenue of $127.6B grew 24.7% reported and 25% in constant currency; FX turned from a one-point tailwind in Q2 to a one-point headwind, which the CFO had flagged in July. Under that headline sit two different businesses. Cloud grew 102.6% to $30.7B and accounted for 62% of the YoY gain (48% in Q2), helped by TPU system deliveries into customer data centers, core GCP and AI infrastructure. Google Services grew 10.6%, its slowest growth since Q1 2025: Search & other +12.0%, YouTube ads +9.2%, Network +1.4% and subscriptions, platforms and devices +11.0%. Every Services line except Network decelerated from Q2.

The geographic table shows the same split. United States revenue grew 39% (it grew 32% in Q2), which matches TPU systems and AI infrastructure landing with US customers first, while EMEA grew 10% in constant currency (12% in Q2) and APAC 15%. Search & other is now 49.7% of Alphabet's revenue, down from 52.8% in Q2. Consolidated growth is more concentrated in one segment, one geography and, in part, one product type (hardware) than at any point in our coverage. Against consensus, revenue missed by 0.3-0.5%: Cloud beat by $1.1-1.6B while Services missed by $2.1B. Search was $1.6B short of the ~$65.0B (+14.9%) the Street expected, its largest shortfall of the AI Mode era after a $0.1B miss in Q2.

Margin Assessment

Gross margin of 59.7% was flat YoY but fell 195bp sequentially from 61.7%, and the cost lines explain why. TAC grew 10%, in line with advertising revenue, so the pressure is entirely in other cost of revenues (+32% to $35.1B): TPU system hardware, depreciation (+49% YoY to $8.4B, +18% QoQ), content and data-center operations. Below gross profit, R&D grew 36% on AI talent and depreciation, sales and marketing 22% on Gemini promotion, and G&A fell 17% only because it laps last year's EC fine. The CFO noted "current year legal charges" inside G&A without a size.

Segment margins diverged. Cloud margin rose to 36.2% despite the hardware mix because the rented bridge capacity "came online only at the very end of Q3"; the Q4 print, not this one, will test our 30-33% bridge-period range. Services margin fell 110bp QoQ to 40.7%, and Alphabet-level activities, which hold the shared Gemini research cost, lost $7.7B (+71% YoY, +32% QoQ). The consolidated result: operating income flat sequentially on $7.8B more revenue. We read that as the cost of running a frontier-model program and a hardware business in the same P&L, not as a one-quarter anomaly, and we lower our margin path accordingly.

EPS & Cash Flow Assessment

GAAP EPS of $2.12 fell 26% YoY. The release quantifies the main distortion: the $8.7B loss on equity securities, which the CFO attributed to SpaceX's share-price decline in the quarter, reduced net income by $6.7B and EPS by $0.55. Adding it back gives $2.67, against $2.12 reported. Q3 2025's $2.87 carried a $10.7B equity gain and the $3.5B EC fine in the other direction, so the YoY comparison is noisy in both periods. Interest expense of $967M (from $143M) and $298M of preferred dividends are now permanent items between operating income and EPS. On that basis, which is how the main consensus providers frame the estimate, EPS missed the ~$3.02 consensus by $0.35 (12%). The shortfall traces to operating income, which missed by about $2.6B (6%), not to items below the line.

The cash flow statement is weaker than the income statement. Operating cash flow fell 12.6% to $42.3B, carrying a $5.2B payment for the 2018 EC fine and a $3.2B inventory build for TPU systems. Capex of $51.4B more than doubled, and free cash flow was ($9.1B), the second negative quarter in a row; trailing twelve-month FCF is $19.7B, down from $53.3B a quarter ago. The quarter was funded externally: $12.4B of common stock through the ATM (12M Class A and 24M Class C shares), $28.7B of senior notes, and no repurchases. Cash and marketable securities rose to $265.1B, of which $90.4B is marketable equity, against $125.1B of long-term debt. The balance sheet now passes $1.0T in total assets and carries $924B of purchase commitments that the CFO described as maximum contractual amounts with some flexibility.

Segment Performance

Segment Mix: Q3 2026

Geographic Mix: Q3 2026

Google Search & Other: $63.35B, +12.0% YoY; flat sequentially

Search grew 12.0%, the bottom of the +12-14% band we set in July and one point above the floor that would have reopened the cannibalization case. Retail made the largest contribution and finance the next. Sequentially, Search was flat ($63.27B to $63.35B); in the same quarter last year it grew 4.4% QoQ. Management named four causes for the 4.8-point deceleration from Q2: FX, which the Chief Business Officer called "the single largest driver"; the lap of the acceleration that began in Q3 2025, which the CFO sized at about three points; APAC retailers hurt by the EU's removal of its de minimis exemption, together with World Cup ad budgets that landed in June rather than July; and the growth of AI Mode queries, which monetize at a lower rate than classic Search.

"Our AI investments are driving increased usage and overall query growth, including commercial queries. In fact, our year-over-year query growth rate is the highest it's been in the last 5 years." — Sundar Pichai, CEO

The two-year view helps. Search grew 14.5% in Q3 2025 and 12.0% now, a two-year stack of 26.5%, down from 28.5% in Q2 and 29.0% in Q1. The two-point stack decline roughly equals the two-point consolidated FX swing between Q2 and Q3, so in constant currency the stack was close to flat. That supports the mechanical reading. The AI Mode remark does not fit it: asked directly in Q&A, management confirmed that the new surface earns less per query than the old one.

Assessment: The pillar held on the number and weakened on the mechanism. Our July thesis said AI surfaces monetize "at parity"; management now says they do not yet, and the ad rollout to international AI Mode and multiturn AI Mode in the US is the attempt to close that gap. With Q4 guided to lap "a couple of points" more plus a larger FX drag, we model Q4 Search at +9-10%. We move Bull-1 to At Risk until the AI Mode ad load shows up in revenue per query.

YouTube Ads: $11.21B, +9.2% YoY

YouTube advertising slowed to +9.2% from +12.9% in the World Cup quarter. Brand advertising made the largest contribution and connected-TV viewing again stood out: the living room now accounts for more than a third of YouTube ads revenue. Direct response kept growing through Demand Gen among small and mid-sized advertisers, and Amazon joining YouTube Shopping lets US creators tag Amazon products in videos, Shorts and live streams, a commerce link that should lift conversion-based ad pricing over time.

"YouTube advertising revenues increased 9% to $11.2 billion, with brand driving the largest contribution and notable strength in living room." — Anat Ashkenazi, CFO

Assessment: Part of the deceleration is calendar: June's World Cup pulled brand budgets forward and the CBO said media and entertainment spend "normalized" after Q2. The rest looks like a maturing ad line with subscriptions taking a larger share of viewer revenue. We model YouTube ads at +8-10% into 2027, below the +12.5% our model carried for Q3.

Subscriptions, Platforms & Devices: $14.29B, +11.0% YoY

Subscriptions, platforms and devices grew 11.0%, down from 15.2% in Q2, while rising 10.7% sequentially. Growth came from YouTube Music and Premium, which passed 150 million subscribers including trials, and from AI plans inside Google One, which the CFO called Alphabet's fastest-growing subscription product. The line also holds Pixel hardware and Play, neither of which management discussed.

"Subscription, platforms and devices revenues increased 11% this quarter to $14.3 billion, driven by strong growth in YouTube subscriptions, particularly in YouTube Music and Premium and strong demand for AI plans in Google One." — Anat Ashkenazi, CFO

Assessment: The sequential jump says the subscription engine is healthy; the YoY slowdown most likely reflects a tougher devices and Play comparison that management did not explain. Paid consumer AI plans remain the cleanest proof that Gemini can earn money outside advertising, and with the Gemini app at 1 billion users the conversion runway is long. We trim our FY27 growth for the line to +11-12% from +13%.

Google Network: $7.46B, +1.4% YoY

Network grew 1.4%, its first YoY growth in at least seven quarters, after a long slow decline as Alphabet shifted spend toward owned surfaces. Management did not comment on it.

Assessment: Immaterial to the thesis at under 8% of Services revenue. We keep modeling a flat-to-declining line and treat this quarter as noise until it repeats.

Google Cloud: $30.71B, +102.6% YoY; margin 36.2%; backlog $561B

Cloud had the best quarter in its history on revenue, growth, operating income and margin. Revenue grew 102.6% (81.8% in Q2), operating income tripled to $11.1B and margin reached 36.2%. Management named four growth sources: GCP AI infrastructure, enterprise AI solutions, core GCP services, and TPU system sales, which are hardware deliveries into customer and partner data centers. GCP was "the vast majority" of revenue, and growth excluding TPU systems was "again incredibly strong," but when an analyst asked for the ex-TPU rate, the Google Cloud CEO answered with an adjective rather than a number. Customer metrics moved up again: 18 product lines above $1B of annual revenue, nearly 75 customers above 10 trillion Gemini Enterprise tokens in the past year, more than 100 million seats sold for knowledge work, and customers who used Gemini Enterprise in Q3 2025 more than quadrupled their spend.

"Google Cloud's backlog was $561 billion in the third quarter, with the majority of the backlog related to typical GCP contracts. We expect to recognize just over 55% of the total backlog as revenues over the next 24 months." — Anat Ashkenazi, CFO

The backlog math tightened in a useful way. The $47B sequential add was slightly below Q2's $52B, but the 24-month conversion share rose from "just over 50%" to "just over 55%," so roughly $309B now converts by Q3 2028, against roughly $257B a quarter ago. We think much of that increase reflects TPU system contracts, which enter backlog when signed and convert as Alphabet delivers hardware it has already built into inventory.

Assessment: Cloud now resembles two businesses inside one segment: a hyperscale cloud growing far faster than any peer, and an accelerator-systems seller with hardware economics. Both are growing, and the combined margin held at 36.2% before the bridge costs arrive. We raise FY26 Cloud to $109.5-112.5B and FY27 to $175-195B. The risk is quality rather than quantity: the more of the growth that comes from TPU hardware, the less each revenue dollar is worth, and management has chosen not to show the split.

Other Bets: $372M revenue, ($2.05B) operating loss

Other Bets revenue grew 8.1% to $372M and the loss widened to $2.0B (from $1.4B) as Waymo scaled. Waymo has now driven more than 300 million fully autonomous miles, operates in 15 major US cities after opening San Antonio, San Diego and Tampa to the public, and plans expansions into Canada, Germany, Singapore and Spain. Isomorphic Labs is using Gemini 4 Argon in drug-design research.

"Waymo now operates in 15 major U.S. cities and just recently opened up service to the public in San Antonio, San Diego and Tampa with planned expansions into Canada, Germany, Singapore and Spain." — Sundar Pichai, CEO

Assessment: Waymo's first international plan is the strategic news; the segment remains a $8B annual loss run rate that the model treats as a cost of optionality. Nothing this quarter changes our view that Waymo's value will be crystallized by outside capital rather than by consolidated revenue.

Alphabet-Level Activities: ($7.66B), loss up 71% YoY

Alphabet-level activities hold the shared AI research and development that builds Gemini, plus corporate costs and certain legal items. The loss grew to $7.7B from $4.5B a year ago and $5.8B in Q2. This line is where the frontier-model program shows up, and it grew faster than any revenue line except Cloud.

"In Alphabet-level activities, the operating loss was $7.7 billion, driven by shared AI R&D expenses." — Anat Ashkenazi, CFO

Assessment: Gemini 4's frontier position (Topic 9) is what this spending buys, and it is visible in Cloud and in the Gemini app. But at a $31B annual run rate and rising, the shared-AI cost line now consumes roughly 70% of a quarter's Cloud operating income, and we assume it keeps growing faster than revenue through 2027.

Key KPIs

Key Topics & Management Commentary

Overall Management Tone: Management was more expansive on product and models than at any call in our coverage, presenting Gemini 4, the 1-billion-user Gemini app and a doubled Cloud as evidence that the AI investment is already paying. It was markedly less forthcoming on money: the ATM issuance went unmentioned, the TPU margin and ex-TPU Cloud growth questions were declined, and the Search deceleration came with four partial explanations rather than one sized bridge. Relative to July, confidence on demand rose and transparency on economics fell.

1. Search at +12%: FX, the Lap, Two Verticals and an AI Mode Monetization Gap

Search's 4.8-point deceleration was the first question on the call and the main reason for the stock's move. Management's account has four parts. FX turned from a tailwind to a headwind and was, per the CBO, the largest single factor. Q3 began lapping the acceleration of late 2025, which the CFO sized at about three points. Two verticals weakened: APAC retailers selling into Europe lost the de minimis exemption, and World Cup budgets landed in June. Finally, AI Mode query growth weighed on revenue because AI Mode monetizes at a lower rate. The four factors as described do not add up neatly: if lapping alone cost three points and FX was larger still, the stated headwinds exceed the observed deceleration, which would imply better underlying growth than the headline. Management offered no bridge to settle it.

"In Google Services, we will continue to lap the search performance acceleration that began in the third quarter last year, which will pressure revenue growth in Q4." — Anat Ashkenazi, CFO

In Q&A the CFO sized the Q4 effect as a further "couple of points" of lapping, on top of a slightly larger FX headwind falling mainly on Search and YouTube. Q4 2025 Search grew 16.7%, so the comparison is the hardest of the year.

Assessment: We take management at its word on the lap and FX, and we take the AI Mode remark seriously because it concedes the monetization gap that our parity assumption ruled out. A pricing gap on a growing surface is fixable with ad load, which management is now adding; a demand gap would not be. Q4 at +9-10% is our base case, and the thesis needs Q1 2027 to show the AI Mode ad expansion lifting revenue per query.

2. Ads Come to AI Mode Abroad, Plus Multiturn Ads at Home

The most important commercial announcement was the expansion of ads in AI Mode to international markets and the launch of multiturn ads in AI Mode in the US. Management backed the move with test data: users find the ads relevant, sponsored and organic product listings in AI Mode now have similar click-through rates, and most AI Mode queries are incremental and often upper-funnel. Supporting tools moved too: AI Max now lets advertisers test budgets and ROI targets across several campaigns at once, Gemini generated 1.5 billion campaign assets in the quarter (7x YoY), and the largest product-listing-ads update in a decade raised user satisfaction by 19%.

"Today, we're announcing the expansion of ads in AI Mode into international markets alongside multiturn ads and AI Mode in the U.S. As always, our approach is deliberate, prioritizing the user experience and engagement first and then scaling the formats that drive proven ROI for advertisers." — Philipp Schindler, SVP & Chief Business Officer

Assessment: This is the lever that answers Topic 1, and the order of events matters: Google built usage first, admitted the monetization gap second, and turned on the ad load third, the same sequence it followed in the shift to mobile. If similar click-through rates hold as the ad load rises, the gap narrows in 2027. We do not put a number on it yet, but it is the single biggest upside variable in Services.

3. Cloud +103%: How Much Is TPU Hardware?

Cloud revenue grew $15.6B YoY. Management listed GCP AI infrastructure, AI solutions and core GCP as the main drivers and called TPU system sales "a meaningful contributor." Growth excluding TPU systems was described only qualitatively: "again incredibly strong" in the CFO's prepared remarks, and "very, very strong" when an analyst asked for the number in Q&A. The cost line gives a partial clue: other cost of revenues rose $8.6B YoY, and the CFO named TPU system hardware first among the causes. The geographic table gives another: US revenue grew 39%.

"GCP accounted for the vast majority of the revenue generated in Q3. The cloud revenue growth rate, excluding the impact of TPU systems, was again incredibly strong." — Anat Ashkenazi, CFO

Demand indicators outside TPU were strong on their own: Gemini Enterprise users spending 4x more, BigQuery query speed up 40% at half the processing cost, Wiz customers adopting AI features at a 95% rate, and named wins at BNP Paribas, Tata Steel, Deutsche Telekom and Etsy.

Assessment: Our working assumption is that TPU systems contributed roughly $5-8B of Q3 Cloud revenue, which would put ex-TPU Cloud growth at roughly 50-70%. That is still exceptional growth, but it is our estimate, not a disclosure, and the decision not to give the number when asked directly is a disclosure choice we note. The 36.2% segment margin with hardware in the mix argues that TPU systems are not dilutive at the operating line yet.

4. TPU Systems Become a Business: ~100 Customers, Oracle, Crux AI and Meta's TorchTPU

The Google Cloud CEO described a deliberate plan to turn TPUs into a merchant platform: build the silicon (TPU 8t for training, 8i for inference, with up to 2.7x better training performance and 80% better inference price-performance than the prior generation); make the software portable (TorchTPU, built with Meta, runs PyTorch models on TPUs); and widen distribution (Google's cloud, customer data centers, and now Oracle and the Blackstone-backed Crux AI joint venture). Named customers ranged from Anthropic and Apple to Citadel Securities, Jump Trading, Physical Intelligence and Caltech.

"Given continued strong and accelerating demand, we've recently entered into partnerships to deliver TPU systems with both Crux AI and Oracle, with more in the works." — Thomas Kurian, CEO, Google Cloud

The CFO explained the cash cycle: a signed TPU contract enters backlog, Alphabet buys components and builds inventory, and cash returns only on delivery and billing. Inventory rose to $13.3B from $10.0B, and the Google Cloud CEO stated that Alphabet procures components only against signed contracts.

Assessment: Bull-4 is confirmed and ahead of the July plan: revenue is "meaningful" a quarter after it started, the channel now includes a rival cloud, and Meta co-built the software bridge that most reduces switching costs from GPUs. The business absorbs working capital ahead of revenue by design, which is part of why FCF is negative, and it is the part we worry about least because it is contract-backed.

5. Cloud Margin at 36.2%: The Bridge Has Not Hit Yet

In July the CFO said rented third-party capacity would create "modest margin pressure" in Q3. It did not, because the capacity arrived only at quarter-end. The pressure moves to Q4, when usage ramps, and stacks with a rising share of TPU hardware.

"The third-party bridging capacity, which we discussed on our prior call, came online only at the very end of Q3 and therefore, had limited negative impact to our margin in the quarter." — Anat Ashkenazi, CFO

For Q4 the guidance was explicit: "We expect margin headwinds in the near term as we utilize this capacity." No size was given.

Assessment: The July expansion trigger (Cloud margin at or above 31% through the bridge) is half-tested: Q3 passed easily without the bridge, and Q4 is the real test. We keep our 30-33% Q4 range. A Q4 margin below 28% would mean the bridge costs more than management implied, and we add that as a new watch item.

6. Capex at $51.4B, $924B of Commitments, and 2027 Deferred Again

Q3 capex rose 114% to $51.4B, about 65% servers and networking (six-year lives) and 35% data centers. The FY26 guide of $195-205B was left unchanged, which implies $63-73B in Q4. The 2027 figure was again deferred, now to the Q4 call, with the same "increase significantly" language, now on its fourth consecutive call. The new disclosure was the size of forward obligations: $924B of purchase commitments and $152B of leases not yet started, which the CFO described as maximum amounts under agreements with some pricing and volume flexibility.

"At the end of the third quarter, we had future purchase commitments of $924 billion and pre-commencement leases and other obligations of $152 billion. These dollar amounts reflect the maximum contractual obligations. However, a number of these agreements have variable pricing and volume flexibility, which could result in lower spend." — Anat Ashkenazi, CFO

Assessment: The commitments are roughly 1.9x Alphabet's expected FY26 revenue and 1.6x its backlog. They do not all land in one year, and the backlog is the demand cover for much of them, but they define the scale of the bet in a way the annual capex guide does not. An implied Q4 rate of $63-73B already annualizes at $252-292B, so FY27 capex above the $270B level in our published trigger is now our base case ($280-320B). The trigger's second condition, "commensurate backlog growth," is therefore the one that decides it, and we will grade it on the Q4 call.

7. Financing: $12.4B Through the ATM, $28.7B of Notes, Zero Buybacks

The release, not the call, disclosed the quarter's financing. Alphabet sold about 36 million common shares (12M Class A and 24M Class C) through its at-the-market program for $12.4B, leaving $27.6B of the $40B program. It issued $28.7B of senior notes. Repurchases stayed at zero for a third quarter. The board declared the $0.22 common dividend and the 6.25% preferred dividend. In July the CFO tied the ATM to offsetting the tax cost of stock-based compensation and said Alphabet was not planning to return to the equity markets beyond it. Net payments for stock-based award activity in Q3 were $5.4B, so the ATM draw was 2.3x that amount.

"Our sizable investments in technical infrastructure and inventory build of TPU systems as we scale that business will put pressure on free cash flows." — Anat Ashkenazi, CFO

Assessment: Our July downgrade trigger was "any equity issuance beyond the disclosed ATM program," and it did not trip: every share came through the ATM. But the pace (31% of the program in one quarter) and the ratio to award payments say the ATM is now a general funding source, not an SBC tax offset. We move Bear-2 to Materializing. A new equity program, or exhaustion of the $40B ATM before mid-2027, would be a thesis event.

8. The Investment Book Turns: An $8.7B SpaceX Markdown

Other income was ($8.2B), driven by an $8.7B loss on equity securities that the CFO attributed to SpaceX's share-price decline in the quarter. After tax it cost $6.7B of net income and $0.55 of EPS. The book remains very large: $90.4B of marketable equity securities inside cash and securities, and $135.2B of non-marketable securities, up from $131.5B in June.

"Other income and expenses was negative $8.2 billion due primarily to unrealized loss from our investment in SpaceX from a decline in its stock price in the third quarter." — Anat Ashkenazi, CFO

Assessment: This is the quarter Bear-3 described in July: the marks that added $136B in the first half run in reverse. The loss was small relative to the gains, but it shows that a now-public holding transmits its daily price straight into Alphabet's EPS. We model on EPS excluding marks and move Bear-3 to Materializing on the mechanism, while noting that the scale so far is modest.

9. Gemini 4 Argon and Carbon: The Model Gap Becomes a Model Lead

Earlier this year management acknowledged a gap in coding and agentic work and named Gemini 4, then in pretraining, as the response. This quarter it shipped Argon two weeks before the call and announced Carbon on the day, which ranks first on the Artificial Analysis Intelligence Index at 59. Developer usage of Argon ramped 3.5x faster than Gemini 3 in its first week despite capacity limits; inside Google, Argon helped teams ship 35% more code with 30% fewer errors. Carbon is going first to trusted cyber defenders and is being tested by the US government. Genie 4, a real-time world model, was announced the day before.

"This rapid progress underscores our incredible model momentum. We are at the point where AI is helping us increase our training and launch velocity. I would expect the next 12 months to see continued acceleration because of RSI." — Sundar Pichai, CEO

Assessment: Bear-4 (model commoditization) recedes from Contained toward resolved. A frontier lead matters here for three concrete reasons: it pulls developer and enterprise API demand into Google Cloud, it improves the AI Mode and Gemini app products that Search monetization depends on, and it raises Google's own engineering productivity. The CEO's reference to recursive self-improvement is a claim we cannot test, so we give it no weight.

10. Agents: Enterprise First, Then the Gemini App's 1 Billion Users

Management laid out an agent roadmap in sequence: the Gemini agent for enterprise first, because enterprises set the highest bar for security and governance; then Search and the 1-billion-user Gemini app "later this year"; then Chrome, Android and Workspace. On commerce, the Universal Commerce Protocol now supports direct checkout for hundreds of thousands of brands, Universal Cart launched in the quarter, and the Agent Payments Protocol sits underneath. The CEO said Carbon is being distilled into more efficient models so agents can run at billion-user scale.

"Agents will operate as a unifying layer across our products, so you will see us extend these agentic experiences beyond Search in the Gemini app to Chrome, Android and our Workspace products as well." — Sundar Pichai, CEO

Assessment: The agent question is the long-term Search question: if agents complete the purchase, who gets paid for the decision? Management's answer is that agents are additive and that a decision with several good options leaves room for advertising. That is plausible but untested. What is tested is distribution: no competitor can put an agent in front of a billion app users, Chrome and Android at once, and the commerce protocols mean Google writes the rules of the checkout.

11. Operating Leverage: Alphabet-Level AI Cost, Depreciation and Hardware

Consolidated operating income was flat QoQ at $40.6B while revenue grew $7.8B. Three cost streams absorbed it. Hardware costs from TPU system sales entered other cost of revenues. Depreciation rose 18% QoQ and 49% YoY, and the CFO again guided it higher with energy and data-center costs. And R&D grew 36% on AI talent and depreciation, with the shared-AI loss in Alphabet-level activities up 32% QoQ.

"Other cost of revenues was $35.1 billion, up 32%, driven by hardware costs primarily related to the sales of TPU systems, increase in depreciation, content acquisition costs and technical infrastructure operations costs." — Anat Ashkenazi, CFO

Assessment: This is the most important change to our model. In July we expected consolidated margin to plateau; it fell instead, and on a clean basis (excluding last year's fine) it is down 210bp YoY. The CFO's case is that ROIC stays "meaningfully above our cost of capital"; the P&L does not yet show it at the consolidated level. We cut FY27 operating income by about 12% and assume no margin expansion until depreciation growth slows.

Guidance & Outlook

Alphabet does not guide revenue or margins, so the outlook comes as directional statements, and this quarter's were tilted negative for Services and positive for Cloud. The CFO quantified the Search lap for the first time (about three points in Q3, a couple more in Q4) and widened the FX headwind, while the Google Cloud CEO described TPU demand as "accelerating," with more partnerships "in the works." On spending, the FY26 capex range held, which the market should read as a mild positive after four consecutive raises, but the 2027 number moved out to February again.

Implied Q4 setup: Revenue of $139-142B (+22-25% YoY) on Services of roughly $104-105B (+8-10%) and Cloud of $34-37B (+92-109%). Capex of $63-73B with operating cash flow near $45-55B implies another negative FCF quarter of roughly ($8B) to ($28B).

Street at: Before the print the Street expected Q3 revenue near $128.0B, Search near $65.0B (+14.9%) and operating income near $43.2B, so the gap sat in Services and costs, not Cloud. We found no reliable published Q4 consensus to set against our implied Q4 range.

Guidance style: Directional and selectively specific. Management quantified the items that explain weakness (the Search lap, FX) and left unquantified the items that would let investors price the strength (ex-TPU Cloud growth, TPU margins, the AI Mode monetization gap). The capex guide has now held for one quarter after four raises.

Analyst Q&A Highlights

Why Record Query Growth Produced Slower Search Revenue

The opening question went straight at the contradiction in the print: query growth at a five-year high and Search revenue growth down almost five points. It asked whether the gap reflects AI Overviews and AI Mode carrying fewer ads while taking a larger share of queries. Management answered with FX first, which it called the largest factor, then vertical effects in retail and media, and only then the AI Mode point, which it confirmed.

Q: "And then, Philipp, just on query growth, the fastest in 5 years but obviously, some deceleration in Search growth. Can you just help us understand the disconnect there? Is it simply that you haven't been showing as many ads basically in AI Overviews and AI Mode, and that's now becoming a bigger percentage of the overall queries?"
— Doug Anmuth, JPMorgan

A: "So first, there was FX. It was the single largest driver of Search deceleration relative to Q2. Second, there were vertical-specific dynamics. In retail, some of our APAC advertisers were impacted by the EU eliminating its de minimis exemption. In media and entertainment, spend normalized after an unseasonably strong Q2, driven by the FIFA World Cup... it's worth noting that the strong growth in AI Mode queries had an impact because of the lower monetization we were seeing in AI Mode."
— Philipp Schindler, SVP & Chief Business Officer

Assessment: The question's hypothesis was right, and management confirmed it as the third of three factors. Ranking FX first is defensible for the sequential change, but the AI Mode effect is the only one of the three that persists, and it was the one left unsized.

How Much the Search Comparison Still Costs

A follow-up asked the CFO to size the comparison effect she had flagged in July and to describe how it runs into Q4 and Q1. She separated the sequential FX effect from the YoY lap and gave a number for each period of the lap: about three points in Q3 and a couple more in Q4.

Q: "Anat, can you help us better understand the search performance comparison factor that you talked about impacting Search? And how do we think about that as we kind of go into 4Q, 1Q just kind of looking at the compares?"
— Brian Nowak, Morgan Stanley

A: "And then when we look at the year-over-year comparisons, this quarter, we started lapping the acceleration we saw in Search growth in 2025, the back end of 2025. We left about 3 points of growth versus Q3, and then we expect to see that continue in Q4, a couple of points of lapping that growth acceleration that we saw at the end of 2025."
— Anat Ashkenazi, CFO

Assessment: This is the most useful number on the call for modeling Search, and it sets up Q4 below +10%. Read with the CBO's ranking of FX as the largest factor, it also implies the underlying trend was better than the headline, but the two answers were never reconciled into one bridge, and the market priced the headline.

What TPU System Sales Do to Revenue, Margin and Capex Intensity

An analyst framed the TPU business in unit terms (roughly $35B of system cost per gigawatt) and asked how much of that Alphabet books as revenue, what margin flows through, and whether moving capacity onto customers' balance sheets lowers Alphabet's own capex intensity. The CFO declined product-level financials and instead described the cash cycle: contract, backlog, inventory build, delivery, cash.

Q: "So if a gigawatt of TPU full system costs around $35 billion for the compute and all the kind of components around it, what percent of that are we actually recognizing as revenue? And what kind of flow-through does that have in terms of margin profile relative to where the Google Cloud segment currently sits?"
— Ross Sandler, Barclays

A: "First, obviously, we don't provide specific financials at the product level... when we signed an agreement with the customer to sell TPU systems, once we sign that agreement, you'll first see that showing up in our backlog and then we start building inventory... So it's not the traditional kind of CapEx as you're thinking about this, and I think that's the essence of your question, but rather a different cycle of cash flow generation... To date, we've had some system sales recorded as revenue, but expect more of it to come in through 2027."
— Anat Ashkenazi, CFO

Assessment: The cash-cycle explanation is useful and confirms that TPU inventory is working capital, not capex. The margin question went unanswered. The capex-intensity half got a framing answer (TPU inventory runs through working capital, "not the traditional kind of CapEx") but no number, and the Google Cloud CEO later added that placing systems in customer and third-party facilities offsets the need to prebuild facilities.

Whether Selling TPU Systems Is a Good Use of Scarce Capacity

The next exchange pressed the same point from the strategy side: if margins cannot be disclosed, why is selling TPU systems a good use of capacity that Google's own products also need? The Google Cloud CEO again declined margins but gave three reasons: price-performance against competitors, scarcity of accelerated compute, and a wider market from placing systems where customers' data already sits. He also stated that all TPU inventory is tied to committed contracts.

Q: "And then I'll go again at TPU margins. We're wondering, do you think this is a really good business for you? And maybe why? If you can't give us the margins, why you think this is a good way to use all this capacity?"
— Justin Post, Bank of America

A: "On TPU systems and margins, we don't break out margins for individual products or infrastructure components... Given the price performance benefits of TPU systems relative to competitors, growing demand for accelerated compute and the scarcity of accelerated compute, we're very focused on continuing to improve margins as we scale the business... So all of the inventory that -- for TPU systems are associated with committed contracts."
— Thomas Kurian, CEO, Google Cloud

Assessment: "Continuing to improve margins" implies the margins today are below where management wants them, which fits the 32% rise in other cost of revenues. The contract-backed inventory statement is the more valuable disclosure: it removes the speculative-build risk from the $13.3B balance.

Cloud Growth Without TPU Systems, and What Capacity Is Holding Back

An analyst asked for Cloud growth excluding TPU system sales, how much growth capacity constraints are holding back, and whether third-party capacity in Q4 implies acceleration. The answer restated the prepared remarks on mix, described how Gemini use pulls through compute, storage, database and security spending, and confirmed that more capacity would serve more demand, without a number for any part.

Q: "And then either Thomas or Anat, just on kind of Cloud growth, is there anything you could share just around the rate of growth of Google Cloud excluding the TPU system sales? And just how much of that has been kind of constrained by capacity? And as some of that 3P capacity comes online in Q4, is it fair to kind of assume an acceleration of that Google Cloud business?"
— Mark Shmulik, AllianceBernstein

A: "So GCP accounted for the vast majority of revenue generated in Q3, and Cloud revenue growth, excluding the impact of TPU system sales was again very, very strong... And we are always, at any point in time, we could obviously serve a lot more demand with capacity, but we are managing all of these parameters and growth carefully."
— Thomas Kurian, CEO, Google Cloud

Assessment: The direct request for the ex-TPU growth rate received the same qualitative answer as the prepared remarks. Management presumably has the number and chose not to give it. The pull-through description (agents consume compute, storage and database) is the better long-run argument for GCP than any single quarter's rate.

How Compute Gets Allocated When Every Business Wants More

A question asked the CEO to restate priorities for scarce compute across Google, and how Cloud weighs its own customers against external TPU sales. The CEO described a long-range planning process with an ROIC framework for each business, rather than a quarter-to-quarter allocation, with the models as the shared foundation.

Q: "Based on some of the comments you've made recently, Sundar, on compute constraints, can you refresh for us your key priorities for the application of compute as it becomes available across the organization?"
— Eric Sheridan, Goldman Sachs

A: "We always take a long-range planning. And so each of our business areas submit long-range plans with a very disciplined ROIC framework. And so I would broadly think of all of this as a long-range planning process, not a short-term allocation exercise."
— Sundar Pichai, CEO

Assessment: In July management gave an explicit order (frontier models, then Google's products, then external customers). This quarter it gave a process instead. The process answer is reasonable, but the explicit order was more useful for judging whether external TPU sales compete with Search and Cloud for the same chips.

Selling the Platform to Competitors

The final question asked why working with companies that compete with Google in models and cloud (the Oracle partnership and the frontier labs buying TPUs are the obvious examples) is an opportunity rather than a long-term risk. The CEO separated products from platforms and placed TPU systems with Chrome, Android and Cloud as platforms that must work for every customer to stay competitive.

Q: "We heard a willingness to work with some of your competitors or companies that we may view as competitors in models and cloud. So I wonder how you think about the philosophy here about work with people that maybe are competing with you and why this strategy is an opportunity, not a risk longer term that makes sense."
— Michael Nathanson, MoffettNathanson

A: "You've seen us work on our models with other companies, including Apple with whom we compete in many, many areas. Similarly, we view our TPU systems as an important platform we are building. We want to be world-class in how we build it. That means you're putting it against market forces, testing it and competing for people who would have the most demanding workloads on top of it."
— Sundar Pichai, CEO

Assessment: The platform framing is consistent with Google's history and with the TorchTPU decision. It also commits Alphabet to supplying rival model labs with the hardware that keeps them competitive with Gemini, a trade-off we accept because the alternative is ceding that demand to NVIDIA.

What They're NOT Saying

1. The ATM draw: $12.4B of common stock was sold through the ATM, 2.3x the quarter's net stock-award payments, and the prepared remarks and Q&A never mentioned it. In July the CFO tied the ATM to SBC taxes; this quarter's pace fits a funding program instead.
2. Cloud growth excluding TPU systems: Described with adjectives in the prepared remarks and again when asked directly in Q&A. Without it, investors cannot separate the cloud business from the hardware business inside a segment growing 103%.
3. TPU system margins: Declined twice, with the Google Cloud CEO saying the company is focused on "continuing to improve margins." The 32% rise in other cost of revenues is the only clue.
4. How much lower AI Mode monetizes: Management confirmed the gap but gave no ratio, no ad-coverage figure and no timeline for parity. It is the single most important unknown for Search.
5. A sized Search bridge: FX ("the single largest driver"), the three-point lap, two vertical effects and AI Mode were offered separately. They were never added up, and as stated they would exceed the observed deceleration.
6. 2027 capex: "Increase significantly" has now been the formulation for four calls. With $924B of commitments disclosed, a range would have helped more than another deferral.
7. The composition of the $924B: No split by type (servers, energy, capacity leases, third-party compute) or by year, and no indication how much is tied to TPU contracts that customers have already signed.
8. SpaceX on both sides: The quarter's equity loss came from SpaceX, and in July an analyst identified SpaceX as the counterparty for the rented bridge capacity that starts weighing on Cloud margin in Q4. Management has not discussed the size of either exposure.
9. DOJ remedies: Silent for a seventh consecutive quarter. The $5.2B payment of the 2018 EC fine appeared only as a cash-flow item.

Market Reaction

• Pre-print setup (Oct 28 close): $324.89, +3.8% YTD from $313.00 at year-end, against the S&P 500's +13.1%; +21.5% over the trailing twelve months but -5.2% over the trailing 30 days, and 19% below the 52-week closing high of $402.62. The stock had already given back its post-Q2 rebound: it closed at $377.65 on August 4 and at $320.58 on October 27, just above its July 23 reaction-day close of $317.69.
• After-hours (Oct 28): The stock fell about 8% on the release and extended to roughly 10% lower as the call progressed, with the after-hours low near 11% down. Premarket on October 29 it traded about 10-11% lower. The pre-print options market had priced a move of roughly 5-6%, so the reaction ran about twice the implied move.
• Reaction session (Oct 29, in progress at publication): Opened at $286.77, down 11.7%, and traded between $280.36 and $288.33; at 1:01 PM ET the stock was $284.22, down 12.5%, on 85.6M shares, already 3.2x the 30-day average daily volume of 26.6M. The session had not closed at publication.
• Peers and market at 1:01 PM ET: Microsoft -1.2%, Amazon -3.3%, Meta +3.5%; the S&P 500 -0.3% and the Nasdaq Composite -1.3%.

A Search-and-cash repricing, not a Cloud repricing. The order of the sell-off tells you what was repriced. The release alone, with revenue a hair light and Cloud ahead, took the stock down about 8%; the call added the AI Mode monetization admission, the Q4 Search lap and the margin headwinds from rented capacity, and the decline deepened. Microsoft reported the same evening with positive free cash flow and rose after hours, which made Alphabet's ($9.1B) look like a company-specific choice rather than an industry condition. Amazon, which reports tonight, is down 3.3% in sympathy on the cash-burn read-across. At 12.5% intraday, the move is larger than any GOOGL close-to-close decline since at least December 2006, including the -9.5% and -9.1% earnings reactions of October 2023 and October 2022; the session had not closed at publication.

The setup made it worse. GOOGL entered the print up less than 4% for the year while the S&P 500 was up 13%, so this was not a crowded long unwinding; it was a stock already discounting the financing debate that then heard three new reasons to extend it: a Search deceleration with a monetization admission attached, an operating-income line that stopped growing sequentially, and an equity draw nobody mentioned. At $284 the stock sits 9% below where it began 2026 and 10.5% below the July reaction-day close, while Cloud revenue is 24% higher than in the quarter that produced that close.

What the price now implies. At roughly $284, Alphabet's equity value is about $3.54T on our estimate of ~12.45B fully diluted shares, and enterprise value is about $3.40T after $140B of net cash (ascribing nothing to the $135.2B non-marketable book). That is near 18x our reduced FY27 operating income, about the multiple the stock carried after the July sell-off, now applied to lower estimates. We think the sell-off prices the next two quarters of Search and FCF correctly and prices the 2027 Cloud ramp too low.

Street Perspective

Debate: Is Search slowing on comparisons and FX, or because AI Mode earns less?

Bull view: The deceleration is mostly arithmetic: an FX swing, the hardest comparison of the year and softness in two verticals, all against the fastest query growth in five years. AI Mode queries are largely incremental, ads on that surface have only just begun to scale internationally and in multiturn formats, and click-through rates already match organic listings. On this view, AI Mode monetization is a 2027 catch-up, not a structural loss, and the AI Mode effect is temporary.

Bear view: Search missed by its widest margin of the AI era and decelerated on a reported two-year-stack basis. Classic link-based queries are growing more slowly than total queries, and management has now conceded that AI Mode monetizes at a lower rate, so every point of query share that moves to AI surfaces dilutes revenue per query. A more agentic 2027 could widen the gap, and the bear camp expects an air pocket over the next one to two quarters before the ad load closes it.

Our take: Both are true, in different proportions over time. The lap and FX explain most of Q3 and Q4, and the two-year stack in constant currency held roughly flat. The AI Mode gap explains less today and matters more each quarter that AI Mode's share of queries rises. Management is now adding ad load to close it, so 2027 is the test year. We hold Bull-1 at At Risk with a dated signpost rather than abandon it.

Debate: Is the cash burn an investment cycle or a structural break in Alphabet's economics?

Bull view: The negative free cash flow is a contract-backed buildout, not a leak. TPU inventory is built only against signed contracts and turns into cash on delivery, more than half of a $561B backlog converts within two years, and the stock trades near 18x next year's operating income on an enterprise-value basis for a company growing revenue 25%. Financing a supply-constrained build with debt and modest equity is rational when the alternative is turning customers away.

Bear view: Two consecutive negative FCF quarters, $12.4B of ATM equity, long-term debt past $125B and a third quarter without buybacks describe a company whose per-share returns are falling even as revenue rises. The bear camp already models 2027 capex of $300B or more, the disclosed commitments exceed $1 trillion including leases, and Microsoft reported positive free cash flow the same evening. Until the cash cycle turns, the stock has no valuation floor that buybacks can defend.

Our take: The bears have the better of the argument on the next 12 months and the bulls on the next 36. Negative FCF, ATM issuance and rising debt are now the operating model through 2027, and the commitments mean the spending is largely fixed. The investment case therefore rests on the $561B backlog converting on schedule, which so far it has. We have cut our estimates and our fair value for the cost of capital; we have not changed our view that the assets being built are earning.

Debate: How much of Cloud's +103% is durable cloud revenue versus one-time hardware?

Bull view: Cloud beat every published revenue estimate, beat on operating income by about 13-14%, and lifted margin to 36% with hardware in the mix. TPU demand is broadening from frontier labs to trading firms, universities and a rival cloud, and the more optimistic camp argues Cloud revenue could exceed Search and YouTube ads combined within a few years. The backlog makes 2027 the year the TPU wave lands.

Bear view: Cloud's growth trailed the most optimistic buy-side hopes, part of it is lumpy hardware revenue whose margins and durability management declines to disclose, and the inventory build and rented capacity mean the margin peak may be behind the segment for now. Growth bought with low-margin hardware deserves a lower multiple than cloud consumption revenue, and investors cannot yet tell how much of the +103% is which.

Our take: The bear is right that the headline overstates the recurring business, and management's refusal to give the ex-TPU rate invites the discount. But the TPU revenue is contract-backed, recurring at the customer level as generations turn over, and arriving with a 36% segment margin. Even on our assumption of 50-70% ex-TPU growth, Google Cloud is the fastest-growing hyperscaler at scale. We value the segment on revenue and margin together, not on growth alone.

Model Update Needed

Valuation impact: At $284.22 (1:01 PM ET), Alphabet's market value is about $3.54T on our estimate of ~12.45B fully diluted shares (12,294M outstanding plus unvested awards and the mandatory convertible preferred on an as-converted basis). Net of $265.1B of cash and marketable securities against $125.1B of long-term debt, and again ascribing no value to the $135.2B non-marketable book, enterprise value is about $3.40T, or 17.9x the $190B midpoint of our revised FY27 operating income. We apply 22-24x FY27 EV to operating income (narrowed from 22-26x: the top of the old range assumed Search compounding in the mid-teens and a self-limiting ATM, and neither holds this quarter) to $185-195B, which gives $4.07-4.68T of EV and $338-387 per share after adding back net cash. Our 12-month fair-value estimate moves to $340-390 (from $390-430). The midpoint implies about +28% from the intraday price. Most of the cut is estimate, not multiple: the FY27 operating-income midpoint falls 12% and the multiple midpoint falls one turn. A return to $400+ needs two things we can see by the Q1 2027 print: Q4 Cloud margin at or above 31% through the bridge, and Search revenue per query rising as AI Mode ads scale.

Thesis Scorecard Post-Earnings

Trigger check (set July 23): Downgrade triggers: Q3 Search below +11% (not tripped, +12.0%); Cloud below +60% with backlog down QoQ (not tripped, +103% and +$47B); any equity issuance beyond the disclosed ATM program (not tripped, all issuance through the ATM); FY27 capex above $270B without commensurate backlog growth (pending, the Q4 call). Position-expansion triggers: Q3-Q4 Cloud margin at or above 31% through the bridge (Q3 passed at 36.2%, Q4 pending); TPU hardware revenue ramping on schedule (met); the stock below $330 while the operational pillars stay confirmed (price met, but Bull-1 moved to At Risk). We keep all four downgrade triggers as published. For the capex trigger, we state the netting now: we will read "commensurate" as backlog growth in Q4 at least matching Q3's $47B; a guide above $270B with a smaller backlog add fires it.

Overall: The thesis weakened on balance. Three of the five bull pillars confirmed and one (TPU) beat our plan, and the model-commoditization bear receded. Against that, the Search pillar moved to At Risk on management's own monetization remark, and both financing bears moved to Materializing. Operationally this is still the best AI-infrastructure franchise in our coverage; financially it is now an externally funded build with negative free cash flow and falling consolidated margin, and our estimates and fair value reflect that.

Action: Maintaining Outperform, our seventh consecutive call, with fair value cut to $340-390 from $390-430 and conviction cut to 6 from 7. No published downgrade trigger fired, and after the 12.5% sell-off the stock has 20-37% upside to the new band. Position sizing stays at 5-7%: one position-expansion trigger fired (TPU), the Cloud-margin trigger is only half-tested, and the price trigger's pillar condition failed with Bull-1 at At Risk. New downgrade triggers, added alongside the four above: Q4 Search growth below +8% (a two-year stack below 25%, below our +9-10% base case); Q4 Cloud margin below 28% (below our 30-33% bridge range); and a new equity program, or exhaustion of the $40B ATM before mid-2027. If the capex trigger and the Q4 Search trigger both fire, we expect to downgrade to Hold. If Q4 Cloud margin holds at 31% or better and Search lands at +9% or better, we expect to restore conviction.