Q2 brings the first revenue from TPU systems delivered to customer data centers, a small amount the ledger puts at by residual, with inventory, primarily TPU system hardware and devices, at . For certain of the agreements the seller backstops data center and power obligations; its data center credit backstops total , not split by agreement, and funding is read as mixed. Cloud revenue rose to and the backlog to ; management names AI solutions and AI infrastructure beside core GCP as important drivers, with no level, rate or ranking for either, so both are left unsized this quarter, Cloud revenue being the ceiling; the Q1 rate and ranking that sized AI solutions are not repeated, a step in the wording rather than the business. Booking Holdings (BKNG, on the ledger) expanded its Cloud commitment.
Capital expenditures were , the vast majority for AI technical infrastructure; the CFO’s bound gives , capital and left out of totals, against a raised 2026 range of to . Free cash flow was negative by , and the company raised equity of with AI infrastructure among the stated uses: the build is now funded beyond operating cash flow. Depreciation rose to and Alphabet-level activities, which the CFO says are driven by shared AI research and development, lost ; the ledger takes the technical infrastructure usage inside that loss out of the shared research size, , since the depreciation and operating cost channels already count it. The build’s cost channels take the exploratory motive: compute goes first to frontier models, Search AI is served with no price change, and Cloud capacity is priced.
Steps from Q1. Channels that open: ads support handled by Gemini agents (the efficiency tell, a share of queries addressed, which as a share by count reads directional and is left unsized), Gemini pitch tools in the sales force, agentic products bringing new small advertisers (a customer count, also unsized), and third-party capacity, already in use at a cost not disclosed and to expand in Q3. TPU systems move from a described to an inferred strength and from exploratory to offensive as revenue begins; capital spending moves from inferred to implied on the CFO’s bound; engineering coding moves from described to directional on one team’s rate. Search and Gemini in the ads systems stay described and unsized, AI being named beside other causes.
Other income was in gains on equity securities, which the 10-Q attributes mainly to SpaceX and an unnamed private company, and the company has of milestone-contingent funding committed to a private company. The sources name no AI lab among the holdings, so these are recorded as context and no equity-mark channel is registered.
Sized channels against the income statement, Q2 2026
9 of 17 channels sized
Each blue mark is one channel's dollars for the quarter; a bar is the range of an estimate. Grey marks are the company's reported lines. The distance between them is the point: how large the AI channel is next to the line it sits in.
Revenuesreported line
$119.80bn
Total costs and expensesreported line
$79.03bn
Cost of revenuesreported line
$45.94bn
Research and developmentreported line
$18.22bn
Sales and marketingreported line
$8.40bn
General and administrativereported line
$6.46bn
Capital expenditures on technical infrastructure for AI: servers, TPUs, GPUs, data centers and networkingspend · implied by management
Depreciation of the AI infrastructure buildspend · our inference
Research and development compensation increase attributed to investment in AI talentspend · our inference
Shared AI research and development in Alphabet-level activities (Google DeepMind and general model development)spend · our inference
Energy and other data center operating costs of the AI buildspend · our inference
TPU system sales: TPU hardware delivered to customer and third-party data centersrevenue in · our inference
Consumer AI plans: Google One subscriptions with access to the most capable Gemini modelsrevenue in · our inference
Gemini-assisted pitch tools in the advertising sales forcerevenue in · our inference
Engineering work done with agentic coding tools (Antigravity, Gemini)cost displaced · our inference
$10mn$100mn$1bn$10bn$100bn$1000bn
AI channel, dollars for the quarter low to high of an estimate reported lineLog scale: each gridline is ten times the one before.
New money and old money, Q2 2026
2 new15 expanded
Each channel is tagged once for whether its money existed before language models, from the company's annual report and call at the start of the period. A bar splits one flow's sized dollars by that tag. The incremental total is the part that would not be there without the models: a new channel counts in full, an expanded one only for what AI added, a relabelled one at zero.
Flow, sized total
Split by novelty
Incremental total
Paid for AI$3.69bn to $8.32bn sized
expanded $3.69bn to $8.32bn
Incremental total $2.83bn to $8.32bnpoint $3.62bn$2.03bn in 1 channel has no traced baseline
Cost displaced by AI$27mn to $410mn sized
expanded $27mn to $410mn
Incremental total $27mn to $410mnpoint $131mn
Revenue arriving through AI$405mn to $5.15bn sized
new $144mn to $2.40bnexpanded $261mn to $2.75bn
Incremental total $160mn to $5.15bnpoint $903mn$790mn in 1 channel has no traced baseline
The sized total counts every channel the company credits to AI, including relabelled money that existed before language models and the ledger's own estimates for it. The incremental total counts relabelled channels at zero. A flow is split by layer where its dollars sit at more than one: end use, compute sold to builders, and hardware. The same dollar can be a buyer's spend, a cloud's revenue and a chipmaker's revenue, so the layers are never added together.
Paid for AI7 channels · $3.69bn to $8.32bn sized · $2.83bn to $8.32bn incremental · 2 not sized
other
Capital expenditures on technical infrastructure for AI: servers, TPUs, GPUs, data centers and networking
28.1% to 35.6% of the quarter’s revenue; capital spending on the build, not added into totals (its depreciation is)
implied by managementdisclosure: bounded· motive: exploratory· before LLMs: expanded
Capital expenditures were , the vast majority on technical infrastructure to support investments in AI (claim c20), and the 2026 range rose to to . Free cash flow was negative by ; an equity raise of is to fund in part capital spending on AI infrastructure, beside notes of , so funding is mixed (claims c23, c24). The size, , applies the CFO’s bound; it is capital, shown and left out of totals. Layer: end-use, the use the filings put first; Google Services carried other costs and expenses, which hold allocated technical infrastructure, of , or after traffic acquisition costs, against for Google Cloud, whose capacity sold to builders is the other use (compute). Part of it is revenue at Nvidia (NVDA).
Evidence: 7 quotes, 11 figures, 2 confounds, 2 from before coverage
Purchases of property and equipment, mostly technical infrastructure, which management says supports the AI opportunities across the company; mostly servers, the rest data centers and networking equipment. Capitalized: it reaches the income statement as depreciation and operating costs, read on their own channels, so this channel is traced and left out of totals. Part of it is accelerator purchases that are revenue at Nvidia (NVDA). Funded from operating cash flow, notes and, from June 2026, an equity raise whose stated use includes AI infrastructure. Layer: end-use, the use the filings put first. The CFO ties the spending to the AI opportunities across the company and the CEO says compute for frontier model development is allocated first; the 10-Q allocates technical infrastructure costs, with their depreciation, to the segments by usage, and Google Services carries several times Google Cloud’s other costs and expenses, also after its traffic acquisition costs are taken out. The other use is the capacity and model access Google Cloud sells to builders, which is compute.
Why this motive
The build serves uses whose tells conflict: compute for frontier model development is allocated first (claim c30), an exploratory tell; AI Mode responses in Search are served at an inference cost with no price change (claim c38), product-defensive; and priced Cloud capacity is sold while demand outpaces supply (claims c25, c16), offensive. The rule takes the least durable, exploratory.
Before LLMs: expanded
Capital spending on technical infrastructure predates the build: in the first quarter of 2024, already tied to AI opportunities, and purchases of property and equipment of in 2024 (an outflow). The baseline is that quarter’s total capital spending, , not an AI share of it, so it is not like for like with the channel’s size; the channel is capital and out of totals, so no total moves on it. The size is the whole of an activity that existed before.
“The significant year-on-year growth in CapEx in recent quarters reflects our confidence in the opportunities offered by AI across our business.”
“With respect to CapEx, our reported CapEx in the first quarter was $12 billion, once again driven overwhelmingly by investment in our technical infrastructure, with the largest component for servers, followed by data centers.”
Capital expenditures expected for 2026, low end of the range · 2026-01-01..2026-12-31
Capital expenditures expected for 2026, high end of the range · 2026-01-01..2026-12-31
Net cash provided by operating activities · 2026-CQ2
Free cash flow shortfall (free cash flow was negative by this amount) · 2026-CQ2
Net proceeds of the June 2026 equity raise (common and mandatory convertible preferred stock) · 2026-CQ2
Net proceeds of senior unsecured notes issued in the quarter · 2026-CQ2
Purchases of property and equipment, change over the prior-year quarter (as a positive outflow) · 2026-CQ2
Google Services other costs and expenses (holds allocated technical infrastructure, traffic acquisition and content costs) · 2026-CQ2
Google Services other costs and expenses less traffic acquisition costs · 2026-CQ2
Google Cloud other costs and expenses (holds allocated technical infrastructure, TPU inventory and other costs) · 2026-CQ2
Reported line it is matched to
Purchases of property and equipment were against a year earlier (outflows shown as negatives).
2026-CQ2: 2025-CQ2:
What else could explain it
line composition: Technical infrastructure also serves Search, YouTube and core Cloud; the CFO attributes it to investments in AI.
other: Part of the spending builds TPU systems for sale, recorded as inventory rather than capital spending, and part is higher supply prices; neither is separated.
Quotes
“CapEx was $44.9 billion in the second quarter, with the vast majority of this spent in technical infrastructure to support our investments in AI.”
“Moving to investments, we are updating our full-year 2026 CapEx guidance range to $195 billion-$205 billion, up from our previous estimate of $180 billion-$190 billion.”
“In June 2026, we issued a combination of Class A stock and Class C stock and mandatory convertible preferred stock for aggregate net proceeds of $49.6 billion, to be used for general corporate purposes, including capital expenditures to scale AI infrastructure and global compute.”
“As we think about our investment needs, as I said, we look at the next year and multiple years out, we first look at how much we can support based on cash from operations. As you've seen in our results today, we continue to generate very healthy, strong cash flow from operations. That's our first source of funding. Then we look at debt, and most recently we did the equity raise.”
“In terms of allocating our TPUs, look, our first priority is making sure we are allocating what we need to compete at the frontier in terms of AGI development.”
Q1 2026bounded · implied by management · exploratory · $28.56bn to $35.70bn
Q2 2026bounded · implied by management · exploratory · $33.67bn to $42.66bn
cost of-revenue
Depreciation of the AI infrastructure build
2.1% to 2.6% of the quarter’s revenue
Incremental total: counts in full.
Matched line moved : the whole line, not this channel.
our inferencedisclosure: direction only· motive: exploratory· before LLMs: expanded
Depreciation rose to , named by the CFO among the drivers of other cost of revenues and research and development (claims c9, c31). The size, , is depreciation above its 2024 quarterly average of times the AI share the CFO's bound on capital expenditures gives, the vast majority for AI technical infrastructure (claim c20), read through the words table; applying it to depreciation is the ledger's step. Layer: end-use, as for the capital spending: Google Services carried other costs and expenses of after traffic acquisition costs against for Google Cloud, on an allocation by usage.
Evidence: 5 quotes, 2 figures, 2 confounds, 1 from before coverage
Depreciation of servers, accelerators and data centers, mostly in other cost of revenues and partly in research and development, allocated to the segments by usage. Management names it as the P&L pressure of the technical infrastructure investment it attributes to AI; the AI share is assumed. Part of it sits in the technical infrastructure usage costs of shared AI research and development; that part is counted here, and an assumed infrastructure share is taken out of the shared research channel so each dollar is counted once. Layer: end-use, as for the capital spending it depreciates: the 10-Q allocates it to the segments by usage and Google Services carries the larger share; the other use is capacity Google Cloud sells to builders (compute).
Why this motive
Carried with the capital spending it depreciates. The build serves uses whose tells conflict: compute for frontier model development is allocated first (claim c30), an exploratory tell; AI Mode responses in Search are served at an inference cost with no price change (claim c38), product-defensive; and priced Cloud capacity is sold while demand outpaces supply (claims c25, c16), offensive. The rule takes the least durable, exploratory.
Before LLMs: expanded
Depreciation of property and equipment was in 2024, a quarter on average, and the anchor call already expected it to rise with technical infrastructure investment. The size is the depreciation above that level, times an assumed AI share. The size is the change AI made, not the whole line.
“Looking ahead, we remain focused on our efforts to moderate the pace of expense growth in order to create capacity for the increases in depreciation and expenses associated with the higher levels of investment in our technical infrastructure.”
Depreciation of property and equipment, prior-year quarter · 2025-CQ2
Reported line it is matched to
Depreciation of property and equipment was against , a rise of .
2026-CQ2: 2025-CQ2:
What else could explain it
line composition: Depreciation covers all property and equipment, offices included.
other: Part of this depreciation sits in the technical infrastructure usage costs of shared AI research and development (claim c44); it is counted here, and the shared research channel takes an assumed infrastructure share out of its size so the dollar is counted once.
Quotes
“In terms of expenses, the significant increase in our investments in technical infrastructure will continue to put pressure on the P&L in the form of higher depreciation expense and related data center operations costs such as energy.”
“Other cost of revenues was $29.8 billion, up 22%, driven by increases in depreciation, inventory costs, primarily from the sales of TPU systems to customers, and content acquisition costs, largely for YouTube.”
“In terms of allocating our TPUs, look, our first priority is making sure we are allocating what we need to compete at the frontier in terms of AGI development.”
Q1 2026direction only · our inference · exploratory · $2.12bn to $2.65bn
Q2 2026direction only · our inference · exploratory · $2.46bn to $3.11bn
operations
Energy and other data center operating costs of the AI build
0.31% to 1.3% of the quarter’s revenue
Incremental total: counts in full.
our inferencedisclosure: direction only· motive: exploratory· before LLMs: expanded
The CFO again names higher data center operating costs such as energy (claim c26), a direction with no level, and the 10-Q lists other technical infrastructure operations costs among the drivers of other cost of revenues, against . The size, , is the ledger’s ratio to the AI depreciation increase, whose AI share comes from the CFO's bound on capital expenditures. Layer: end-use, as for the depreciation.
Evidence: 3 quotes, 2 figures, 1 confound, 2 from before coverage
Energy, network capacity and equipment costs of running the technical infrastructure, which management names beside depreciation as the P&L pressure of the AI investment. No amount is given; the size is the ledger’s, built as a ratio to the depreciation increase. Layer: end-use, as for the depreciation it is built from; the other use is capacity Google Cloud sells to builders (compute).
Why this motive
Carried with the depreciation of the same capacity. The build serves uses whose tells conflict: compute for frontier model development is allocated first (claim c30), an exploratory tell; AI Mode responses in Search are served at an inference cost with no price change (claim c38), product-defensive; and priced Cloud capacity is sold while demand outpaces supply (claims c25, c16), offensive. The rule takes the least durable, exploratory.
Before LLMs: expanded
At the anchor other cost of revenues already held technical infrastructure operating costs, energy among them, with no amount given, and management expected expenses to rise with the infrastructure investment. The size is a share of the cost rise above the anchor level, so it is sized as an increment. The size is the change AI made, not the whole line.
“Looking ahead, we remain focused on our efforts to moderate the pace of expense growth in order to create capacity for the increases in depreciation and expenses associated with the higher levels of investment in our technical infrastructure.”
Other cost of revenues, prior-year quarter · 2025-CQ2
What else could explain it
line composition: Other cost of revenues holds content acquisition, compensation and TPU inventory costs beside running costs; no running-cost amount is given.
Quotes
“In terms of expenses, the significant increase in our investments in technical infrastructure will continue to put pressure on the P&L in the form of higher depreciation expense and related data center operations costs such as energy.”
“In terms of allocating our TPUs, look, our first priority is making sure we are allocating what we need to compete at the frontier in terms of AGI development.”
Q1 2026direction only · our inference · exploratory · $319mn to $1.33bn
Q2 2026direction only · our inference · exploratory · $369mn to $1.56bn
research
Shared AI research and development in Alphabet-level activities (Google DeepMind and general model development)
0.72% to 3% of the quarter’s revenue
Incremental total: counts at zero at the point and in full at the high end, with no traced baseline.
our inferencedisclosure: bounded· motive: exploratory· before LLMs: expanded
The CFO says the Alphabet-level loss of , against , was driven by shared AI research and development (claim c27). The line is a ceiling; the words table reads the filing’s primarily as a share, and the size, , is that share of it after an assumed part for the technical infrastructure usage the depreciation and operating cost channels count is taken out.
Evidence: 4 quotes, 2 figures, 2 confounds, 2 from before coverage
The costs the company does not allocate to segments, which the filings say primarily reflect shared AI research and development: compensation and technical infrastructure usage costs of developing its general AI models, beside corporate initiatives and shared costs. The line is a ceiling; the AI share is assumed. Its technical infrastructure usage costs (depreciation and running costs of the build) are already counted by the depreciation and operating cost channels, so the size takes an assumed infrastructure share out and keeps the rest, mostly compensation. Layer: end-use, the use the filings put first: the general models serve Search, YouTube and the Gemini app in Google Services before Google Cloud, whose model access sold to builders (compute) is the other use; the line is not allocated to segments, so the build’s allocation decides it.
Why this motive
Research framing: frontier model development, now a Gemini 4 pre-training run, gets compute first (claims c29, c30).
Before LLMs: expanded
At the anchor the same line held AI-focused shared research and development, including general AI model development, and was in 2024 (a loss). Only an assumed share of the line is AI research and development, so any baseline taken from it would rest on an untraced share; none is set, and the channel counts as undetermined in the incremental total. The line also widened during 2024, when general AI model teams moved in from Google Services. The size is the whole of an activity that existed before.
“These costs primarily include certain AI-focused shared R&D activities, including development costs of our general AI models; corporate initiatives such as our philanthropic activities; corporate shared costs such as certain finance, human resource, and legal costs, including certain fines and settlements.”
“General AI model development teams previously under Google Research in our Google Services segment are reported within Alphabet-level activities prospectively beginning in the second quarter of 2024.”
line composition: The line also holds corporate initiatives and shared finance, human resources and legal costs.
other: Its technical infrastructure usage costs are counted by the depreciation and operating cost channels; an assumed share for them is taken out of this size (claim c44).
Quotes
“In Alphabet-level activities, the operating loss was $5.8 billion, driven by shared AI R&D expenses.”
“In terms of allocating our TPUs, look, our first priority is making sure we are allocating what we need to compete at the frontier in terms of AGI development.”
Research and development compensation increase attributed to investment in AI talent
0.97% to 2.2% of the quarter’s revenue; overlaps another channel, not added into totals
our inferencedisclosure: direction only· motive: exploratory· before LLMs: expanded
Research and development rose , which the CFO attributes to compensation for AI talent and to depreciation; the 10-Q gives the compensation increase as (claim c31). The size, , is the ledger’s; it overlaps shared AI research and development and is left out of totals.
Evidence: 2 quotes, 3 figures, 1 confound, 1 from before coverage
The year-over-year rise in research and development compensation, which the CFO attributes to investment in AI talent; the 10-Q gives the dollar rise without naming AI. Much of it is paid inside shared AI research and development, so it is read as overlapping that channel and left out of totals.
Why this motive
Carried: hiring in AI with no measured return named (claims c31, c32).
Before LLMs: expanded
At the anchor research and development compensation was already rising for Google DeepMind and Cloud; research and development expenses were in 2024. A budget redirected to AI work is expanded; the size is part of a year-over-year rise, so it is sized as an increment. The size is the change AI made, not the whole line.
“On an adjusted basis, operating expenses were up 5%, reflecting first in R&D, an increase in compensation expense primarily for Google DeepMind and Cloud, and second in sales and marketing, a slight increase year-on-year, reflecting increases in compensation expense primarily for Cloud sales.”
Q1 2026direction only · our inference · exploratory · $1.00bn to $2.25bn
Q2 2026direction only · our inference · exploratory · $1.16bn to $2.61bn
marketing
Advertising and promotion for the Gemini app and AI products
Not sized
The marketing rise is credited to the Gemini app and Search together and nothing separates the AI part, so the channel is not sized (methodology, AI named beside another cause).
described, no sizedisclosure: direction only· motive: exploratory· before LLMs: expanded
Sales and marketing rose to from on investments to support the Gemini app and Search (claim c33); the 10-Q gives the rise in advertising and promotional activities as . The outlook speaks of marketing for AI products (claim c32). Under the joint-cause rule the channel reads described and unsized.
Evidence: 2 quotes, 1 figure, 2 confounds, 1 from before coverage
Marketing the CFO says supports the Gemini app and Search; the 10-Q gives the rise in advertising and promotional activities. Search is named beside the Gemini app and nothing separates the AI part, so the channel is unsized.
Why this motive
Carried: promotion of a largely free assistant with paid tiers, no measured return (claims c33, c32).
Before LLMs: expanded
Advertising and promotion predates the Gemini app: in 2024. The movement is part of the rise in the line over the prior year, so it is sized as an increment. The size is the change AI made, not the whole line.
“For the years ended December 31, 2022, 2023, and 2024, advertising and promotional expenses totaled approximately $9.2 billion, $8.7 billion, and $8.7 billion, respectively.”
Q1 2026direction only · described, no size · exploratory
Q2 2026direction only · described, no size · exploratory
cost of-revenue
Third-party compute capacity bought as a bridge while internal capacity is built
Not sized
The Q2 cost is not disclosed and use is to expand in Q3. No amount, provider, volume or price is given, and the only commitment of the kind in the filing, a short-term lease from the third quarter, is not tied to this capacity, so nothing permits a ballpark: the channel is inscrutable this quarter, not zero.
inscrutabledisclosure: described· motive: offensive· before LLMs: expanded· layer: compute
The channel opens this quarter. The CFO plans to expand the use of third-party capacity in Q3 as a bridge while internal capacity is built, which implies some use already in Q2 at a cost not disclosed, at a cost that pressures Cloud margins, and the CEO says it serves very large Cloud customers (claims c53, c55, c56). Its money may be revenue at another capacity seller on the ledger; the provider is not named.
Evidence: 6 quotes, 1 figure, 1 confound, 1 from before coverage
Capacity from other providers that the company uses to serve Cloud customers while its own capacity is built; the cost of providing capacity sold to builders, so the layer is compute. The CFO plans to expand its use in Q3 2026, which implies some use already in Q2; the Q2 cost is not disclosed, and the cost of the capacity pressures Cloud margins. The provider is not named in the company’s words; an analyst named a deal with SpaceX, which management did not confirm.
Why this motive
Capacity bought to keep serving very large Cloud customers while supply is constrained (claims c53, c56).
Before LLMs: expanded
At the anchor capacity came from owned equipment and lease arrangements, with operating lease commitments of . Buying capacity from others to serve existing cloud demand is expanded, sized as a level with no traceable quarter before AI. The size is the whole of an activity that existed before.
“We make investments in land, buildings, and servers and network equipment through purchases of property and equipment and lease arrangements to provide capacity for the growth of our services and products.”
Non-cancelable commitment of a short-term lease beginning in the third quarter of 2026, approximately · as-of 2026-06-30
What else could explain it
other: An analyst named a third-party compute deal with SpaceX; management did not confirm the provider. A short-term lease with a commitment of about begins in the third quarter of 2026, but the filing does not say it is this capacity.
Quotes
“Given the supply-constrained environment, we plan to expand the use of third-party capacity in Q3 as a bridging strategy while we build out more internal capacity.”
“I think on the bridge deal, the main thing I would say is, look, on the margin, there are very, very large customers of ours on Cloud who we are trying to support them through this extraordinary moment.”
“Additionally, in June 2026, we entered into a short-term lease agreement with a non-cancelable commitment of approximately $5.8 billion, which will commence in the third quarter of 2026.”
“The incremental opportunities they are bringing to us, while a short-term cost over a few months may be very high, in the lifetime of the deal, as we bring more capacity on, is highly ROI positive.”
Cost displaced by AI2 channels · $27mn to $410mn sized · $27mn to $410mn incremental · 1 not sized
engineering · cheap to verify
Engineering work done with agentic coding tools (Antigravity, Gemini)
0.02% to 0.34% of the quarter’s revenue
Incremental total: counts in full.
our inferencedisclosure: direction only· motive: exploratory· before LLMs: expanded
The CEO says Antigravity has accelerated how the company builds, and one Chrome team is on track to deliver faster through model-driven refactoring (claims c47, c48). A single project’s rate makes the state directional. Employees rose to from . The size, , is the ledger’s.
Evidence: 2 quotes, 4 figures, 1 confound, 1 from before coverage
Internal software engineering with the company’s own agentic coding tools, which management says lets engineers build at a faster velocity, and which the CFO names among the ways the company drives efficiency. No saving is measured; one team’s delivery speed-up is given in Q2. Research and development compensation keeps rising, so the size is the cost avoided against what the same output would have cost, the ledger’s own.
Why this motive
One team’s projected speed-up is the only measure and no cost line is shown to move, so the passing-mention reading holds (claims c47, c48).
Before LLMs: expanded
At the anchor streamlining operations through AI was one of several cost work streams, and research and development expenses were in 2024. Putting LLM tools into existing engineering work is expanded, sized as an increment against the prior rate. The size is the change AI made, not the whole line.
“Then the other work streams we've talked to you about in the past, like all of the work around technical infrastructure, which Sundar alluded to, streamlining operations within the company through the use of AI, what we're doing with procurement with our suppliers and vendors, which he also referenced, the work you've seen on real estate optimization.”
“As just one example, a team in Chrome is now on track to accelerate delivery by 8x , compressing a two-year timeline into three months through model-driven refactoring.”
Q2 2026direction only · our inference · exploratory · $27mn to $410mn
customer support · cheap to verify
Ads customer support handled by Gemini-powered agents
Not sized
The quarter's only measure is a share by count of support interactions: Gemini-powered agents autonomously address of support queries (claim c50), with no cost, rate per unit or headcount. A share of queries by count sizes nothing (the methodology rule for counts).
described, no sizedisclosure: direction only· motive: efficiency· before LLMs: expanded
The channel opens this quarter. Gemini-powered agents now autonomously address of ads support queries, freeing staff for complex cases (claim c50). A share of queries by count reads directional and sizes nothing, and no cost is given, so the channel is left unsized.
Evidence: 2 quotes, 2 figures, 1 confound, 2 from before coverage
Advertiser support queries addressed autonomously by Gemini-powered agents, with the support teams moved to complex cases. Management gives the share of support queries the agents address in Q2 2026; no cost is given. Customer service staff sit partly in sales and marketing.
Why this motive
An implemented tool to which the company attributes a measured operating result, the share of support queries addressed autonomously (claim c50).
Before LLMs: expanded
At the anchor sales and marketing already held certain customer service functions, the line was in 2024, and streamlining operations through AI was already a cost work stream. LLM agents in existing support work are expanded, sized as an increment. The size is the change AI made, not the whole line.
“Then the other work streams we've talked to you about in the past, like all of the work around technical infrastructure, which Sundar alluded to, streamlining operations within the company through the use of AI, what we're doing with procurement with our suppliers and vendors, which he also referenced, the work you've seen on real estate optimization.”
Share of ads support queries addressed autonomously by Gemini-powered agents · as-of 2026-07-22
Sales and marketing · 2026-CQ2
What else could explain it
other: Support staff are redeployed to complex cases rather than reduced, so no cost line is shown to fall.
Quotes
“Our ads customer support teams use Gemini-powered agentic solutions that now autonomously address 75% of support queries, freeing them to solve our customers' most complex challenges.”
Revenue arriving through AI8 channels · $405mn to $5.15bn sized · $160mn to $5.15bn incremental · 5 not sized
product revenue
Google Cloud AI solutions: products built on Gemini and other generative models (Vertex AI model access, Gemini Enterprise, agent platform, AI security agents)
Not sized
This quarter's sources name AI solutions as one of several drivers of Cloud growth with no rate or ranking of its own: core GCP, AI solutions and AI infrastructure were all important drivers (claim c2), Cloud was powered by AI infrastructure and AI solutions (claim c1), and the 10-Q says the increase was led by AI solutions, AI infrastructure and core GCP services together (claim c65). The growth rate and the largest-contributor ranking given in Q1 are not repeated. Nothing separates AI solutions' part (the methodology rule for AI named beside another cause, which holds at the upstream companies). The step from Q1 comes from the wording, not from a change at the company. Cloud revenue, , is quoted as the ceiling.
described, no sizedisclosure: direction only· motive: offensive· before LLMs: new· layer: compute
Management names AI solutions with core GCP and AI infrastructure as important drivers of Cloud growth of to , and model APIs now process about tokens a minute (claims c2, c17). The growth rate and ranking given in Q1 are not repeated, so nothing separates AI solutions' part of the rise and the channel is left unsized this quarter; the step from Q1 comes from the wording, not from the business. Cloud revenue is the ceiling. Booking Holdings (BKNG, on the ledger) expanded a multi-year Cloud commitment (claim c66): that money is a ledger company's spend, in Google Cloud revenue and possibly in part in this channel.
Evidence: 11 quotes, 8 figures, 3 confounds, 1 from before coverage
Revenue Google Cloud earns from products built on its generative models: model access and tokens through Vertex AI and the API, Gemini Enterprise seats, the agent platform, and Gemini-powered security agents. Management gives a year-over-year growth rate for revenue from products built on its GenAI models in Q1 2026 and calls AI solutions the largest contributor to Cloud growth; no level is given. It sits inside Google Cloud segment revenue, which is a ceiling and never its size. Some buyers are AI-native developers; most named buyers are enterprises. Layer: compute, the use the filings put first: the CFO credits AI solutions growth to demand for the company’s models, sold to builders as model access through Vertex AI and the API. The other use is applications sold to businesses (Gemini Enterprise, agents, security agents), which is end-use; the filings give no split, so the channel is not divided by layer.
Why this motive
Carried from Q1 with the same tell: separately priced AI products management names among the important drivers of Cloud growth (claims c2, c19).
Before LLMs: new
At the anchor the company already sold generative AI through Vertex AI and AI Studio, used by more than developers, and Gemini for Google Cloud and Workspace; Google Cloud revenue was in the first quarter of 2024 and no AI solutions revenue was given. Products built on generative models could not exist without LLMs, so the money is new.
“And we shared that more than 1 million developers are now using our generative AI across tools, including AI Studio and Vertex AI.”
Tokens per minute processed by the model APIs, approximately · as-of 2026-07-22
Cloud customers that each processed more than a trillion tokens in the past year, approximately · TTM as-of 2026-06-30
Google Cloud remaining performance obligations (backlog) · as-of 2026-06-30
Google Cloud backlog, change over the prior quarter end · 2026-CQ2
Google Cloud revenues · 2026-CQ2
Google Cloud revenues, prior-year quarter · 2025-CQ2
Google Cloud revenues, change over the prior-year quarter · 2026-CQ2
Google Cloud revenue growth over the prior-year quarter · 2026-CQ2
What else could explain it
acquisition: Wiz, acquired in March, is now inside Google Cloud for a full quarter and sold with AI-powered security, which the CEO says nearly all its customers use (claim c67); it is a security business, not an AI company, and its revenue sits in the non-AI remainder, nothing of it credited here.
line composition: Google Cloud revenue also holds core GCP services, Workspace, AI infrastructure and, from this quarter, TPU systems.
mix shift: Core GCP services and AI infrastructure are named as drivers of the same growth (claim c2); the parts are not split.
Quotes
“Cloud revenue grew 82%, powered by strong demand for AI infrastructure and AI solutions.”
“Cloud revenues were up 82% to $24.8 billion, driven primarily by GCP, which grew faster than cloud overall. Core GCP, AI solutions, and AI infrastructure were all important drivers of growth.”
“Demand for our models is translating to strong token usage across developers and enterprise customers, and we continue to be supply constrained, a sign of momentum and rapid adoption.”
“We're also seeing strong interest in our AI-powered security platform, which is differentiated because it integrates threat intelligence, cyber response prioritization with Wiz, AI-automated scanning, code remediation, and monitoring.”
“These services provide access to solutions such as artificial intelligence (AI) offerings including our enterprise AI infrastructure, Vertex AI platform, and Gemini Enterprise; cybersecurity offerings; and data and analytics solutions.”
“Google Cloud saw a meaningful acceleration in growth as revenues increased 82% to $24.8 billion, led by an increase in Google Cloud Platform (GCP) across enterprise AI Solutions and enterprise AI Infrastructure, as well as core GCP services.”
Q1 2026direction only · our inference · offensive · $2.62bn to $4.81bn
Q2 2026direction only · described, no size · offensive
product revenue
Google Cloud AI infrastructure: TPU and GPU capacity rented to AI labs, capital markets firms and enterprises
Not sized
AI infrastructure is named as one of several drivers of Cloud growth with no rate or ranking of its own: core GCP, AI solutions and AI infrastructure were all important drivers (claim c2), and the 10-Q names them together (claim c65). Nothing separates its part (the methodology rule for AI named beside another cause, which holds at the upstream companies). Cloud revenue, , is quoted as the ceiling.
described, no sizedisclosure: direction only· motive: offensive· before LLMs: expanded· layer: compute
The CEO names AI infrastructure buyers: an AI lab (Ineffable Intelligence), AI builders, a financial services group, pharmaceutical and robotics companies (claim c15). Counterparty is mixed: labs paying from investor capital, other firms from operations. The sources name no lab buyer as a company Alphabet has put money into (the 10-Q's equity gains are on SpaceX and an unnamed private company), so the labs' part reads investor capital under the lab rule, and the payer mix makes funding mixed. Strong growth with no level makes the state directional. AI infrastructure is named beside AI solutions and core GCP as drivers of one rise with no rate or ranking of its own, so nothing separates its part and the channel is left unsized; Cloud revenue, , is the ceiling. Where the payer is a lab, it is the buildout's money moving.
Evidence: 9 quotes, 5 figures, 3 confounds, 3 from before coverage
Accelerator capacity (TPUs and Nvidia GPUs) sold as a cloud service for training and serving models. Named buyers include AI labs (Thinking Machines Lab, Ineffable Intelligence), capital markets firms (Hudson River Trading, Deutsche Börse Group), robotics and pharmaceutical companies; management does not split them. No level is given; Google Cloud segment revenue is a ceiling. Renting accelerator capacity predates LLMs. Layer: compute. The stored sources name no lab among the buyers as a company Alphabet has put money into, so under the lab rule the labs’ part of the funding is investor capital; with other buyers paying from operations, funding is mixed.
Why this motive
Priced capacity sold while demand still outpaces supply (claims c25, c16).
Before LLMs: expanded
At the anchor TPUs and Nvidia GPUs were already rented through Google Cloud and more than of funded generative AI startups were customers; Google Cloud revenue was in the first quarter of 2024 with no AI infrastructure figure. Accelerator rental predates LLMs and some payers are not AI labs, so the tie-break gives expanded, sized as a level with no traceable quarter before AI. The size is the whole of an activity that existed before.
“Today, more than 60% of funded GenAI startups and nearly 90% of GenAI unicorns are Google Cloud customers.”
“We offer an industry-leading portfolio of NVIDIA GPUs along with our TPUs. This includes TPU v5p, which is now generally available, and NVIDIA's latest generation of Blackwell GPUs.”
“Our AI-optimized infrastructure allows us to use, and offer our customers, a range of AI accelerator options, including our own custom-built Tensor Processing Units (TPUs).”
“Cloud revenues were up 82% to $24.8 billion, driven primarily by GCP, which grew faster than cloud overall. Core GCP, AI solutions, and AI infrastructure were all important drivers of growth.”
“We are seeing strong growth and demand for our AI infrastructure offerings from leading labs such as Ineffable Intelligence, next-generation AI builders, including Kakao, financial services like Deutsche Börse Group, pharmaceutical companies such as Pfizer and Roche, and robotics and spatial intelligence companies such as World Labs.”
“Demand for our models is translating to strong token usage across developers and enterprise customers, and we continue to be supply constrained, a sign of momentum and rapid adoption.”
“I think on the bridge deal, the main thing I would say is, look, on the margin, there are very, very large customers of ours on Cloud who we are trying to support them through this extraordinary moment.”
“These services provide access to solutions such as artificial intelligence (AI) offerings including our enterprise AI infrastructure, Vertex AI platform, and Gemini Enterprise; cybersecurity offerings; and data and analytics solutions.”
“Google Cloud saw a meaningful acceleration in growth as revenues increased 82% to $24.8 billion, led by an increase in Google Cloud Platform (GCP) across enterprise AI Solutions and enterprise AI Infrastructure, as well as core GCP services.”
Q1 2026direction only · described, no size · offensive
Q2 2026direction only · described, no size · offensive
product revenue
TPU system sales: TPU hardware delivered to customer and third-party data centers
0.2% to 1.7% of the quarter’s revenue
Incremental total: counts at zero at the point and in full at the high end, with no traced baseline.
our inferencedisclosure: described· motive: offensive· before LLMs: expanded· layer: hardware
TPU systems were delivered to customer data centers for the first time and revenue began, a small amount of the agreements, with growth accelerating meaningfully even without it (claims c3, c4). Inventory, primarily TPU system hardware and devices, rose to from at year end (claim c64). For certain of the agreements the seller backstops data center and power obligations; its data center credit backstops total , with a further agreed, not split by agreement (claims c12, c13). Funding is mixed: capital markets firms pay from operations, frontier labs in certain cases from investor capital, and the seller’s credit supports some of the data center and power obligations, not the purchase price. The sources name no lab buyer and no Alphabet investment in one. The size, , is the ledger’s residual.
Evidence: 14 quotes, 6 figures, 2 confounds, 2 from before coverage
Multi-gigawatt agreements, signed by Q1 2026, to supply TPU systems (hardware, software, installation and support) to a limited number of customers for their own or third-party data centers; management names capital markets firms and, in certain cases, frontier AI labs. Revenue began in Q2 2026, with the significant majority expected in 2027, and the agreements sit in the Cloud backlog. In connection with certain of these agreements the company provides credit backstops to support third-party data centers and power infrastructure: the backstops cover payment obligations relating to data centers and power, not the purchase price, which the buyers pay (capital markets firms from operations, frontier labs from investor capital), so funding is mixed. The sources name no lab buyer and no Alphabet investment in one; the backstops are the seller’s credit beside the buyers’ own money, and the more conservative reading keeps funding mixed. Layer: hardware.
Why this motive
A separately priced product now recognized as revenue, which the CFO frames as an expansion of the addressable market (claims c3, c10).
Before LLMs: expanded
At the anchor TPUs were offered to customers only as cloud capacity inside the company’s own infrastructure, and the inventory in cost of revenues was for devices. Selling the accelerator as hardware is a new form of the compute the anchor already rents, so the tie-break gives expanded, sized as a level with no traceable quarter before AI. The size is the whole of an activity that existed before.
“We offer an industry-leading portfolio of NVIDIA GPUs along with our TPUs. This includes TPU v5p, which is now generally available, and NVIDIA's latest generation of Blackwell GPUs.”
“Our AI-optimized infrastructure allows us to use, and offer our customers, a range of AI accelerator options, including our own custom-built Tensor Processing Units (TPUs).”
Inventory (primarily TPU system hardware and devices) · as-of 2026-06-30
Inventory at the start of the year · as-of 2025-12-31
Google Cloud remaining performance obligations (backlog) · as-of 2026-06-30
Maximum potential future payments under data center credit backstops · as-of 2026-06-30
Further backstops agreed for data center and energy supply build-out, estimated by the company · as-of 2026-06-30
Google Cloud revenue growth over the prior-year quarter · 2026-CQ2
What else could explain it
other: For certain of these agreements the seller backstops payment obligations of third-party data centers and power infrastructure, so part of the buyers’ build rests on the seller’s credit; the backstops do not cover the purchase price.
line composition: The residual also absorbs any error in the assumed growth of the rest of Cloud.
Quotes
“We also began to recognize revenues from TPU system sales, which we delivered to customer data centers for the first time in Q2. Cloud revenue growth accelerated meaningfully even after excluding the impact of TPU system sales.”
“We start building inventory to be able to sell those systems. You see that impact on the cash from operations because we built ahead, obviously, as we're building that business and ramping up.”
“We continue to expect to recognize a relatively small portion of the revenues from our existing TPU system sales agreements this year, ramping as we exit 2026.”
“We have signed a limited number of agreements to supply TPU systems to customers who require or provide on-premises infrastructure for specialized, high-scale workloads. In the second quarter of 2026, we began recognizing revenues from these agreements, with the significant majority to be recognized in 2027.”
“Other cost of revenues was $29.8 billion, up 22%, driven by increases in depreciation, inventory costs, primarily from the sales of TPU systems to customers, and content acquisition costs, largely for YouTube.”
“The way to think about it is this is an expansion of our total addressable market, so it expands the opportunities by providing solutions to customers that need those systems in their data centers.”
“To the extent people want it as infrastructure, we are balancing it by increasingly looking at opportunities to put TPUs in their data centers or in other data centers, like the project we are doing with Blackstone, et cetera.”
“As of June 30, 2026, we provided backstops in the form of financial guarantees and credit derivatives with maximum potential amount of future payments of $7.6 billion and $43.8 billion, respectively.”
“We have also entered into an agreement to provide an estimated $24.1 billion of future backstops to support the build-out of data center and energy supply infrastructure, subject to finalization of terms with data center providers.”
“To meet the AI compute capacity demands of our customers, we are engaging in the supply of TPU systems which may increase our costs and operational complexity.”
“Inventory consists primarily of hardware related to TPU systems for sale to enterprise customers and devices, which primarily include the Pixel family of products.”
Consumer AI plans: Google One subscriptions with access to the most capable Gemini models
0.12% to 2% of the quarter’s revenue
Incremental total: counts in full.
our inferencedisclosure: direction only· motive: offensive· before LLMs: new
The CFO says Google One growth was driven by demand for AI plans, within subscriptions, platforms and devices of against , and the Gemini app has monthly users (claims c34, c35). Growth credited to AI plan demand, with no level, keeps the state directional. No AI plan count or revenue is given; the size, , is the ledger’s.
Evidence: 2 quotes, 3 figures, 1 confound, 1 from before coverage
The AI tiers of Google One sold to consumers through the Gemini app, recorded in Google subscriptions, platforms, and devices. Management says demand for AI plans drove Google One growth and that the Gemini app is a sizable contributor to it, without a subscriber count or revenue for the AI plans.
Why this motive
Carried: separately priced AI plans whose demand drove Google One (claim c34).
Before LLMs: new
At the anchor Google One had crossed paid subscribers and the AI premium plan with Gemini Advanced had just been introduced; no revenue for it was given. A priced LLM assistant could not exist before LLMs.
“In Q1, we introduced a new AI premium plan with Gemini Advanced.”
Google subscriptions, platforms, and devices revenues · 2026-CQ2
Google subscriptions, platforms, and devices revenues, prior-year quarter · 2025-CQ2
Gemini app monthly active users · as-of 2026-07-22
What else could explain it
bundling: The AI plans bundle storage and other Google One benefits with model access.
Quotes
“Subscription platforms and devices revenues increased 15% this quarter to $12.9 billion, due to strong growth in both YouTube subscriptions, particularly YouTube Music and Premium, and Google One, which was driven by demand for AI plans.”
Q1 2026direction only · our inference · offensive · $120mn to $2.10bn
Q2 2026direction only · our inference · offensive · $144mn to $2.40bn
search discovery
AI Overviews and AI Mode in Search: query growth and new ad formats inside AI answers
Not sized
AI is named beside other causes of Search growth and nothing separates its part, so the channel is not sized (methodology, AI named beside another cause); the line’s rise is kept in the note.
described, no sizedisclosure: direction only· motive: product-defensive· before LLMs: expanded
AI Mode passed monthly users and, like AI Overviews, is said to add queries overall; monetization on queries showing AI Overviews is called encouraging and new formats are being tested in AI Mode (claims c36, c39). Google Search & other revenue was against , a rise of , with paid clicks up and cost per click up ; the 10-Q explains the rise by queries, advertiser spending and ad formats without naming AI, and no AI share is given. Under the joint-cause rule the channel reads described.
Evidence: 6 quotes, 4 figures, 2 confounds, 2 from before coverage
Unpriced LLM features in Search that management says drive query growth, including commercial queries, with ads tested and deployed inside AI Mode (Direct Offers, contextual site links, sponsored links in answers). The money is advertisers’ spend in Google Search & other; management credits Search growth to many parts of the business together and does not separate AI’s part, and the 10-Q explains the line by queries, advertiser spending, ad formats and currency without naming AI.
Why this motive
Carried: unpriced features users now expect, whose serving cost management keeps lowering, with ad formats inside AI Mode still being tested (claims c38, c40).
Before LLMs: expanded
At the anchor AI Overviews were starting to reach the main results page, with higher search usage in testing; Google Search & other revenue was in the first quarter of 2024. A feature lifting an existing revenue line is expanded, sized as an increment. The size is the change AI made, not the whole line.
“And now, we are starting to bring AI overviews to the main Search results page.”
“Most notably, based on our testing, we are encouraged that we are seeing an increase in search usage among people who use the new AI Overviews, as well as increased user satisfaction with the results.”
Paid clicks, change over the prior-year quarter · 2026-CQ2
Cost per click, change over the prior-year quarter · 2026-CQ2
AI Mode monthly active users, a floor · as-of 2026-07-22
Google Search & other revenue growth over the prior-year quarter · 2026-CQ2
What else could explain it
other: Management credits Search growth to many parts of the business working together beside Gemini integration (claim c41); nothing separates the AI part.
one time item: The CFO names strong ad growth related to the World Cup, mainly in YouTube.
Quotes
“Just like AI Overviews, AI Mode is driving an incremental increase in Search queries overall, and we are now sending billions of clicks to websites every week through AI features in Search.”
“Thanks to our engineering and hardware optimizations, this quarter, we reduced the cost of AI Mode responses to its lowest level since launch, even as we have brought more advanced AI capabilities.”
“We continue to be encouraged with monetization performance on queries that show AI Overviews, even as we've expanded AI Overviews to more commercial queries.”
“Look, our 17% year-over-year growth rate in the Q2 Search and other revenues was really driven by many parts of our business working well together and very, very deep Gemini integration.”
Q1 2026direction only · described, no size · product-defensive
Q2 2026direction only · described, no size · product-defensive
product ranking
Gemini in the ads systems: ad relevance, Smart Bidding, AI Max and Performance Max
Not sized
Management gives advertiser outcomes and adoption counts, not a revenue effect for the company, and credits Search growth to several causes together; nothing separates the AI part, so the channel is unsized (methodology, AI named beside another cause).
described, no sizedisclosure: described· motive: offensive· before LLMs: expanded
AI Max has advertisers, advertisers in AI-powered campaigns see an average of more conversions or value, and shopping ads showed a improvement in relevance (claims c43, c44, c45). AI Max also unlocks searches that were not monetizable before (claim c46). The anchor already credited AI and LLMs in bidding and matching; coverage gives measures of AI-attributed demand, so the channel is read as expanded, and it stays unsized. Booking Holdings (BKNG, on the ledger) is partnering on AI-powered ad formats (claim c66); its ad spend is a ledger company’s money.
Evidence: 7 quotes, 3 figures, 2 confounds, 4 from before coverage
Models applied to ad matching, bidding and creative across Search and YouTube, and the AI-powered campaign types sold to advertisers (AI Max, Performance Max). Management gives relevance and conversion rates for advertisers and adoption counts, not a revenue effect for the company.
Why this motive
Measured relevance and conversion gains attributed to Gemini in the ads systems (claims c45, c44).
Before LLMs: expanded
At the anchor Smart Bidding used AI, Broad Match used LLMs, Gemini generated Performance Max assets, and advertisers using automatically created assets saw more conversions on average; Google Search & other revenue was in the first quarter of 2024. The LLM tools in the anchor date from after LLMs, so their presence there does not make coverage a rename, and coverage gives the measures the cohort reading asks for, AI-attributed demand for an existing product: an attach count, a cohort conversion rate and a volume of searches newly monetizable. Expanded, sized as an increment. The size is the change AI made, not the whole line.
“We've talked about how solutions like Smart Bidding use AI to predict future ad conversions and the value in helping businesses stay agile and responsive to rapid shifts in demand, and how products like Broad Match leverage LLMs to match ads to relevant searches and help advertisers respond to what millions of people are searching for.”
“In February, we rolled Gemini into PMax. It's helping curate and generate text and image assets so businesses can meet PMax asset requirements instantly.”
Advertisers that have adopted AI Max · as-of 2026-07-22
Additional conversions or value on Search for advertisers in AI-powered campaigns, on average · 2026-CQ2
Improvement in showing highly relevant shopping ads · 2026-CQ2
What else could explain it
relabel: Part of the claimed effect is machine-learning bidding and matching that the anchor already credited with conversion gains of the same kind; only the LLM part is new.
other: Search growth is credited to many parts of the business together.
Quotes
“We're applying Gemini models really across our entire ads infrastructure, whether it's ads quality, advertising tools, ads in new AI experiences.”
“Those who adopt our AI-powered campaigns like AI Max or Performance Max, see an average of 15% more conversions or value on Search at a similar ROAS.”
“Look, our 17% year-over-year growth rate in the Q2 Search and other revenues was really driven by many parts of our business working well together and very, very deep Gemini integration.”
Gemini-assisted pitch tools in the advertising sales force
0.01% to 0.61% of the quarter’s revenue
Incremental total: counts in full.
our inferencedisclosure: direction only· motive: offensive· before LLMs: expanded
The channel opens this quarter. of the sales team uses Gemini-assisted tools weekly, with up to higher win rates on customized pitches (claim c51). A usage share and an upper rate with no level make the state directional. No revenue is attributed; the size, , is the ledger’s, a share of Google advertising revenue of .
Evidence: 2 quotes, 3 figures, 1 confound, 2 from before coverage
Gemini-assisted tools the sales team uses weekly to build customized pitch narratives, with a win-rate gain management gives as an upper figure. The money is advertisers’ spend in Google advertising won on those pitches; no revenue is attributed, so the size is the ledger’s.
Why this motive
A measured movement in conversion that management attributes to Gemini tools, a higher win rate, given as an upper figure (claim c51).
Before LLMs: expanded
At the anchor sales and marketing held the cost of sales staff, sales compensation was rising for Cloud sales, and the line was in 2024. LLM tools put into existing selling are expanded, sized as an increment: the revenue won above the prior win rate. The size is the change AI made, not the whole line.
“On an adjusted basis, operating expenses were up 5%, reflecting first in R&D, an increase in compensation expense primarily for Google DeepMind and Cloud, and second in sales and marketing, a slight increase year-on-year, reflecting increases in compensation expense primarily for Cloud sales.”
Agentic solutions bringing new small and medium advertisers to Google Ads
Not sized
The quarter's only measure is a count of customers: hundreds of thousands of new customers reached year to date by the agentic solutions for small and medium businesses (claim c52), with no spend or revenue. A count of customers sizes nothing (the methodology rule for counts).
described, no sizedisclosure: direction only· motive: offensive· before LLMs: expanded
The channel opens this quarter. Agentic solutions for small and medium businesses reached hundreds of thousands of new customers year to date (claim c52). A count of customers with no dollar level reads directional and sizes nothing, so the channel is left unsized.
Evidence: 1 quote, 1 figure, 1 confound, 2 from before coverage
Agentic products for small and medium businesses that management says expanded its reach by hundreds of thousands of new customers year to date. The money is the new advertisers’ spend in Google advertising; no revenue is attributed, so the size is the ledger’s, built from the customer count. The sales team’s Gemini tools are a separate channel.
Why this motive
A customer count management attributes to its agentic products, new small and medium advertisers reached year to date (claim c52), a measured movement in customers.
Before LLMs: expanded
At the anchor businesses of all sizes already relied on Google Ads, and Google Search & other revenue was in the first quarter of 2024. Selling ads to small businesses predates LLMs; the new customers the agentic products bring are the increment. The size is the change AI made, not the whole line.
“We also know businesses of all sizes around the world rely on Google Ads to find customers and grow their businesses”
Q2 2026. Growing slower than revenue: cost of revenues (+17.7%), sales and marketing (+18.3%), general and administrative (+24.0%), total costs and expenses (+21.3%). A displaced cost shows up as a line that stays under the dashed revenue line. These are the audited lines, as first reported; nothing here is attributed to AI by the filing. Not drawn: general and administrative, which one-time items move by more than 60% in a quarter; the values are in the table below.
Cost of revenuesResearch and developmentSales and marketingTotal costs and expensesRevenue