Business Overview

Alphabet Inc. (NASDAQ: GOOG / GOOGL)

2 September 2026

Evidence base: web research only — no documents were provided for this run.

Primary sources: Alphabet Form 10-Q for the quarter ended 30 June 2026; Form 10-K FY2025;

quarterly earnings releases Q1 2025 – Q2 2026 and Forms 8-K to 10 August 2026;

competitor filings; Synergy Research; StatCounter; WPP Media; Pew Research; US court records.

This is a business and industry primer for a long-term investor.

It is not a valuation and not a recommendation.

1. Executive Snapshot

2. What the Company Does

Alphabet solves two unrelated problems for two unrelated customers. For an advertiser, the problem is that most advertising is bought before anyone has expressed a want; Google sells the moment after the want is stated. For an enterprise, the problem is that training and serving AI models requires accelerators, power and floor space that almost nobody can procure at scale; Google Cloud rents that stack out. The two businesses share an engineering base and almost nothing else economically.

The advertising unit, traced end to end

The unit is one paid click. An advertiser sets bids in an auction. A user types a query. Google ranks ads against that query and serves them. Nothing is billed until the user clicks. On the click, the advertiser is charged the auction-clearing price, and Google pays a share of that price away as traffic acquisition cost — to the browser, handset maker or carrier that supplied the query, or to the third-party publisher that supplied the page. What is left is Google's revenue on the unit, and it converts at a very high rate because the marginal cost of ranking one more query is close to nothing.

In Q2 2026 that arithmetic ran as follows. Google advertising revenue was $81,629m and total traffic acquisition cost was $16,179m, a TAC rate of 19.8%. So roughly four-fifths of every advertising dollar was retained before any operating cost. Volume, not price, is doing the work: paid clicks rose 13% year on year while cost-per-click rose 3%, a ratio of roughly four to one.

Alphabet discloses only percentage changes in clicks and price, never absolute counts. Revenue per click therefore cannot be computed from the filings, and no figure claiming to do so should be trusted.

What each line actually is

Alphabet reports three segments — Google Services, Google Cloud and Other Bets — but the revenue disaggregation inside Services is where the business is visible. FY2025 broke down as follows.

Source: Alphabet Form 10-K FY2025, revenue disaggregation note. Percentages of consolidated revenue of $402,836m; hedging gains and losses excluded, so the column does not sum to 100%.

Search & other is the business. It is queries on google.com plus Gmail, Maps and Play, and it is owned inventory — Google controls the surface, the auction and the user relationship. Google Network is the opposite: ads Google sells on other people's websites and apps through AdSense, AdMob and Ad Manager. Network carries a far higher TAC rate because most of the revenue is paid straight to the publisher, and it has now declined for sixteen consecutive quarters, from a peak of $9,305m in Q4 2021 to $7,303m in Q2 2026 while group revenue roughly doubled.

That decline is the single most informative line in the disclosure. The part of the business that is shrinking is the part where Alphabet is an intermediary rather than a principal, and it is the low-margin part. Some of the Services margin expansion over the last three years is simply mix: the worse business got smaller.

The cloud unit, and a new one

Google Cloud sells three things: consumption-based infrastructure and AI platform services, Workspace subscriptions, and — from the second quarter of 2026 — TPU systems sold as hardware. The revenue-recognition policy was rewritten in the Q2 2026 10-Q to name "the sale of TPU systems" as the source of product revenue, and inventory rose from $2,439m at December 2025 to $9,991m at June 2026, $6.7bn of that in Q2 alone. Management said delivery to customer data centres began in the quarter and that most of the revenue will land in 2027. Alphabet is now building accelerators to sell, not only to run.

The unit here is a contracted accelerator-hour rather than a click, and the volume is visible in advance. Revenue backlog reached $519.5bn at 30 June 2026, of which $513.9bn was Google Cloud — roughly 6.6 times the segment's last twelve months of revenue.

Other Bets, and what is leaving

Other Bets earned $382m in Q2 2026 against an operating loss of $1,799m; the FY2025 loss was $7,515m, widened by a $2.1bn valuation-based employee compensation charge for Waymo. Waymo passed 500,000 fully autonomous rides a week and raised $16.0bn in February 2026, the significant majority funded by Alphabet. Google Fiber is being contributed to a new entity for $1.5bn in cash, a $2.0bn note and a 49.99% stake, with $7.1bn classified as held for sale and closing expected late in 2026. The segment is being narrowed toward the one asset with a plausible path to scale.

3. Industry, Competitive Position & Moat

Alphabet sits inside two industries with different structures, and its position in each is moving in opposite directions.

Advertising: a profit pool that sits with whoever owns the surface

Global advertising revenue excluding US political spend is forecast to grow 8.9% in 2026 to $1.3 trillion, with the US up 11.9% (WPP Media, This Year Next Year midyear forecast, June 2026). Dentsu's December 2025 forecast put growth at 5.1% on a narrower definition; the two disagree, and the gap is definitional rather than a dispute about direction. WPP puts traditional and generative search combined at 21.8% of all advertising revenue in 2026.

The economics of the value chain are simple. Profit accrues to whoever owns the surface where purchase intent is expressed, not to the intermediaries who move budget across surfaces. Alphabet's own numbers demonstrate the point better than any industry study: its owned surfaces run at a 41.9% segment operating margin on a last-twelve-months basis while its intermediary business shrinks.

Barriers to entry in search are frequently overstated in one direction and understated in another. Advertiser switching costs are close to zero — a media buyer can move budget between Google, Meta and Amazon within a day, and nothing in the filings suggests otherwise. The barrier that actually binds is on the user side: a click-relevance feedback loop trained on more than five trillion queries a year, which no entrant can bootstrap, coupled with owned distribution through Android and Chrome.

On scale Alphabet remains unmatched. Its last-twelve-months advertising revenue of roughly $315bn exceeds Meta's $201bn and Amazon's $68.6bn of advertising revenue combined. But both grew faster: Amazon advertising rose 22.1% and Meta 22.2% against 17.9% growth in Alphabet's Search & other line in H1 2026. Alphabet is losing share of a growing digital advertising pool. eMarketer projected in April 2026 that Meta would pass Google in digital ad revenues for the first time, though on a net measure that is not comparable with the gross figures above.

Cloud: third by size, first by rate of change

Synergy Research put enterprise cloud infrastructure spend at $129bn in Q1 2026, up 35% year on year, with the annualised run rate passing $500bn for the first time. Share was AWS 28%, Microsoft Azure 21% and Google Cloud 14%, with the Big Three taking roughly two-thirds of the market. Other trackers place Google between 11% and 14% depending on methodology; the ranking is not in dispute.

Source: each company's own filings — Microsoft Form 10-K FY June 2026, Amazon Form 10-K FY2025, Alphabet Form 10-Q Q2 2026. Segment definitions differ; Microsoft's Intelligent Cloud includes enterprise services beyond Azure.

Google Cloud is 60% of AWS's size and growing roughly 3.7 times as fast, with the margin gap to AWS now about four percentage points and closing. That is the clearest competitive fact in the whole picture, and it is measured against competitors' own reported figures rather than the company's characterisation.

What would be hardest for a well-funded rival to reproduce

Not the model, and not the data centre. Both are being replicated with capital. What is hard is owning all four layers at once — the frontier model, the accelerator, the data centre and the power — because each layer owned removes both a supplier's margin and a supplier's queue.

Alphabet has DeepMind and Gemini, the TPU, its own data centres, and after the $5.9bn acquisition of Intersect in March 2026, a renewable energy developer. Microsoft rents its frontier model from a partner and buys merchant silicon. Amazon has custom silicon but no frontier model of its own. Meta has models but no cloud business through which to sell them. On the evidence available, Alphabet is currently the only one of the four that owns every layer.

The industry rewards exactly that. In an environment where accelerators, grid interconnects and land are the binding constraints — not software and not demand — the winner is the firm that has removed the most intermediaries between a customer's workload and an electron. Alphabet's own actions confirm which constraint it believes binds: it bought a power developer, signed energy contracts with take-or-pay minimums running to 2054, and posted $43.8bn of credit-derivative notional backstopping third-party data-centre developers.

Testing the company's account against outside evidence

• Claim: AI is not cannibalising Search — partially supported. Alphabet's own paid clicks rose 13% in Q2 2026 and StatCounter puts Google at roughly 89–90% of conventional search, down from about 91% in 2023. But Pew Research Center clickstream data found users clicked a result on 15% of visits without an AI summary and only 8% of visits with one, and Apple testified in May 2025 that Safari searches fell for the first time in over 22 years. The query is safe; the click is the exposed variable. Management's language has also softened, from monetising AI surfaces at "approximately the same rate" in mid-2025 to being "encouraged with monetization performance" by Q2 2026, with no parity claim attached.
• Claim: a differentiated full-stack position in cloud — supported. Third by share but growing 3.7 times faster than AWS with a margin within four points of it, on competitors' own numbers.
• Claim: demand exceeds supply — supported. The backlog series and management's repeated statement that it is compute constrained are disclosures against interest, not marketing. Pichai on the Q1 2026 call: cloud revenue would have been higher had Alphabet been able to meet demand.
• Claim: an unassailable advertising position — contradicted at the margin. Alphabet earns more advertising revenue than Meta and Amazon combined and is still growing at a mid-teens rate, but both rivals grew faster in the most recent comparable period.

The regulatory shape of the moat

Judge Mehta's remedies decision of 2 September 2025, entered as final judgment on 5 December 2025 and partly effective from February 2026, is materially narrower than what the Department of Justice sought. Alphabet will not divest Chrome or Android; the court found that plaintiffs had overreached. Critically, Google may still pay for placement — the Apple, Samsung and carrier payments continue. What it may not do is sign exclusive default agreements for Search, Chrome, Assistant or Gemini, or condition revenue-share benefits on placement for more than a year at a time. It must also license certain search index and user-interaction data to qualified competitors and offer syndication, though not advertising data. A six-year term applies under a Technical Committee. Both sides have appealed to the D.C. Circuit and briefing was still under way in late August 2026.

The practical effect on the moat is smaller than the headline. The mechanism the court removed was exclusivity; the mechanism that actually holds the position — users choosing Google when offered a choice — was untouched. The European choice-screen precedent is the relevant evidence: roughly 97% of users still selected Google. What could weaken the position is not this judgment but the ad-tech case, which remains undecided and is dealt with in Section 7.

4. Growth Engine

Alphabet grew 24% year on year in Q2 2026, its twelfth consecutive quarter of double-digit growth and the fastest rate since 2021. Revenue rose $23.4bn against the same quarter of 2025. Where that money came from matters more than the headline, because two of the four contributors are structural and one is shrinking.

Source: Alphabet Form 10-Q for the quarter ended 30 June 2026, revenue disaggregation note. Shares of growth are calculated against the $23,368m increase in the five lines above and exclude Other Bets and hedging.

Reported growth versus organic growth

Two acquisitions closed in March 2026 and both sit inside the reported rate, so the headline needs separating. Wiz, acquired for $29,467m and consolidated into Google Cloud from 11 March, had crossed $1bn of annual recurring revenue before close according to press reporting — the figure is not disclosed in Alphabet's filings. At roughly $250–300m a quarter it accounts for perhaps two to three percentage points of Cloud's 82% growth, leaving organic Cloud growth around 79–80% (inferred). Intersect adds power generation capacity and property, not revenue. At group level the acquisitions contribute well under one percentage point of the 24%. This is overwhelmingly a grown number, not a bought one — but Alphabet does not disaggregate the contribution, so the split is an estimate rather than a disclosure.

The drivers, ranked

• Cloud backlog conversion — structural. Google Cloud is the largest single contributor and the fastest-moving. Growth accelerated from 32% in Q2 2025 to 48% in Q4 2025 to 82% in Q2 2026, and the backlog went from $108.2bn a year ago to $513.9bn. Because the backlog is contracted, roughly half of it converts to revenue within 24 months on management's guidance, which means near-term Cloud growth is largely already sold. The constraint is delivery, not demand.
• Search query volume — structural. Search grew 17% in Q2 2026 on paid clicks up 13% and cost-per-click up 3%. This is a volume story with a modest price tail, which is the healthier of the two configurations: it means more queries are being monetised rather than existing queries being repriced. It is also the driver that the bear case says should already be failing, and it is not.
• US geographic concentration — cyclical. The United States supplied roughly 63% of the entire year-on-year revenue increase in Q2 2026 and crossed 50% of group revenue for the first time. US growth went from 11.8% a year earlier to 32.1% while EMEA decelerated from a 21.4% peak to 15.0%. The concentration is real and it is partly circular: much of the incremental US demand is AI-sector capital spending, funded by the same cycle Alphabet is selling into.
• Consumer subscriptions — management-driven. Paid subscriptions reached 350m by Q1 2026 from 325m at the end of 2025, led by Google One and YouTube Premium, and YouTube passed $60bn of combined advertising and subscription revenue in FY2025. This is the most predictable line in the company and the least discussed.
• TPU systems sold to third parties — management-driven, not yet visible. First deliveries occurred in Q2 2026 and management expects the majority of revenue in 2027, so this driver is almost entirely absent from current figures. Inventory building from $2.4bn to $10.0bn in six months is the leading indicator. Margins have been explicitly refused: "We don't break out margins for any specific products."

5. Margin, Cash & Capital Allocation

How the margin is made

Alphabet's consolidated operating margin of 34.0% in Q2 2026 is an average of three very different things. Google Services earned $39,544m on $94,540m of revenue, a 41.8% margin. Google Cloud earned $8,814m on $24,768m, a 35.6% margin, up from 20.7% in the same quarter a year earlier. Other Bets lost $1,799m.

Against all of that sits an unallocated Alphabet-level cost of $5,789m in the quarter, described in the filings as shared AI research and development plus corporate fines and settlements. That line has grown from $1,299m for the whole of FY2022 to $16,760m in FY2025 and $11,178m in the first half of 2026 alone. It is where frontier model development hits the profit and loss account before it reaches capital expenditure, and it is unallocated precisely because no segment can be asked to carry it.

The two segments have travelled in opposite directions across the full record. Google Cloud's operating margin ran (11.9%) in FY2021, (7.3%) in FY2022, 5.2% in FY2023, 14.1% in FY2024 and 23.7% in FY2025 before reaching 31.3% on a last-twelve-months basis. Google Services over the same six periods ran 37.1%, 32.6%, 35.2%, 39.8%, 40.7% and 41.9%. One business found its floor and climbed 43 points; the other dipped through the 2022 advertising trough and recovered to a new high.

Cloud's margin expansion is the clearest evidence of operating leverage in the business. In H1 2026 the segment added $10.4bn of operating income on $18.9bn of incremental revenue, an incremental margin of roughly 55%. Management has flagged one offset: leasing third-party capacity as a bridge to its own build "will put some pressure on operating margins for Cloud."

The advertising margin is being helped by mix rather than by pricing. Traffic acquisition cost fell from 20.6% of Google advertising revenue in Q2 2025 to 19.8% in Q2 2026, not because distribution partners accepted less but because Google Network, which pays away most of its revenue to publishers, keeps shrinking. That tailwind ends when Network stops shrinking or becomes too small to matter.

The financial shape of the change

Source: Alphabet Forms 10-K FY2022, FY2024 and FY2025 and Form 10-Q Q2 2026; free cash flow as the company defines it, being operating cash flow less purchases of property and equipment. Comparability: FY2025 includes a $3.5bn European Commission ad-tech fine in Q3 and a $2.1bn Waymo compensation charge in Q4; the LTM column includes Wiz from 11 March 2026 and an approximately $2.1bn PriceRunner damages accrual. FY2023 is omitted deliberately — the four columns chosen show the trough, the pre-build baseline, the last full year and the current run rate.

The table contains the whole argument. Revenue growth accelerated and the operating margin expanded, which is unusual at this scale. Free cash flow did the opposite: it is now below where it stood in FY2021, when the company was a third smaller. Capital expenditure is what separates the two, and it has moved from 11% of revenue to 30%.

For the full arc, capital expenditure ran $24,640m in FY2021 and $32,251m in FY2023, against free cash flow of $67,012m and $69,495m in those same two years. Nothing about the pre-2024 record predicts the present one.

Cash conversion has inverted

Operating cash flow of $185,675m over the last twelve months is the highest in the company's history. Free cash flow of $53,273m is the lowest since before 2021. The quarterly path shows when it broke: $24.5bn in Q3 2025, $24.6bn in Q4, $10.1bn in Q1 2026 and negative $5,855m in Q2 2026 — the first negative free cash flow quarter on record, with $39,069m of operating cash flow against $44,924m of capital expenditure.

The reported figure also understates what has been committed. At 30 June 2026, $29,113m of property and equipment sat in accounts payable and accrued liabilities rather than in cash paid, $122,814m of assets were recorded as not yet in service, and contractual commitments — largely long-term supply agreements for infrastructure and inventory components — stood at $707.0bn, with a further $85.2bn of leases signed but not yet commenced. The spending disclosed in the cash flow statement is the part that has already cleared.

Depreciation is the mechanism that converts all of this into the profit and loss account, and it is running far behind the asset base. It rose from $15,311m in FY2024 to $21,136m in FY2025 and reached $13,586m in the first half of 2026 alone. Capital expenditure is running at 5.25 times depreciation on an LTM basis, which means the asset base is growing more than five times faster than it is being written off. The servers carry a six-year useful life adopted in 2023; the Q2 2026 10-Q discloses no change to it.

Where the cash went, and the change in who has first claim on it

Ranked over the last twelve months: capital expenditure $132.4bn, acquisitions $35.3bn, share repurchases $17.4bn, dividends $10.3bn, and net debt repayment of nothing — Alphabet was a large net issuer. The ranking within the first half of 2026 is starker still, because repurchases were exactly zero against $28,306m in the same period of 2025, with $69.5bn of the April 2025 authorisation left unused.

The funding side changed more than the spending side. In June 2026 Alphabet issued equity for cash for the first time in its history: 29m Class A and 29m Class C shares at roughly $353 in a public offering, a $10.0bn private placement to a Berkshire Hathaway affiliate, and 19m shares of 6.25% mandatory convertible preferred, for aggregate net proceeds of $49.6bn. It bought capped calls at a $1.0bn premium to limit conversion dilution below about $530. It also authorised a $40.0bn at-the-market programme, undrawn at 30 June, intended primarily to fund employee equity tax obligations. Long-term debt face value went from $49,085m at December 2025 to $101,085m at June 2026 across six currencies.

What the ranking reveals is a deliberate reordering of claims on cash. Internal reinvestment now comes first, and shareholders come last — to the point where the company is issuing stock at the same time as it has stopped buying it back. Diluted share count rose from 12,198m to 12,309m year on year, and the 2021 Stock Plan reserve grew by a further 200m Class C shares at the June annual meeting. The dividend was raised 5% to $0.22 a quarter, which is a small gesture beside a $150bn swing in capital intensity. This is behaviour, not a verdict: management is buying an option on being the infrastructure provider of the next decade and is willing to dilute existing holders to pay for it.

Three items post-date the reported figures and belong here. On 10 August 2026 Alphabet closed a $25.0bn US dollar senior note offering across ten tranches at a weighted average coupon of about 5.61% and weighted average maturity of 15.5 years, taking long-term debt face value to roughly $126.1bn; Q2 interest expense was already $1,278m against $261m a year earlier, and the August tranche adds roughly $1.4bn a year. The $40bn at-the-market programme had not begun at the balance sheet date and was expected to start in Q3 2026. The Google Fiber contribution is expected to close late in 2026.

The other half of reported earnings

Diluted earnings per share of $9.11 in Q2 2026 is not an operating number. Alphabet booked a $99,031m gain on equity securities in the quarter and $135,946m in the first half, and it states directly that $6.26 of the $9.11 came from that gain, along with $21.9bn of associated tax. Over the last twelve months these marks represent close to half of pre-tax income. Roughly $94.1bn of the position is now SpaceX stock, restricted from sale in tranches through the third quarter of 2027; the rest is a non-marketable holding the filings identify only as "a private company," widely reported to be Anthropic though not named in any filing (inferred). The same line was a loss of $5,519m in FY2022.

6. Cyclicality, Constraints & What to Monitor

Alphabet has an advertising cycle it has already survived once and a capital cycle it has never experienced. The two need separating because the second is now the larger exposure.

Where the business sits against its own peak and trough

The advertising trough is documented: in Q4 2022 revenue growth fell to 1.0% and the operating margin to 23.9%, as post-pandemic advertising normalised. Revenue never actually declined. That is the downside case for the advertising business, drawn from its own record rather than from a forecast, and it is survivable.

Against that record, the business today sits near the top on every measure. Revenue growth of 24.2% in Q2 2026 is the fastest since 2021, and the operating margin of 34.0% is close to the highest the company has reported.

But the margin peak has already passed. It reached 36.1% in Q1 2026 and fell 2.1 points sequentially, with operating income growing 2.7% quarter on quarter while revenue grew 9.0%. Depreciation reached the margin one quarter before it reached the annual figures. A record margin at a peak and a record margin at a trough are opposite facts, and this is the former, one quarter after the turn.

The constraints that bind

Supply, not demand, has been the binding constraint on every earnings call from Q2 2025 to Q2 2026, with no easing date offered. Pichai stated on the Q1 2026 call that the company is compute constrained in the near term and that cloud revenue would have been higher had it been able to meet demand. The backlog series — $108bn, $158bn, roughly $240bn, $462bn, $514bn — is the measurable form of that statement.

The downside mechanism this business would actually feel is not an advertising recession. It is that depreciation on a six-year server life arrives on a fixed schedule while backlog conversion depends on delivery and utilisation. At 5.25 times depreciation, the charge is effectively contracted for; the revenue that justifies it is not. If utilisation of the 2026–2028 vintages disappoints, the operating margin compresses on a timetable management can only alter by revisiting the useful-life assumption — an accounting judgement rather than an operating fix. Notably, server useful lives were not discussed on any of the last five earnings calls, and no analyst raised them.

Two further constraints are contractual rather than physical. The energy service agreements carry terms of two to 26 years with obligations running to 2054, take-or-pay minimums and substantive termination fees; and Alphabet guides capital expenditure and nothing else — no revenue, margin or tax guidance was given on any of the last five calls, so the only forward number the company stands behind is the one that consumes cash.

Durable versus borrowed

Leading indicators, and where they are published

• Quarterly Cloud backlog additions. Form 10-Q, Note 2, "Revenue Backlog." The single best forward read on Cloud. Additions fell from roughly $220bn in Q1 2026 to $54bn in Q2 — the sharpest change in the disclosed series, and worth watching before the revenue line reflects it.
• Paid clicks and cost-per-click. Form 10-Q, Management's Discussion, monetisation metrics table. The bear case on AI requires paid-click growth to turn negative. It has not.
• Depreciation as a percentage of revenue, against capex guidance. Consolidated statement of cash flows, and the capex guidance given on each earnings call. Guidance has been raised at five consecutive calls, from $75bn for FY2025 to $195–205bn for FY2026.
• Server useful life. Form 10-K, summary of significant accounting policies. Any extension beyond six years would flatter reported margins without changing cash; any shortening would do the reverse.
• Cloud market share. Synergy Research Group's quarterly hyperscaler tracker, against each company's reported segment growth.
• Court dockets. D.C. Circuit docket for the search appeal; Eastern District of Virginia for the ad-tech remedies judgment.

7. Risks, Unknowns & Questions for Deeper Work

Cyclicality and the depreciation schedule are covered in Section 6. What follows is what those do not capture, ordered by the degree to which the risks compound one another.

• The ad-tech remedy is still undecided, and it is the one that could remove an asset. Judge Brinkema found in April 2025 that Google's publisher tools unlawfully excluded rivals. The remedies proceeding ran in September 2025, closing arguments were heard in November 2025, and as of the Q2 2026 10-Q no judgment had been issued. The Department of Justice sought structural remedies and the company states these "could have a material adverse effect." The honest read is two-sided: Google Network is $29.8bn of FY2025 revenue, so a forced divestiture would remove about 7% of revenue — but it is the lowest-margin 7%, and removing it would raise the group margin. The Texas Attorney General case follows on from whatever is decided.
• The click, not the query, is the billable event. Alphabet is paid when a user engages with an advertisement, not when a question is answered. Every query Gemini resolves completely is a query that produced no auction. Pew's clickstream evidence — a 15% click rate without an AI summary against 8% with one — describes the mechanism precisely. Paid clicks are still growing 13%, so this has not happened at the level that matters. But the metric to watch is not query volume, which Alphabet emphasises, and not engine market share, which is stable. It is paid clicks, and the company reports that figure only as a percentage change.
• The equity marks can reverse, and Alphabet is now issuing equity against them. Non-marketable equity carried at $124.3bn was written up by $87.9bn in a single quarter on observable third-party funding rounds, which are procyclical by construction. A reversal would not touch operating income, but it would hit reported earnings and shareholders' equity — and Alphabet is now issuing stock against that equity base and has committed a further $20.0bn of contingent funding to the same counterparty through 2030. The exposure and the funding obligation move together.
• Dilution and depreciation compound. The at-the-market programme removes the share shrinkage that net settlement of employee equity used to provide; the stock plan reserve grew by 200m Class C shares; diluted shares are already rising. Set that against a depreciation charge scheduled to grow faster than revenue for at least two years, and earnings per share can fall while revenue grows. Neither element alone is alarming. Together they are the mechanism by which a good operating year becomes a poor reported one.
• Backlog concentration is undisclosed. There is $513.9bn of Cloud backlog and $707.0bn of supply commitments, and Alphabet discloses no customer concentration for the Cloud segment at all. If a material share of the backlog sits with a small number of AI laboratories funded by the same capital cycle that is funding Alphabet's own build, then the asset and the obligation are correlated, and a single counterparty failure would strike both sides of the balance sheet. Nothing in the filings establishes that this is the case; nothing in the filings rules it out either.

What the sources could not answer

These are findings, not gaps in the research. Each would need resolving before an investor could form a thesis on this business.

• Absolute paid clicks and revenue per click. Only percentage changes are disclosed, so the unit economics of the core business cannot be reconstructed from public filings.
• Google Cloud customer concentration. No segment-level disclosure exists, and no customer exceeds 10% at group level.
• The maintenance versus growth split of capital expenditure. Not disclosed. Without it, the sustainable level of free cash flow after the build cannot be estimated.
• TPU system revenue and its gross margin. Management explicitly declined: "We don't break out margins for any specific products." Whether merchant silicon carries hardware margins or cloud margins is a first-order question for 2027.
• Wiz's revenue contribution. Not disaggregated, so the organic Cloud growth rate is an estimate.
• The identity of the private company behind $124.3bn of non-marketable holdings. Named nowhere in the filings.
• Whether the six-year server life still reflects the economics of AI-era hardware. No change disclosed, and the question was not raised on any of the last five earnings calls.
• Segment revenue by geography. Alphabet reports geography only at consolidated level, so the US surge cannot be attributed between advertising and cloud.
• The share of AI-assistant usage displacing search. Published estimates range from under 1% of referral traffic to roughly 17% of digital queries depending on methodology; the spread is too wide for any figure to be usable.

8. Investor Takeaways

• What this business really is. This is no longer an advertising company that happens to own a cloud business. It is an advertising business funding the construction of an AI infrastructure business, and the construction is now large enough to determine the reported numbers.
• The core economic engine. An auction that retains roughly 80c of every advertising dollar after distribution costs and converts 42c of every revenue dollar to operating profit, attached to a cloud business now converting 36c and growing at 82%.
• The main growth lever. Conversion of $513.9bn of contracted Cloud backlog, roughly half of it scheduled within 24 months. Near-term cloud growth is largely already sold; the constraint is delivery.
• What could break the story. Depreciation on a six-year life arriving faster than the utilisation that justifies it, while the buyback stays off and the share count rises. That combination makes falling earnings per share on rising revenue an arithmetic possibility rather than a bearish opinion.
• What to monitor. Quarterly Cloud backlog additions, which fell from roughly $220bn to $54bn between Q1 and Q2 2026; paid-click growth, currently 13%; and depreciation as a percentage of revenue.