Alphabet Inc.

01 — Executive summaryNet income of $112.2 billion. Operating income of $40.8 billion. Both are correct.

Alphabet reported revenues of $119.8 billion for the quarter ended 30 June 2026 — a 24% increase, the fastest rate of growth the company has recorded since 2021, and its twelfth consecutive quarter of double-digit expansion. Operating income rose 30% to $40.8 billion. Google Cloud grew 82%, a rate no business of its size has sustained in the history of enterprise computing, and did so at a 35.6% operating margin, up 18.5 percentage points in eight quarters. Google Search & other grew 17%. Diluted earnings per share were $9.11, up 294%.

Every one of those numbers is accurate and none of them describes the same company. The $9.11 of earnings contains $6.26 that arrived because two private companies Alphabet owns pieces of were marked up in value during the quarter. Operating income of $40.8 billion sat above an "other income" line of $98.0 billion. Net income of $112.2 billion was 2.8 times the profit the business actually generated from selling things to people.

This report is about the gap between those two companies, and about what the market is paying for each of them. The question is not whether Alphabet is executing — it demonstrably is, and the evidence is stronger than the sell-side summary suggests. The question is what remains of Alphabet's earnings once the marks are removed, and whether $345.02 per share leaves anything on the table for that residual.

Our answer, compressed: Alphabet is running the best operating play of its life at the exact moment its reported earnings stopped being a useful measure of it. The operating business is worth more than the market's scepticism about AI search disruption implies. The reported earnings are worth considerably less than the trailing multiple of 17.3 implies. And the capital now required to sustain the operating momentum has already consumed the entire free cash flow of the company, stopped the buyback dead, and turned Alphabet into a regular issuer of both debt and equity for the first time in its history as a public company.

Six findings follow from the primary record. Each is developed in full later; each is stated here with the number that supports it.

1. Google Cloud is the most important business in the company, and it is not being bought

Google Cloud grew 63% in Q1 2026 and 82% in Q2 2026. Revenue went from $12.3 billion to $24.8 billion year over year. What makes this more than a demand story is what happened to the margin at the same time: Google Cloud's operating margin went from 17.8% in Q1 2025 to 35.6% in Q2 2026. Operating income in the segment grew 4.5× on revenue that grew 2.2×. A business buying growth through price or through unprofitable capacity cannot do that. The segment now generates a fifth of group operating profit from a fifth of group revenue, and the revenue backlog — $513.9 billion of Google Cloud commitments at 30 June 2026, up from $155 billion three quarters earlier — is contracted rather than forecast.

There is a caveat attached to that backlog number and it matters. In the first quarter of 2026 Alphabet changed its definition of revenue backlog to include contracts with an original expected term of one year or less. Some of the growth in the series is therefore definitional rather than commercial, and the company has not quantified how much. We treat the backlog as directional evidence of demand, not as a contracted revenue schedule.

2. Reported earnings have decoupled from the operating business, and the mechanism is a mark

Alphabet holds a portfolio of equity stakes that it carries under the ASC 321 measurement alternative: at cost, adjusted upward when an observable transaction establishes a higher value for the same issuer. The carrying value of that portfolio was $124.3 billion at 30 June 2026 against a cumulative initial cost of $47.6 billion — a $77.7 billion markup that exists entirely because other people raised money at higher valuations. Separately, Alphabet holds $94.1 billion of SpaceX shares, classified as marketable equity securities following SpaceX's 2026 listing, of which $80.0 billion is subject to short-term transfer restrictions and $14.1 billion is restricted through the third quarter of 2027.

Because these are equity securities rather than equity-method investments, every revaluation runs through the income statement. In Q2 2026 the gain on equity securities was $99.0 billion pre-tax, adding $77.1 billion to net income and $6.26 to diluted earnings per share. In Q1 2026 the equivalent figure was $36.9 billion. Over the twelve months to June 2026 the non-operating line contributed $151.6 billion against operating income of $147.6 billion.

3. The capital cycle has turned free cash flow negative, and it is accelerating

Alphabet generated $39.1 billion of operating cash flow in Q2 2026 and spent $44.9 billion on property and equipment. Free cash flow was −$5.9 billion — the first negative quarter since the company listed in 2004. Capital expenditure has gone from $32.3 billion in 2023 to $52.5 billion in 2024 to $91.4 billion in 2025, and is guided to $195–205 billion in 2026, a range raised twice during the year. The CFO has said 2027 will "significantly exceed" 2026.

This is a deliberate choice and management has been explicit about the rationale: demand is contracted, capacity is scarce, and the company will not have enough of it. The bear response is equally explicit and we take it seriously: the depreciation on $200 billion of annual capital expenditure arrives immediately, while the revenue it is built to serve — particularly TPU system sales — is largely a 2027 recognition event. Depreciation of property and equipment already rose 42% year over year in Q2 2026, to $7.1 billion.

4. Alphabet is now a borrower and an issuer, and the buyback is gone

Long-term debt went from $46.5 billion at 31 December 2025 to $98.2 billion at 30 June 2026. In June 2026 the company raised $49.6 billion of equity across Class A, Class C and mandatory convertible preferred stock, and authorised an at-the-market programme for a further $40.0 billion of stock, of which nothing had been sold. Share repurchases, which ran at $61.5 billion in 2023 and $62.2 billion in 2024, fell to $45.7 billion in 2025 and to zero in the first half of 2026 — a zero in Q2 2026 against $13.2 billion in the same quarter a year earlier.

A company that stops buying its own stock, triples its debt and sells equity while its shares trade below their May high is telling you something about the relative return on the marginal dollar inside the business versus outside it. That is not a criticism of the decision. It is a statement about what the decision implies for the cash return to shareholders over the next three years, which is materially lower than it was.

5. On operating earnings the shares are not cheap, and the trailing multiple is an artefact

The trailing price-to-earnings ratio of 17.3 is the single most misleading number attached to this security. It is computed on trailing diluted earnings per share of $19.93, of which the great majority is unrealised equity marks. On our estimate of clean FY2026 earnings — operating income of roughly $164 billion, taxed at 19.5%, on 12.4 billion diluted shares — Alphabet earns approximately $10.65 per share, which puts the shares at 32.4× 2026 clean earnings and 27.2× our 2027 estimate. Forward P/E exceeding trailing P/E is the tell: it happens when the trailing figure contains something the forward figure does not, and here that something is $99 billion of paper gains.

6. Every divestiture has failed, and the regulatory cost is nonetheless compounding

On 2 September 2026 Judge Leonie Brinkema rejected all three of the Department of Justice's structural remedies in the ad-technology case, declining to order the divestiture of AdX or the open-sourcing of the DFP auction logic. In the search-distribution case Judge Amit Mehta had already refused Chrome and Android divestiture in September 2025. The breakup thesis is, for now, dead in the United States. It is not dead in Europe: on 23 July 2026 the Commission fined Google €890 million under the Digital Markets Act in two separate decisions, the same month Alphabet paid $5.2 billion on the Android decision the EU courts upheld on 2 July 2026.

Conduct remedies are slower and less dramatic than breakups, and they are not free. Default payments to distribution partners — historically the most valuable single input into Google's search economics — are the target of the Mehta remedy. The honest position is that the market has stopped pricing structural risk at the moment the structural risk was replaced by something harder to model and harder to reverse.

02 — Company & business modelThree businesses, one balance sheet, and a fourth that is not a business at all

Alphabet reports three segments. The first, Google Services, is advertising and the consumer products that surround it: Search, YouTube, the Google Network of third-party placements, Android, Chrome, Maps, Play, devices, and a subscription bundle of Google One, YouTube Premium, YouTube TV and NFL Sunday Ticket. It generated $94.5 billion of revenue in Q2 2026, 78.9% of the total, and $39.5 billion of operating income, 96.9% of the group's profit from segments before unallocated costs.

The second, Google Cloud, sells infrastructure, platform services, Workspace and — increasingly and importantly — TPU systems, the company's own AI accelerators, which appear in the segment's product revenue line. It generated $24.8 billion in the quarter, 20.7% of revenue and 21.6% of segment operating profit. Three years ago it was 9% of revenue and loss-making.

The third, Other Bets, is a portfolio: Waymo in autonomous driving, Wing in drone delivery, Verily in life sciences, Isomorphic Labs in AI drug discovery, and X, the moonshot factory. It generated $382 million of revenue and lost $1.8 billion.

The fourth thing is not a segment at all and does not appear on any of the three. It is Alphabet's investment portfolio — SpaceX, a large stake in a private company Alphabet declines to name but which third-party reporting and the arithmetic both identify as Anthropic, and a tail of smaller positions. It generated no revenue, employs no one, and contributed $99.0 billion of pre-tax income in the June quarter. It is the largest single source of Alphabet's reported earnings growth, and it is invisible in the segment disclosure.

The structure matters because the four parts have almost nothing in common as investments. Google Services is a mature, high-margin, cash-generative advertising franchise with a competitive moat that is being questioned rather than broken. Google Cloud is a capital-hungry infrastructure business in the middle of a growth curve that resembles AWS circa 2016, funded by the cash the first business generates. Other Bets is venture capital inside a public company. The investment portfolio is a closed-end fund that happens to be attached to a search engine, marked to market through the income statement because of an accounting standard rather than a management decision.

How the money actually moves

Advertising revenue enters through three doors. Search & other — the largest, at $63.3 billion in Q2 2026 — is paid mostly on a cost-per-click basis by advertisers bidding in an auction against a query. YouTube ads, at $11.1 billion, is sold on a mix of auction and reservation. Google Network, at $7.3 billion and shrinking, is the remnant of the AdSense business that places ads on other people's websites; Alphabet pays the publishers a revenue share, which appears as traffic acquisition cost.

The cost structure of Search is unusual and worth stating plainly, because it is the source of the entire regulatory case against the company. Alphabet pays distribution partners — Apple foremost among them, but also browser vendors and carriers — to make Google the default search engine. Those payments sit inside traffic acquisition costs, which were $16.2 billion in Q2 2026, or 19.8% of Google advertising revenue. The Department of Justice proved at trial that these payments were the mechanism by which Google maintained its search monopoly; Judge Mehta's remedy targets them directly. Their size has never been separately disclosed, and the reported estimate of roughly $20 billion a year for the Apple agreement alone is a court figure, not a company one.

On the other side, Google Cloud is a straightforward consumption business: customers pay for compute, storage, network and increasingly for AI inference, metered by usage. Its economics are dominated by two variables — the utilisation of capital already deployed, and the unit cost of the silicon underneath. Alphabet controls the second of those to a degree no competitor does, because it designs its own chips. That is the subject of section 13.

What Alphabet is not

Two things are commonly asserted about Alphabet that the disclosure does not support, and both matter for valuation. The first is that Google is a search company. It is not, in revenue terms: Google Cloud is now 20.7% of revenue and growing at 82% while the rest of the company grows at 15%, which means the segment mix is shifting by roughly two percentage points of revenue per quarter. On the current trajectory Google Cloud is a third of Alphabet by 2028.

The second is that Alphabet's AI spending is an expense. It is not, in the accounting sense that matters most: the majority of it is capital expenditure that becomes an asset and is depreciated over five to six years for servers and longer for buildings. That treatment flatters near-term earnings and defers the cost, which is a legitimate accounting outcome rather than a manipulation — but it means the operating margin of 34% reported today is being supported by depreciation that has not yet arrived.

03 — The financial recordFour years, and the year the two lines crossed

The cleanest way to see what has happened to Alphabet's income statement is to plot two lines that ought to move together and watch them separate. From 2021 through 2024, net income tracked operating income with a stable, modest discount reflecting interest income, a small tax rate differential and occasional one-off charges. In 2025 the lines crossed. Over the twelve months to 30 June 2026 the gap is $96.6 billion, and it is widening by roughly $30 billion a quarter.

The table below sets out the four-year record. Read the last two rows together: operating income grew 75% from 2023 to the trailing twelve months, while net income grew 231%.

Three observations follow from the table, and they set up the rest of this report.

The operating margin has moved, but only in 2026

Alphabet's consolidated operating margin sat between 27% and 32% for three years and then jumped to 36.1% in Q1 2026 and 34.0% in Q2 2026. That is a real improvement, and it is worth understanding where it came from, because not all of it is durable.

Roughly half of the 2026 improvement is mix: Google Cloud went from 15% of revenue in Q2 2025 to 20.7% in Q2 2026, and its margin went from 20.7% to 35.6%, so the segment contributes disproportionately more profit per dollar of revenue than it did. The other half is operating leverage on a fixed cost base in Google Services, where revenue grew 15% while cost of revenues grew 18% — but research and development grew 32%, sales and marketing grew 18% and general and administrative grew 24%, all faster than revenue. The leverage is not in the expense lines. It is in the mix.

That distinction matters for what comes next. Mix-driven margin expansion reverses if the growing segment's margin stops expanding, and Google Cloud's margin is about to meet a depreciation schedule built on $200 billion a year of capital expenditure. Management already flagged that greater use of third-party capacity will put "modest pressure" on the Cloud margin in Q3 2026.

The non-operating line is not a rounding error

Other income (expense), net was $1.4 billion in 2023 and $151.6 billion over the trailing twelve months. The composition, from the 10-Q, is almost entirely gains on equity securities: $99.0 billion in Q2 2026, $36.9 billion in Q1 2026, and smaller amounts in the preceding two quarters. Interest income was $1.4 billion in the quarter and interest expense $1.3 billion, so the company's financing position is now roughly neutral — a change from the years when Alphabet earned a meaningful return on its cash pile.

The Q2 2026 disclosure also identifies $2.1 billion of legal charges relating to a Stockholm Patent and Market Court decision in a private action brought by PriceRunner: $1.5 billion of principal damages accrued in general and administrative expenses within Google Services, and $581 million of interest and costs in other income (expense), net. This is the kind of item that, three years ago, would have been the largest non-operating item in the quarter. In Q2 2026 it is 2% of the line.

The cash flow statement is where the story changes

Revenue and earnings can be argued about. Cash cannot. Alphabet's operating cash flow grew from $101.7 billion in 2023 to $164.7 billion in 2025 — a genuine, high-quality compounding of the franchise. But capital expenditure grew faster, from $32.3 billion to $91.4 billion, and then more than doubled again in the trailing twelve months to $132.4 billion. Free cash flow, which was $69.5 billion in 2023 and $73.3 billion in 2025, fell to $53.3 billion on a trailing basis and turned negative in the June quarter.

This is the arithmetic that governs everything in sections 10, 11 and 12. Alphabet is not short of money — it has $242.5 billion of cash and marketable securities and a AAA-equivalent balance sheet. It is short of free money, which is a different problem and one that has already changed the company's capital allocation behaviour.

04 — Segment deep diveGoogle Services: the franchise that was supposed to break

The central question about Alphabet's largest segment has been the same for two years: does generative AI destroy the economics of search? The bear case is intuitive and well documented. An assistant that answers a question directly does not send the user to a website, does not serve a search result page full of ads, and does not generate a click. If AI Overviews and AI Mode intercept enough queries, the click economy that funds the entire open web — and Google's advertising revenue with it — erodes.

Two years into the transition, the revenue line says this has not happened yet. Google Search & other grew 17% in Q2 2026 and 19% in Q1 2026 — the two fastest rates since 2022, both recorded after the global rollout of AI Overviews and AI Mode. The segment's operating margin, at 41.8% in Q2 2026, is roughly where it has been for three years.

That is a genuinely surprising result and it deserves to be taken seriously rather than explained away. It is also not the whole picture, because the same two years produced a large body of third-party evidence that the underlying click economy is deteriorating. Zero-click searches are estimated at roughly 68% in early 2026, up from about 60% in 2024. Publisher referral traffic from Google search has fallen substantially and was still falling in September 2026. Organic click-through rates on queries that trigger an AI Overview have been measured as materially lower than before the feature existed.

The two sets of facts are not reconcilable from public disclosure, and the reason is specific: Alphabet does not disclose query volume, click volume, monetised click counts, or revenue per query. It discloses revenue, and revenue is growing. Anyone — bull or bear — who quotes a "search revenue per query trend" is inferring it from something else.

Three ways both things can be true

There are at least three mechanisms that would allow declining clicks and rising revenue to coexist, and it is worth being explicit that we cannot distinguish between them with the available data.

First, AI Mode queries may be more commercially valuable. A user who asks an assistant a follow-up question about a product is further down the purchase funnel than a user who typed two keywords. If AI Mode queries are longer, more specific and more commercial, revenue per query can rise even as the number of clicks per query falls. Alphabet has said AI Mode drives incremental query growth and that the cost of an AI Mode response is at its lowest level since launch. Both statements are consistent with this mechanism and neither proves it.

Second, the loss may be concentrated in queries that were never monetised well. Informational queries — definitions, facts, recipes — were always the least valuable to advertisers. If AI Overviews absorb those and leave commercial queries to the traditional result page, the revenue effect is much smaller than the click effect. This is the most plausible explanation and it is also the most hopeful one for Alphabet.

Third, the growth may be coming from somewhere other than query economics. Search & other includes properties beyond the search box — Maps, Discover, and increasingly AI Mode itself — and Alphabet has been adding advertising inventory to AI surfaces. If the company is monetising the same users through more surfaces rather than more valuable queries, the growth is real but the durability question is unchanged.

What we can say with confidence is narrower than either camp would like. The revenue line has not broken, the margin has not broken, and the segment is generating $39.5 billion of operating income a quarter on $94.5 billion of revenue. But the segment's margin fell 350 basis points sequentially in Q2 2026 even as revenue grew 15%, and the cost of serving AI answers — inference — sits inside cost of revenues rather than in research and development, which means it is a permanent variable cost rather than a fixed investment. The economics of an AI answer and the economics of a ten-blue-links result page are not the same, and Alphabet has not yet had to demonstrate which one it is running.

What we watch

Three disclosures would resolve most of the uncertainty, and none of them currently exists. Revenue per query, if disclosed, would settle the mechanism question immediately. Monetised click counts would do the same. A segment-level disclosure of inference cost — even as a percentage of Google Services cost of revenues — would tell us whether AI answers carry a structurally lower gross margin than the result page they replace. Until then, the most informative proxy available is the Google Services operating margin, and the most informative thing that can happen to it is a sustained decline that is not explained by a fine or a charge. In Q2 2026 it declined 350 basis points. That is one quarter, not a trend, and we are not going to treat it as one.

The other item worth watching is traffic acquisition cost. At 19.8% of advertising revenue in Q2 2026, TAC is the line that a conduct remedy attacks. If the Mehta remedy succeeds in reducing or eliminating default payments, TAC falls and Google Services margin rises — the remedy would, in the near term, be accretive to earnings while destroying the distribution advantage that justified the payments. That is an unusual and under-discussed feature of the case.

05 — Segment deep diveYouTube and subscriptions: the second franchise nobody values separately

YouTube is the largest video platform in the world and it has never had a valuation attached to it. That is a deliberate outcome of Alphabet's reporting structure: YouTube's advertising revenue is a single line inside Google Services, its subscription revenue is buried inside "Google subscriptions, platforms, and devices," and its costs — content, infrastructure, and the revenue share paid to creators — are not disclosed at all. The company tells us the advertising number because it has to, and nothing else.

What we do know is that in 2025, revenue across YouTube advertising and subscriptions exceeded $60 billion for the first time, and that Alphabet had more than 325 million paid subscriptions across consumer services at year-end 2025, led by Google One and YouTube Premium. YouTube advertising alone was $40.4 billion in 2025 and is running at $42.6 billion on a trailing basis. If YouTube were a standalone company it would be among the twenty largest media businesses in the world.

The growth rate is the interesting part, and it is not flattering. YouTube advertising grew 13% in Q2 2026, 11% in Q1 2026, 9% in Q4 2025. Search & other grew 17%, 19% and 17% over the same three quarters. YouTube is growing more slowly than Search, which is the opposite of what the "AI kills search" thesis would predict, and it is worth asking why.

The short-form problem

The answer is almost certainly short-form video. YouTube Shorts is the platform's answer to TikTok, and it is enormously successful on engagement metrics and much less successful on monetisation metrics. Short-form inventory monetises at a lower rate per hour watched than long-form, because the ad load is lighter, the sessions are shorter, and the brand-safety premium that long-form commands does not fully transfer. As Shorts grows as a share of total watch time, the blended revenue per hour falls even as total watch time rises.

This is not a crisis — 13% growth on a $42.6 billion run-rate is an excellent business. But it does mean YouTube is not the segment that will offset a search deceleration if one arrives. It is growing slower than the segment it would need to defend.

Subscriptions are the more interesting story

The "Google subscriptions, platforms, and devices" line was $12.9 billion in Q2 2026, up 15% year over year, and $48.0 billion for 2025. This is a genuinely different kind of revenue from advertising: recurring, contracted, and less exposed to the advertising cycle. It includes Google One, YouTube Premium and Music, YouTube TV, NFL Sunday Ticket, Play store fees, and device sales.

Growth of 15% on a $48 billion base is a meaningful achievement and the line has been quietly compounding faster than advertising for several years. It also carries a strategic benefit that does not show up in the income statement: a subscriber relationship is the distribution channel through which Gemini reaches consumers without going through an app store. Alphabet disclosed 950 million Gemini app monthly active users in Q2 2026, and the paid subscription base is the part of that funnel where willingness to pay has already been demonstrated.

The caveat is that Alphabet does not disaggregate this line, and the components have very different economics. Device sales are low-margin hardware. Play store fees are nearly pure margin and are under sustained antitrust attack — the $700 million settlement with US state attorneys general received final approval in August 2026. NFL Sunday Ticket is a licensed content cost that has generated litigation of its own. Fifteen percent growth on a blended line can conceal a subscription business growing at 25% and a hardware business going backwards.

Why it matters for the whole

The reason to spend time on YouTube and subscriptions in a report about earnings quality is that they are the parts of Alphabet least exposed to the two risks that dominate the bear case. They are not search, so the AI Overviews mechanism does not apply to them in the same way — YouTube already answers the question "what should I watch" without sending the user anywhere. And they are recurring revenue, so a deterioration in advertising spend does not translate directly into a deterioration in the top line.

Together they are roughly $60 billion of annual revenue growing at low-to-mid teens, inside a company whose valuation is dominated by an argument about search. That is an under-appreciated floor, and it is one of the reasons we do not accept the strongest version of the bear case.

06 — Segment deep diveGoogle Cloud: 82% growth, and the margin says it is real

Nothing else in this report is as important as this section, and nothing else in the quarter was as surprising. Google Cloud grew 82% in Q2 2026, from $13.6 billion to $24.8 billion. In Q1 2026 it grew 63%. In Q4 2025 it grew 48%. Three quarters ago it grew 34%. On a business that passed a $70 billion annual run-rate at the end of 2025, that trajectory has no recent precedent in enterprise infrastructure.

Growth rates of this shape invite the suspicion that they are being bought — that Alphabet is discounting compute, or taking loss-making capacity deals, or pulling forward revenue from customers who would have arrived anyway. The margin answers that question, and the answer is no.

Google Cloud's operating margin went from 17.8% in Q1 2025 to 35.6% in Q2 2026. Operating income went from $2,177 million to $8,814 million — a 4.5× increase while revenue grew 2.2×. Incremental revenue has been coming in at a higher margin than the existing business, which is the definition of a business scaling into its cost base rather than one buying its way into a market. A company discounting to win share cannot expand margin by 18.5 percentage points over eight quarters while growing 82%.

Management attributes the expansion principally to a higher mix of in-house silicon — TPU for training and inference, Graviton-equivalent custom Arm cores for general compute — which carries a lower unit cost than the merchant alternative. The margin line is the evidence for that claim, and it is the strongest single argument in the bull case.

The backlog, and the definition change inside it

Google Cloud's revenue backlog — Alphabet's term for remaining performance obligations — was $513.9 billion at 30 June 2026, against $519.5 billion for the company as a whole. It increased by more than $50 billion sequentially, from roughly $460 billion at 31 March 2026. Three quarters earlier, at 30 September 2025, it was $155 billion. Just over half of the backlog is expected to be recognised as revenue within 24 months, with the remainder thereafter.

We have to attach a caveat, and it is not a small one. In the first quarter of 2026, Alphabet changed its definition of revenue backlog to include contracts with an original expected term of one year or less. The company disclosed the change in the 10-Q; it did not quantify the effect. That means the jump from $155 billion to $460 billion over two quarters contains an unknown amount of definitional expansion, and the disclosed figure is not comparable to the pre-2026 series without adjustment. We treat the backlog as evidence of demand — which it clearly is — rather than as a contracted revenue schedule.

Supply, not demand, is the constraint

Management's consistent message through 2026 has been that the business is supply-constrained. The CFO said on the Q2 2026 call that demand continues to outpace supply across the industry, and that the company would not have enough capacity in 2026 or 2027 to serve the demand it can already see. That is the stated justification for the capital expenditure programme, and it is a defensible one: if you cannot serve contracted demand, the rational response is to build.

It also creates a specific and under-discussed risk. A supply-constrained business that is building as fast as it can will show excellent margins, because every unit of capacity it brings online is immediately sold. The test of the Cloud margin is not what it does while supply is scarce. It is what it does in the first quarter when supply catches up with demand, and whether the capacity built at 2026 prices can be sold at 2028 prices. Alphabet has not been in that position in the AI era, and neither has anyone else.

On the current numbers, Google Cloud is annualising at roughly $99 billion and generating about $35 billion of annualised operating income. It is the fastest-growing large business in the company by a wide margin and the second-largest contributor to group operating profit. It is also consuming the majority of the capital expenditure, and its margin is the line most exposed to the depreciation that follows. Both things are true and the valuation in section 20 has to hold them at the same time.

07 — Segment deep diveOther Bets: Waymo is now the only one that matters

Other Bets generated $382 million of revenue in Q2 2026 and lost $1,799 million. On any financial reading it is immaterial: 0.3% of revenue, and a loss that rounds to nothing against a $40.8 billion operating profit. It is nonetheless worth a section, because it contains Waymo, and Waymo has quietly become one of the most valuable autonomous-vehicle businesses in the world.

The disclosure we have is thin and mostly qualitative. Waymo disclosed approximately 500,000 paid weekly rides in Q1 2026 — the last hard operational figure the company has given — and reached 14 US cities by 1 September 2026 after opening Denver, San Diego and Tampa. It operates roughly 4,000 vehicles and has stated a target of one million weekly rides by the end of 2026. London and Tokyo were announced in June 2026 as the first international markets.

In February 2026, Waymo announced an investment round of $16.0 billion, the significant majority of it funded by Alphabet, at a reported valuation of $126 billion. For the quarter ended 31 December 2025, Alphabet recognised a $2.1 billion employee compensation charge based on estimated stock valuation — a charge that runs through research and development and is the single largest reason Other Bets' loss widened to $3.6 billion that quarter. That is a genuinely informative number: it is Alphabet telling us, through its own compensation accounting, roughly what it thinks its own employees' Waymo equity is worth.

The safety record is the growth constraint

Waymo's driving record is, on the published data, substantially better than that of human drivers on a per-mile basis. That is not the binding constraint. The binding constraint is that every incident involving a robotaxi is a national news event, and the incident cadence has picked up. The National Highway Traffic Safety Administration opened a probe after a Waymo struck a child near a school in Santa Monica in January 2026. There was a voluntary recall of 3,791 vehicles in May 2026 for a flooded-road defect, a further recall in June 2026, and an active National Transportation Safety Board investigation into Waymo's behaviour around school buses in Austin.

None of these individually threatens the business. Collectively they define the speed at which it can expand, because each one triggers a regulator, a city council and a news cycle. The pattern in autonomous vehicles has been consistent for a decade: the technology arrives faster than the permitting does. Waymo's 14-city footprint is a permitting outcome as much as a technical one, and the international expansion into London and Tokyo will be slower than the domestic one.

Wing, Verily, Isomorphic and X

The rest of the portfolio is small and mostly not disclosed. Wing has passed one million home deliveries through partnerships with Walmart, DoorDash and Papa Johns. Isomorphic Labs raised more than $2 billion. Verily and X received no material disclosure in the quarter. Combined they are unlikely to be material to a $4.2 trillion company within the forecast horizon, and we do not ascribe value to them in our base case.

The correct way to think about Other Bets financially is as a call option purchased with a known annual premium. The premium is roughly $7 billion a year of operating losses plus the capital Alphabet puts in — $16 billion to Waymo in a single round in February 2026. The option is on autonomous transportation becoming a large market and Waymo winning it. If it works, it is worth more than the premium by an order of magnitude. If it does not, the losses are absorbed without difficulty. What it is not is a business that will change Alphabet's earnings within the horizon of this report, and any valuation that attributes meaningful near-term earnings to Other Bets is attributing something the segment has never produced.

08 — The equity marks$47.6 billion in, $124.3 billion on the books, and no ownership percentage disclosed

This is the section that changes the report. Alphabet's reported earnings are, to a degree that most summaries of the company do not convey, a function of its investment portfolio rather than its operating businesses. Understanding the mechanism is a prerequisite for valuing anything else.

The accounting standard is ASC 321. Equity securities with readily determinable fair values are carried at fair value with changes recognised in income. Equity securities without readily determinable fair values — private company stakes — may be carried under the measurement alternative: at cost, less impairment, adjusted upward or downward on observable transactions for identical or similar instruments of the same issuer. That last clause is the whole mechanism. When a private company Alphabet has invested in raises money at a higher valuation, Alphabet does not wait for an exit. It marks its own position up, and the gain goes through the income statement in the quarter the round closes.

The numbers from the Q2 2026 10-Q are unambiguous. At 30 June 2026 Alphabet's measurement-alternative portfolio had a total initial cost of $47,642 million, cumulative upward adjustments of $85,732 million, cumulative downward adjustments including impairments of $9,115 million, and a carrying value of $124,259 million. Twelve months earlier that portfolio — combined with the $7.2 billion of equity-method investments that sit alongside it — was $68.7 billion on the balance sheet, and the year before that it was smaller still.

The 10-Q states that as of 30 June 2026 this portfolio "primarily consists of our investment in a private company." Alphabet does not name it and does not disclose its ownership percentage. The arithmetic and third-party reporting both point to Anthropic, in which Alphabet first invested $300 million for a stake in April 2023 and into which it has since committed as much as $40 billion, including a $10 billion up-front tranche announced in April 2026 with the remainder milestone-based. We treat the identification as well-founded but not company-confirmed, and we do not rely on the ownership percentage because it has never been disclosed.

SpaceX is the other half of the position

SpaceX is no longer a private company in the accounting sense. Following its 2026 listing, Alphabet's holding is classified as marketable equity securities and carried at fair value: $94.1 billion at 30 June 2026, comprising $80.0 billion of shares subject to short-term restrictions on the ability to sell and $14.1 billion subject to long-term restrictions through the third quarter of 2027. Alphabet's initial investment was $900 million in January 2015 at a valuation of roughly $12 billion.

Taken together, the two positions are carried at approximately $218 billion. That is 34% of Alphabet's $640.5 billion of book equity, and it is more than the entire book value of the company as recently as 2023.

What the marks did to the income statement

In Q2 2026, other income (expense), net was $97,983 million. Within it, the gain on equity securities was $99,031 million. The 10-Q's footnote is precise about the effect: the gain "increased the provision for income tax, net income, and diluted net income per common share by $21.9 billion, $77.1 billion, and $6.26, respectively." In Q1 2026 the equivalent gain was $36.9 billion. The gross unrealised gain on measurement-alternative securities alone was $77,544 million in Q2 2026 and $114,161 million in the first half.

The half-year cash flow statement makes the same point in a different place: "loss (gain) on debt and equity securities, net" is a $(135,803) million adjustment to reconcile net income back to operating cash flow. In other words, $135.8 billion of the $174.8 billion of net income reported in the first half was a non-cash gain on securities, removed before arriving at the operating cash the business actually produced.

The asymmetry nobody discusses

There is a feature of this accounting that receives far less attention than the gains, and it is the reason we do not simply add the stake value to our target and move on. The marks are asymmetric in timing and symmetric in direction. When Anthropic or SpaceX raises money at a higher valuation, Alphabet books a gain immediately, because the observable transaction is a raise. When the same company's valuation falls — or when the private funding market closes — there may be no observable transaction at all, and the carrying value stays where it was until an impairment test produces a writedown. Gains arrive quarterly and visibly. Losses arrive late and in lumps.

SpaceX mitigates this, because it is now listed and its mark moves with a public price every quarter in both directions. The private position does not. In a market where AI laboratory valuations have compounded at a rate that has no historical parallel — from $61.5 billion in March 2025 to a reported $350 billion in April 2026 for Anthropic, by third-party account — the probability of a downward revaluation at some point in the next three years is not trivial, and the accounting will not deliver it gradually.

09 — Quality of earningsThree numbers that do not appear in the headline

Reported earnings per share of $9.11 is the number that was reported. It is worth restating what it is made of, because the composition is more informative than the level.

Of the $9.11, $6.26 is attributable to the net gain on equity securities after tax — a figure the company discloses directly. That leaves $2.85 from everything else, including operating income, interest, other non-operating items and a tax rate that is itself affected by the mark. On the same basis, Q1 2026's reported $5.11 contained approximately $2.35 from a $36.9 billion equity gain, leaving $2.76.

Two consecutive quarters at roughly $2.80 of clean earnings is a useful anchor. Annualised and grown modestly, it implies something in the region of $11.50 to $12.00 of clean earnings per share for 2026 — which, at $345.02, is a multiple of approximately 29×. That is the number a shareholder is actually paying, and it is not the number that appears in any published screen.

Number one: the tax rate is not a tax rate

Alphabet's effective tax rate was 19.1% in Q2 2026, up from 16.9% in Q2 2025. The 10-Q attributes the increase "primarily to higher pre-tax earnings from unrealised gains on equity securities that are deferred tax liabilities at the statutory tax rate." In plain terms: the marks are taxed at a higher rate than the operating business, and as the marks grow the blended rate drifts toward the statutory rate. The operating business's own tax rate is materially lower than 19.1%.

This has a second-order consequence worth flagging. Alphabet disclosed $12.1 billion of gross unrecognised tax benefits at 30 June 2026, of which $10.3 billion would affect the effective rate if recognised. It also noted that the OECD's 15% global minimum tax rules, including the "Side-by-Side Safe Harbor" introduced in January 2026, did not have a material effect in the first half but "could be affected" as more countries enact them. Tax has been a modest tailwind for large US technology companies for a decade. It is unlikely to remain one.

Number two: stock-based compensation is growing faster than revenue

Stock-based compensation was $7,957 million in Q2 2026 against $5,998 million in Q2 2025 — a 33% increase, against revenue growth of 24%. For the half year it was $14,708 million against $11,514 million. Alphabet had $59.1 billion of unrecognised compensation cost related to unvested restricted and performance stock units at 30 June 2026, to be recognised over a weighted-average period of 2.7 years.

The reason it is growing is visible in the grant data. Units unvested at 31 December 2025 had a weighted-average grant-date fair value of $159.75. Units granted in the first half of 2026 had a weighted-average grant-date fair value of $292.21. Alphabet is granting the same employees roughly the same number of units at nearly twice the accounting cost, because the share price roughly doubled. The expense is real, it is non-cash, and it is a transfer from shareholders to employees that does not appear in free cash flow.

The interaction with the buyback is the point. Through 2024, Alphabet repurchased enough stock ($62.2 billion) to more than offset dilution and reduce the share count. In the first half of 2026 it repurchased nothing, and diluted shares went from 12,198 million to 12,309 million year over year. Stock-based compensation of roughly $30 billion a year is now being absorbed by shareholders rather than bought back with the company's own cash.

Number three: the segment disclosure excludes the most important cost

Alphabet-level activities — the unallocated bucket that holds shared AI research and development, corporate initiatives and certain legal charges — lost $5,789 million in Q2 2026, up 72% year over year. That is where the cost of building Gemini sits. It is excluded from Google Services margin, excluded from Google Cloud margin, and shown only as a consolidated deduction. The segment margins in sections 4 and 6 are therefore not the full cost of the businesses they describe, and the Google Services margin of 41.8% would be roughly 6 percentage points lower if the AI research bill were allocated to it in proportion to revenue.

We are not arguing this is improper. Shared research genuinely is shared, and the accounting choice is defensible. But a reader comparing Google Services' 41.8% margin to Microsoft's productivity segment margin is comparing two different things, and the gap between them is partially a disclosure choice.

10 — Capital intensity$200 billion a year, and the depreciation that has not arrived yet

Alphabet will spend between $195 billion and $205 billion on property and equipment in 2026. To put that in context: it is more than the company spent in the four years from 2021 to 2024 combined. It is roughly 40% of 2026 revenue. It is larger than the gross domestic product of roughly 130 countries. And the CFO has said that 2027 will "significantly exceed" it.

The progression is the striking part. Capital expenditure was $32.3 billion in 2023, $52.5 billion in 2024 and $91.4 billion in 2025. Guidance for 2026 was set at $175–185 billion in February 2026, raised to $180–190 billion, and raised again to $195–205 billion in July. In a single year the company has revised its own capital plan upward by more than $20 billion, twice, and each revision has been justified by the same observation: demand exceeds supply.

Alphabet is not alone in this, and the comparison is the right frame. Amazon has guided to $220 billion for 2026. Microsoft and Meta, on Farstar estimates, will spend roughly $300 billion between them. The four companies together will deploy approximately $725 billion of capital in a single year on AI infrastructure, a 77% increase over 2025.

There are two ways to read that number and both are defensible. The bull reading is that four of the best-capitalised companies in history are responding to contracted demand they can see, that Alphabet's $513.9 billion of Google Cloud backlog is evidence rather than hope, and that the infrastructure being built is the input to a technology cycle that has years to run. The bear reading is that this is the largest coordinated capital deployment in corporate history, that it is being financed partly with debt and equity by companies that until recently needed neither, and that the depreciation will arrive on schedule whether or not the revenue does.

The depreciation arithmetic

The reason depreciation is the crux is that it is the mechanism by which capital expenditure becomes an earnings charge, and Alphabet's has only just started moving. Depreciation of property and equipment was $4,998 million in Q2 2025 and $7,104 million in Q2 2026 — a 42% increase. For the half year it was $13,586 million against $9,485 million. On an annualised basis the company is running at roughly $28 billion of depreciation against $321.2 billion of net property and equipment.

If capital expenditure settles at $200 billion a year and servers are depreciated over five to six years, the steady-state depreciation charge on each annual cohort is between $33 billion and $40 billion — before counting data-centre buildings, which are depreciated over much longer lives. The company is currently recognising roughly $28 billion in total. The gap between where depreciation is and where it is heading is the single largest mechanical headwind to Alphabet's operating margin over the next three years, and it is arithmetic rather than opinion.

FY2026 depreciation and operating cash flow are Farstar estimates. FY2026 free cash flow assumes the $200 billion midpoint of capex guidance and the first-half run-rate extrapolated.

The free cash flow line in that table is the one to watch. Alphabet generated $69.5 billion of free cash flow in 2023, $72.8 billion in 2024 and $73.3 billion in 2025 — remarkably stable, and a figure that supported both the dividend and $45–62 billion a year of buybacks. On our estimates the company will generate approximately negative $20 billion of free cash flow in 2026, which means the capital returns of the last three years were funded by a cash flow stream that no longer exists.

What management says the money buys

The company's defence is that this is contracted rather than speculative, and the backlog supports it. Management has also disclosed the composition: roughly 60% of Q2 2026 capital expenditure was servers and roughly 40% was data centres and networking. That mix matters because servers depreciate over five to six years and buildings over twenty to forty, so a capex programme weighted toward servers converts into earnings charges faster.

There is also a recognition-timing feature that deserves attention. Google Cloud generates product revenue "primarily from the sale of TPU systems," and the CFO said on the Q2 2026 call that the majority of TPU system sales revenue will be recognised in 2027. The cost of building the capacity to serve those sales is being incurred in 2026. That is not a problem in a present-value sense — a sale is a sale whenever it is recognised — but it does mean that 2026 operating margin is being flattered relative to 2027, and that the bulls and bears are looking at different years when they argue about whether the returns are there.

11 — Balance sheet & financingFrom a net-cash fortress to a regular issuer, in eighteen months

For most of its life as a public company, Alphabet's balance sheet was a curiosity: a business generating more cash than it could deploy, accumulating a securities portfolio, and carrying debt that was small enough to be a rounding error. That era ended in the second half of 2025 and the change has been abrupt.

Long-term debt was $46,547 million at 31 December 2025 and $98,165 million at 30 June 2026. In November 2025 Alphabet issued $24.8 billion of senior unsecured notes; in Q2 2026 it issued a further $20.3 billion. The CFO described the trajectory on the Q2 call as debt rising "from around $16 billion a year ago to a level approaching $100 billion." Interest expense went from $261 million in Q2 2025 to $1,278 million in Q2 2026 — a 5× increase, and one that will keep rising as the notes roll into full periods.

The equity raise is the more consequential decision

In June 2026 Alphabet raised $49.6 billion of net proceeds across Class A stock, Class C stock and a new 6.25% mandatory convertible preferred stock, sold in combination. The stated use of proceeds was "general corporate purposes, including capital expenditures to scale AI infrastructure and global compute." It also authorised an at-the-market programme to sell up to $40.0 billion of additional stock, on which nothing had been sold at the quarter end, with proceeds "primarily intended to be used to meet tax obligations associated with employee equity grants."

Alphabet had not issued equity to fund operations since its IPO in 2004. The decision to do so now, while the shares traded below their May 2026 high and while the company held $242.5 billion of cash and marketable securities, is the clearest available signal of how management is thinking about the capital cycle. A company does not sell equity unless it believes the equity is a cheaper source of funds than the alternatives — or unless it believes the scale of the opportunity is larger than its ability to fund it internally. Both readings are available and they are not mutually exclusive.

There is one further item on the asset side that deserves a mention because it is growing quickly and is easy to miss. Goodwill went from $33.4 billion at 31 December 2025 to $57.8 billion at 30 June 2026, and intangible assets from $1.3 billion to $9.1 billion. The cash flow statement shows $33.7 billion of "acquisitions, net of cash acquired, and purchases of intangible assets" in the first half. The great majority of that is the $32 billion acquisition of Wiz, the cloud and AI security platform, which closed on 11 March 2026 — the largest acquisition in Alphabet's history.

Wiz is strategically coherent: security is a genuine differentiator in enterprise cloud and Alphabet has been the third-place vendor that enterprises are least likely to make their primary cloud. Whether it earns its price is unknowable for several years, and Alphabet will not disaggregate its contribution. What is knowable is that it consumed a third of a year's free cash flow, at a moment when free cash flow had already gone negative, and that the $32 billion is now goodwill and intangibles rather than cash.

12 — Capital returnsBuybacks went from $62 billion a year to zero, and nobody announced it

The most consequential change in Alphabet's capital allocation is also the least discussed. Share repurchases, which had been a defining feature of the Alphabet investment case for a decade, went to zero in the first half of 2026.

The progression is stark. Alphabet repurchased $61.5 billion of stock in 2023 and $62.2 billion in 2024. In 2025 it repurchased $45.7 billion — already a 27% reduction. In Q1 2026 it repurchased nothing. In Q2 2026 it repurchased nothing, against $13.2 billion in the same quarter a year earlier. The company still has $69.5 billion of authorisation remaining; it simply is not using it.

There was no announcement, no explanation in the earnings release, and no question of a formal suspension. The buyback stopped because the cash that funded it was redirected, and the direction is disclosed in the same cash flow statement: purchases of property and equipment of $80.6 billion in the first half of 2026 against $39.6 billion in the first half of 2025.

The dividend has taken up some of the slack, and less than it appears

Alphabet introduced a dividend in 2024 at $0.60 per share for the year, raised it to $0.83 in 2025, and is paying $0.22 per quarter in 2026 — an annualised $0.88, a yield of approximately 0.26%. It also began paying $12.15 per quarter on the new mandatory convertible preferred stock, equivalent to roughly $0.60 per depositary share, and paid $86 million of preferred dividends in Q2 2026.

Total dividends paid were $7.4 billion in 2024, $10.0 billion in 2025 and $5.2 billion in the first half of 2026. Against a company that was returning $62 billion a year through buybacks two years ago, a $10 billion annual dividend is not a substitute. Total capital returned to shareholders fell from $69.6 billion in 2024 to $55.8 billion in 2025 and is running at roughly $10 billion annualised in 2026.

It is fair to note the counter-argument, which management has made implicitly rather than explicitly: if the incremental return on $200 billion of AI infrastructure exceeds the return on retiring stock at 32× clean earnings, then cutting the buyback is the value-maximising decision. That is a coherent position. It is also unfalsifiable for at least two years, and it requires the capital to earn a return that is not yet visible in any disclosed metric. The company's own decision to raise equity at the same time suggests it is not certain about the answer either.

13 — Custom siliconTPU is the most consequential thing Alphabet has built since Search

There is one strategic asset inside Alphabet that does not appear anywhere in the financial statements as a distinct item and that may be worth more than several of the businesses that do: the Tensor Processing Unit, Alphabet's family of custom AI accelerators. Understanding why it matters requires understanding what it changes about the company's economics.

The industry default for AI compute is to buy NVIDIA accelerators at merchant prices. That model works, and it is expensive. A hyperscaler buying merchant silicon pays a gross margin to the vendor on every unit, has limited ability to differentiate its offering at the hardware level, and is exposed to allocation and lead times it does not control. Alphabet designs its own chips, has them fabricated by TSMC through Broadcom, and deploys them in its own data centres. The saving on unit cost flows directly into Google Cloud's gross margin — which is precisely what the margin expansion in section 6 looks like.

The strategic question is whether Alphabet sells TPUs or uses them. The answer, at present, is that it does both but overwhelmingly the latter. TPUs reach external customers primarily through Google Cloud as an infrastructure offering rather than as merchant silicon sold for deployment elsewhere. That is a deliberate choice with a clear logic — the customer buys the chip, but they buy it inside a contract that also buys the network, the storage, the software stack and the managed services, and switching costs are correspondingly higher — and a clear cost, which is that it caps the addressable market for the silicon at the size of Google Cloud rather than the size of the AI accelerator market.

Anthropic is the proof point

The largest confirmed external commitment is Anthropic. Broadcom disclosed a $21 billion Anthropic TPU order in December 2025, including roughly $10 billion of Ironwood racks, and Anthropic publicly expanded its Google and Broadcom compute partnership on 6 April 2026, targeting approximately one million TPUs and one gigawatt of capacity. Alphabet agreed in April 2026 to invest up to $40 billion in Anthropic in cash and compute.

Read that sequence carefully, because it describes a structure that recurs throughout this industry: Alphabet invests in Anthropic, Anthropic commits to buy TPUs from Alphabet through Broadcom, and the capital returns to Alphabet as revenue. The revenue is real and the chips are real. But a portion of the demand is being financed by the supplier, which means the demand is not entirely independent evidence of market-clearing prices. The same structure appears in the NVIDIA report in this series from the other direction, and it is the most important thing to understand about AI infrastructure revenue in 2026.

14 — Competition IAI models: Google has the distribution, OpenAI has the users

Alphabet's position in frontier AI is genuinely strong and genuinely second. Both halves of that sentence are load-bearing, and the public debate tends to pick one.

On users, OpenAI leads. Sensor Tower estimates put ChatGPT at 46.4% of consumer AI assistant monthly active users in May 2026 — the first time it has fallen below half — with Gemini at 27.7% and Claude at 10.3%. The absolute figures from the same source are 1.1 billion monthly actives for ChatGPT, 662 million for Gemini and 245 million for Claude. Alphabet's own disclosure for Q2 2026 is 950 million Gemini app monthly active users, up from 750 million at the end of 2025 and 650 million in Q3 2025. The gap between 662 million and 950 million is definitional — app-only measurement versus the wider ecosystem — and both numbers are quoted by different sides of the debate. We use the direction rather than the level.

On capability, the picture is cyclical rather than structural, which is a meaningful change from eighteen months ago. Gemini 3 shipped in November 2025, Gemini 3.1 Pro in February 2026, and Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber in July 2026. Google has led on some public benchmarks within the same year that it has trailed on others. Gemini 3.5 Pro was still in testing as of mid-2026, which is a slip by the company's own cadence, and Gemini 4 pre-training has begun. The honest read is that no laboratory now holds a durable capability lead for more than a quarter.

Where Alphabet's advantage actually is

The durable advantage is not the model. It is the distribution and the full stack, and both are harder to replicate than a benchmark result. Gemini is inside the search box that a billion people use by default. It is inside Android, Chrome, Workspace and YouTube. It is now inside Siri, through an Apple agreement announced in January 2026 and reported at roughly $1 billion a year, which is a remarkable reversal for a company whose default-search payments to Apple are the subject of an antitrust remedy. Alphabet reported that Gemini Enterprise is used by nearly 90% of the Fortune 100, that roughly 500 Google Cloud customers each processed more than a trillion tokens in the past year, and that its models process approximately 22 billion API tokens per minute, up from 16 billion in Q1 2026 and 7 billion in Q3 2025.

That last series is the most underrated number in the AI debate. API token throughput quadrupling in three quarters is a measure of real, paid usage rather than user counts, and it is growing faster than any competitor's disclosed equivalent.

The criticisms that land

Three bear arguments on AI are worth taking seriously. The first is that Alphabet has lost senior AI talent at a rate that matters: Noam Shazeer left for OpenAI in June 2026, John Jumper left for Anthropic, and in the most significant change of the cycle, Demis Hassabis moved from chief executive of Google DeepMind to chair of Google DeepMind and chief scientist of Alphabet on 5 August 2026, with Jeff Dean departing and Koray Kavukcuoglu taking a new role. The reorganisation reads as a promotion of the scientific vision, and it is also the largest senior AI departure of the cycle. Organisational churn at that level is a leading indicator of strategic disagreement, and it is not visible in any financial metric.

The second is product quality. Alphabet faced a user backlash in June 2026 over increased hallucination after a June 3 app update, and a separate backlash in May 2026 that tightened safety filters had made the assistant unhelpfully cautious. Both are the ordinary friction of shipping at scale, and both are also the kind of thing that causes a user to switch. In a market where ChatGPT has 1.1 billion monthly users and switching costs are close to zero, quality regressions are more dangerous than they would be in enterprise software.

The third is monetisation, and it is the one we find most under-examined. Claude reportedly converts a meaningfully higher share of its users to paid subscriptions than either ChatGPT or Gemini. Alphabet's Gemini revenue is not disclosed separately, and neither is its cost. We know from the Alphabet-level activities line that shared AI research and development costs $5.8 billion a quarter and rising 72% a year. We do not know what the revenue against it is, and Alphabet's decision not to disclose it is itself informative.

The synthesis: Alphabet is very likely to be one of the two or three companies that matter in consumer AI, because it owns the distribution channels through which most of the world reaches a computer. It is not obviously going to be the company that captures the most value from it, because the model layer is competitive, the consumer price point is zero, and the enterprise price point is being contested by every large software vendor simultaneously. Being a probable survivor is not the same as being the winner, and the share price is closer to pricing the second than the first.

15 — Competition IIAdvertising: Google is about to lose a title it has held for two decades

Google has been the largest seller of digital advertising in the world since it passed Yahoo in 2004. eMarketer forecast in April 2026 that Meta will overtake it in 2026 — the first time that has happened. The forecast is a third-party estimate rather than a disclosed fact, and the margin is narrow, but the direction has been consistent for four years: Google's share of global digital advertising has drifted down from roughly 30% in 2021 to about 26.4% in 2026.

It is important to be precise about what is and is not happening, because "Google is losing advertising share" is used to imply several different things. Google's advertising revenue is growing — 14% year over year in Q2 2026, with Search & other at 17%. It is not shrinking. What is happening is that the total market is growing faster, because two categories are expanding that Google barely participates in.

The pattern in that table is the whole argument. Google is dominant in the category it invented and which the antitrust cases attacked. It is a strong second in social and video, an absent third in retail media, and it is losing the third-party network business outright — Google Network declined 1% in Q2 2026 and has now declined for five consecutive quarters, from $7,954 million in Q4 2024 to $7,303 million in Q2 2026.

The retail media gap is the one that compounds

Retail media is the fastest-growing category in advertising and it is built on something Google cannot manufacture: first-party purchase data. An advertiser buying a sponsored product placement on Amazon knows the ad was served to someone who bought the category last month. Google's search advertising infers intent from a query. The inference is very good and it is not the same thing, and as measurement standards tighten and privacy regulation reduces third-party signal, the gap widens rather than narrows.

Alphabet's response is to build first-party data through its own surfaces — YouTube watch history, Maps location, Play purchases, and the subscription base — and to use AI to extract more signal from less data. That is a real strategy. It is also the strategy every other advertising platform is running, and Google's starting position in commerce-adjacent data is weaker than Amazon's.

The offsetting factor is that AI raises the value of a query

The counter-argument, and it is a strong one, is that the advertising formats Google is shipping inside AI Mode are more valuable per impression than the ten-blue-links page they replace. An assistant that has a conversation about a purchase decision has more context about the user's intent than a search result page, and context is what advertisers pay for. Alphabet has said that AI Mode drives incremental query growth and that the cost of serving an AI Mode response is at its lowest since launch.

What we do not have is the number that would settle it: revenue per AI Mode session versus revenue per traditional search session. Until Alphabet discloses it, the advertising-share debate will continue to be argued with proxies. Our position is that the share loss is real and structural, that it is happening in categories where Google was never dominant, and that it does not by itself imply a decline in Google's advertising revenue — only a decline in its share of a market growing faster than it is.

16 — Competition IIICloud: the third-place vendor is the only one accelerating

The cloud market has had a stable shape for a decade: AWS first, Microsoft second, Google third and far behind. The shape has not changed. The velocity has, dramatically.

In Q2 2026, Google Cloud grew 82%. AWS grew 37%, which was itself the fastest rate that business has recorded in eighteen quarters, on a base of roughly $169 billion of annual run-rate revenue. Azure's growth has been reported in the high thirties. Google Cloud is growing at more than twice the rate of the market leader, on a base that is roughly a quarter of the size.

Share estimates are third-party (Synergy Research-derived) and two versions circulate for Q2 2026: one has AWS 31% / Azure 24% / Google 12%, another has AWS 28% / Microsoft 20% / Google 15%. The direction — Google gaining, Microsoft slipping — is consistent across both. AWS revenue is derived from the run-rate disclosed in the companion Amazon report in this series.

The strategic question is whether 82% growth is sustainable or whether it is a step change caused by a small number of very large AI contracts. We lean toward the second reading, and the reason is arithmetic: at $24.8 billion a quarter, Google Cloud would need to add roughly $20 billion of quarterly revenue year over year to sustain an 82% rate. The backlog supports several quarters of that, but the backlog is concentrated, and the concentration is in exactly the two customers whose own business models depend on continued capital raising.

What Alphabet has that the others do not

Two things, and they are related. The first is TPU, discussed in section 13: Alphabet is the only one of the three large clouds that designs its own AI accelerator at scale, which gives it a lower unit cost of inference and a degree of supply independence the others lack. The second is that Alphabet is the only one of the three whose own AI models are competitive at the frontier, which means it can sell an integrated offering — model, tooling, infrastructure — rather than renting compute to companies that then buy models from someone else.

The disadvantage is equally clear and it is the reason Google Cloud has been third for a decade: enterprise customers have already made their commitments. Multi-cloud is real, but the primary cloud is sticky, and the primary cloud is usually AWS or Azure because those relationships were established earlier. Alphabet's route into the enterprise has been AI — customers who need frontier models and cannot get enough capacity elsewhere will accept a second cloud relationship — and that route works precisely as long as capacity is scarce. If capacity stops being scarce, the urgency that brought those customers in stops being a reason to stay.

Microsoft's position deserves a note because it bears on Alphabet's. Microsoft's share is estimated to be slipping for the first time, and the reason is the same one that helped Alphabet: OpenAI's exclusivity with Microsoft ended on 27 April 2026, and OpenAI subsequently added AWS and Google Cloud as providers. Microsoft's loss of a captive customer is Alphabet's gain, and it is a gain that depends on a commercial arrangement rather than a technical advantage. Those are the least durable kind.

17 — The disruption questionSearch under AI: what would have to be true for the bear case to be right

This is the section that decides the valuation, and it is the one where the available evidence is genuinely insufficient to reach a conclusion. We are going to set out what the bear case requires, what the evidence shows so far, and where the burden of proof sits.

The mechanism of the bear case is straightforward. Search advertising is sold against clicks. A traditional result page produces clicks: paid clicks on the ads, organic clicks on the results. Both are monetisable — the first directly, the second by keeping publishers producing the content that gives people a reason to search. An AI answer produces no click. The user gets the answer, closes the tab, and the entire economic chain that the page supported — the ad auction, the publisher's ad revenue, the referral economy — never happens.

The evidence that this is occurring is substantial and comes almost entirely from outside Alphabet. Zero-click searches are estimated at roughly 68% in early 2026, up from about 60% in 2024. Publisher referral traffic from Google has fallen materially and was still falling in September 2026. Studies of click-through rates on queries that trigger an AI Overview have measured declines in organic clicks of more than half, with paid clicks falling further.

The evidence that it is not yet occurring at the revenue line is equally substantial and comes entirely from Alphabet. Search & other grew 17% in Q2 2026 and 19% in Q1 2026. Google Services operating margin was 41.8%, within its three-year range. Total advertising revenue grew 14%.

Reconciling the two

As set out in section 4, there are at least three mechanisms that would allow both sets of facts to be simultaneously true, and Alphabet's disclosure does not let us distinguish between them. But there is a fourth possibility that deserves to be stated because it is the least comfortable one: the revenue is lagging the traffic. Advertising contracts are annual, budgets are set in advance, and a large share of Google's search revenue is committed spend. If the underlying click economy deteriorated in 2025 and 2026, the revenue effect could be deferred into 2027 by the structure of the contracts. That would mean the revenue line is a lagging indicator, and its continued strength is not evidence against the mechanism — only evidence that the mechanism has not yet been transmitted.

We cannot prove this and we are not asserting it. We are noting that it is the one version of the bear case that the current revenue data cannot refute, because the data arrives too late.

Where the burden of proof sits

Our position is that the burden of proof has shifted. Two years ago the question was whether AI would disrupt search; today the question is whether the disruption has begun. Alphabet has now delivered five consecutive quarters of accelerating search growth through the entire rollout of AI Overviews and AI Mode. That is meaningful evidence and it should be weighted accordingly. But the company has also made it impossible to test the mechanism directly, by declining to disclose query volumes, click volumes or revenue per query, and by declining to disclose the cost of serving an AI answer.

A shareholder is therefore being asked to accept a thesis on the strength of the one number the company chooses to publish, while the numbers it declines to publish are the ones that would test it. That is not an accusation of bad faith — companies disclose what regulation requires and what they judge useful — but it is a real limitation on the confidence any analysis can have, and it should be reflected in the multiple rather than in the forecast.

18 — Cost structure & people198,933 employees, and the AI bill that sits outside every segment

Alphabet ended the second quarter of 2026 with 198,933 employees, up from 187,103 a year earlier and from 190,820 at the end of 2025. That is a 6.3% increase over twelve months, and it is a more complicated number than it looks, because it is the net of simultaneous hiring and rolling reductions.

The reductions in 2026 were not announced as a programme. Platforms and Devices offered buyouts in January; Cloud restructured in February; further Platforms and Devices cuts followed in March and April; the Global Business Organization and People Operations offered buyouts in mid-2026; and there were quieter reductions in Cloud, including in the Threat Intelligence Group and Mandiant, in July. Around 4,500 employees petitioned the chief executive for severance guarantees in July 2026. No official annual total has been disclosed, and third-party trackers estimate the reductions at roughly 1% to 2% of headcount. The company is hiring faster than it is cutting, and doing both continuously.

The financial consequence is visible in the expense lines. Research and development grew 32% year over year to $18.2 billion in Q2 2026, faster than the 24% growth in revenue. Sales and marketing grew 18% to $8.4 billion and general and administrative grew 24% to $6.5 billion. Cost of revenues grew 18% to $45.9 billion. Only cost of revenues grew slower than revenue, and only modestly.

The most informative single line in the segment disclosure is the one nobody quotes. Alphabet-level activities lost $5,789 million in Q2 2026, up 72% year over year from $3,372 million, and up 108% from $2,783 million in Q4 2024. This bucket holds the shared cost of developing Alphabet's general AI models — employee compensation and technical infrastructure usage for shared research — plus corporate initiatives and certain legal charges. It is excluded from every segment margin the company reports.

Stock-based compensation is the quiet transfer

Alphabet recognised $7,957 million of stock-based compensation in Q2 2026, up 33% year over year and growing faster than revenue for the second consecutive year. The unrecognised balance is $59.1 billion, to be recognised over 2.7 years. The driver, as set out in section 9, is that Alphabet grants units at a share price roughly twice what it was two years ago, so the same compensation policy produces twice the accounting expense.

The economics of this are unambiguous and frequently elided. Stock-based compensation is a real cost paid in real dilution. Through 2024, Alphabet's buyback was large enough to absorb it and reduce the share count. In the first half of 2026, with the buyback at zero, diluted shares rose from 12,198 million to 12,309 million year over year. A shareholder's claim on the business is being diluted by roughly 1% a year, and the mechanism that used to prevent it has been redirected to capital expenditure.

Productivity, honestly measured

Revenue per employee was approximately $2.27 million on a trailing basis in Q2 2026, against $2.04 million a year earlier — an 11% improvement. That is a good number and it is worth noting because the popular narrative about large technology companies is that headcount growth has outrun revenue growth. At Alphabet it has not, in the last year.

The caveat is that revenue per employee is a poor measure of productivity in a company that is simultaneously capitalising $200 billion of infrastructure. The labour input per dollar of AI capacity produced is falling, which is the point of the capital programme, and the productivity gain will show up in the capital line rather than the headcount line. What a shareholder should watch is not headcount but the ratio of operating income to total capital employed, which is the measure that captures both. On our numbers that ratio is deteriorating: operating income of approximately $164 billion against property and equipment of $321 billion and rising fast.

19 — Regulation & antitrustEvery breakup has failed. The meter has not stopped.

Two years ago the existential risk in the Alphabet investment case was structural: a court ordering the divestiture of Chrome, or Android, or the advertising exchange. As of September 2026, every structural remedy proposed by the United States government has been rejected, and the market has correctly repriced that risk. The mistake would be to conclude that the regulatory problem has gone away. It has changed shape.

The search case: liability, remedies, appeal

Judge Amit Mehta found in August 2024 that Google had illegally maintained a monopoly in general search services. The remedies opinion followed on 2 September 2025 and rejected divestiture of Chrome and Android in favour of behavioural relief. A 95-page opinion on 5 December 2025 finalised the judgment, and Judge Mehta largely sided with the government on two points that matter commercially: the sharing of Google's search index and user interaction data with competitors under licence, and the syndication of search and search-text advertising. He rejected Google's proposed veto rights over who could receive the data and its demand that syndicating rivals display results identically.

The order entered into force in early 2026 and a Technical Committee with the power to grant exceptions was stood up. The precise compliance dates are contested in secondary sources and we have not been able to verify them against the docket; the default-payment prohibition — the remedy that attacks the roughly $20 billion a year Alphabet pays Apple for default placement — is the economically material provision and readers should treat its operative date as unconfirmed.

Google filed a notice of appeal in January 2026 targeting the data-sharing mandate. The Department of Justice and a group of states cross-appealed on 3 February 2026, seeking Chrome divestiture and termination of the Apple default agreement. Google's opening brief was filed on 22 May 2026. No oral argument has been scheduled and briefing will likely run into 2027. The licence terms and privacy safeguards for data sharing were still not finalised as of May 2026.

The ad-technology case: the breakup that did not happen

Judge Leonie Brinkema found liability in April 2025 — that Google had illegally monopolised publisher advertising technology. On 2 September 2026 she issued the remedies order, and it rejected all three of the government's structural requests: the divestiture of AdX, the open-sourcing of the DFP final-auction logic, and a contingent DFP divestiture. Most of the behavioural remedies were accepted as modified. The memorandum opinion is sealed, so the reasoning and the precise content of the accepted remedies are not public.

The practical consequence is that AdX and DFP remain integrated and the take rate is unchanged. Parties had to move for redactions by 16 September 2026, and a single joint proposed final judgment is due by 2 October 2026, with unresolved provisions to be filed side by side. Google has signalled an appeal on liability since April 2025. Follow-on damages actions by OpenX, PubMatic, Magnite, Index Exchange, Raptive, Vox Media, The Atlantic and Teads, along with claims in the United Kingdom and Sweden, are unaffected by the remedy outcome and will proceed.

Europe is now the venue that matters

With structural remedies dead in the United States, the European Commission becomes the only regulator plausibly capable of breaking the company up. It has been active. On 5 September 2025 it issued an ad-technology decision under Article 102; the public version was released in January 2026 and structural remedies remain on the table, with Google's annulment action pending before the General Court. On 23 July 2026 the Commission fined Google €890 million in two separate Digital Markets Act decisions — €460 million for self-preferencing its own services in search results across shopping, hotels, transport and sports, and €430 million for anti-steering restrictions in Google Play. And on 2 July 2026 the EU courts upheld the 2018 Android decision, with $5.2 billion of fine and interest, previously accrued, paid that month.

The United Kingdom is running a parallel process. The Competition and Markets Authority opened a consultation on conduct requirements for Google general search and search advertising under the Digital Markets, Competition and Consumers regime on 28 January 2026. That is a forward-looking framework rather than a penalty, and it is the template that other jurisdictions are likely to follow.

20 — ValuationThe multiple depends entirely on which earnings you use, and only one of them is real

Alphabet's share price on 16 September 2026 was $345.02, against a market capitalisation of $4.22 trillion and a 52-week range of $235.84 to $408.61. The shares are roughly 15% below their May 2026 high and up 9% for the year to date. Sixty-three analysts carry an average rating of "Strong Buy" with a consensus target of $428.07, implying 24% upside.

The reported multiples are, in our view, close to meaningless. Trailing price-to-earnings is 17.3 on trailing diluted earnings per share of $19.93. That figure contains the equity marks. Enterprise value to EBITDA is 23.7, which is distorted in the same direction and by the same mechanism, since EBITDA includes the $151.6 billion of non-operating income while excluding depreciation that is rising 42% a year. Price to free cash flow is 79.2, which is the only one of the three that is arguably too high, and it is too high because free cash flow has collapsed rather than because the price has risen.

The single most useful diagnostic is this: forward price-to-earnings (25.9) is substantially higher than trailing price-to-earnings (17.3). That inversion happens when the trailing figure contains something the forward figure does not. Here it is $135.8 billion of non-cash securities gains in the first half alone.

Building clean earnings

Our valuation starts from operating income, applies a normalised tax rate, and ignores the equity marks in the earnings base entirely — treating the portfolio instead as a separately valued asset. That is the only structure that produces a comparable multiple.

FY2026E and FY2027E are Farstar estimates. FY2024 and FY2025 clean net income excludes other income (expense), net, and applies a 19.5% normalised tax rate to operating income; reported figures include it. The normalised rate is our assumption and is higher than the 17–19% Alphabet has reported on operating earnings historically, reflecting the OECD minimum tax framework.

Two things follow from that table. First, the shares trade at 32.4× our 2026 clean estimate and 27.2× our 2027 estimate — not 17.3× anything. Second, the clean earnings growth implied by the model is 25% in 2026 and 20% in 2027, which is a genuinely high-quality growth rate for a company of this size. The question is whether it justifies the multiple, and the answer depends on whether you believe the 2027 margin.

The margin is the whole argument

Our FY2027 operating margin assumption of 33.4% is only 60 basis points above 2026 and roughly flat against 2025. It assumes that Google Cloud's margin expansion offsets the depreciation drag from $200 billion of annual capital expenditure and the continued growth of the unallocated AI research bill. That is not a heroic assumption, and it is not a conservative one either. It is the assumption that carries the most risk in this model.

If the operating margin instead compresses to 28% by 2027 — the bear case, driven by depreciation arriving faster than Cloud margin can expand — FY2027 clean EPS falls to approximately $9.20, and the multiple at today's price becomes 37.5×. That is the arithmetic behind our bear scenario.

Valuing the portfolio separately

The equity stakes are carried at approximately $218 billion: $94.1 billion of SpaceX shares, $124.3 billion of measurement-alternative securities that primarily consist of one private company, and $7.2 billion of equity-method investments and other. Three discounts are appropriate. The SpaceX position is marketable but $94.1 billion of it is subject to transfer restrictions, of which $80.0 billion expire in the short term. The private position has no observable price except at funding rounds. And a realised gain would be taxed.

Applying a 30% discount for illiquidity and transfer restrictions, and a further 21% for tax, gives $120.5 billion of after-tax value, or approximately $9.72 per share. In our base case we add that to the operating value. We do not capitalise any further markup.

Scenario values

The bear case assumes search disruption transmits into revenue, Google Cloud growth halves, and depreciation compresses the margin. The bull case assumes Cloud sustains high growth into 2027, TPU economics continue to expand margin, and the private portfolio is revalued upward again. Both are plausible. The 2.7× spread between them is the honest statement of how much of this company's value depends on a question that cannot be resolved from the disclosure.

21 — The bull caseSix arguments we take seriously

1. Google Cloud has broken out, and the margin proves it is not being bought. 82% growth at a 35.6% operating margin, with operating income up 4.5× while revenue grew 2.2×, is the strongest evidence available that Alphabet has found a second engine. The segment now generates a fifth of group operating profit and is growing 3.4× faster than the rest of the company. If it sustains even half its current growth rate, it becomes a third of Alphabet by 2028.

2. The search disruption thesis has now been tested for two years and has failed on the revenue line. Five consecutive quarters of accelerating Search & other growth through the global rollout of AI Overviews and AI Mode is meaningful evidence. The segment margin is within its three-year range. The bear case requires a transmission mechanism that has not yet appeared in any disclosed number.

3. TPU is a genuine structural advantage that no competitor can replicate quickly. Alphabet is the only large cloud provider that designs its own AI accelerator at scale. That gives it a lower cost of inference, supply independence, and an integrated offering — model, tooling, infrastructure — that competitors selling rented compute cannot match. The Anthropic relationship, including a $21 billion order disclosed by Broadcom and a target of one million TPUs, is confirmation that the silicon is competitive externally, not merely internally.

4. The balance sheet can absorb the capital cycle without distress. $242.5 billion of cash and marketable securities, $98.2 billion of debt, $144.3 billion of net cash, and a business generating $185.7 billion of trailing operating cash flow. Alphabet is financing an unprecedented capital programme from a position of strength rather than necessity, and it has multiple sources of funding — cash, debt, equity, and an at-the-market facility — that it has not yet needed to use in full.

5. The equity portfolio is real value that the market is not paying for on a clean-earnings multiple. $218 billion of SpaceX and private AI stakes sit on the balance sheet at approximately 34% of book equity. If the private position continues to appreciate at anything like its recent rate, or if Anthropic lists, the value crystallises. On our clean-earnings multiple of 27.2× FY2027, a shareholder is paying very little for an option on that outcome.

6. Gemini's distribution is worth more than its market share. Gemini reaches users through the search box, Android, Chrome, Workspace, YouTube, and now Siri. The API token throughput series — 7 billion per minute in Q3 2025, 16 billion in Q1 2026, 22 billion in Q2 2026 — is a measure of paid usage growing far faster than any disclosed user count. Alphabet does not need to win the consumer AI race outright; it needs to be one of two or three survivors, and its distribution makes that close to certain.

22 — The bear caseSix arguments we cannot dismiss

1. The earnings that make the stock look cheap are not earnings. $6.26 of Q2 2026's $9.11 diluted earnings per share came from marking up private and newly-listed equity. Over the trailing twelve months, non-operating income of $151.6 billion exceeded operating income of $147.6 billion. Any valuation that uses the trailing multiple of 17.3 is valuing a company that does not exist, and the direction of that error is always the same: it makes Alphabet look cheaper than it is.

2. Depreciation is arriving on a $200 billion annual capital programme. The charge is currently running at roughly $28 billion annualised against capital expenditure guided to $195–205 billion and rising further in 2027. On five-to-six-year server lives, the steady-state charge is $33–40 billion per annual cohort. The gap between where depreciation is and where it is heading is arithmetic, not opinion, and it lands directly on the operating margin that the entire bull case rests on.

3. Free cash flow is negative, the buyback is zero, and the company is now an issuer. −$5.9 billion in Q2 2026, the first negative quarter since the 2004 IPO. Repurchases at zero in both quarters of 2026 against $13.2 billion in Q2 2025. $49.6 billion of equity raised in June, plus a $40 billion at-the-market authorisation, plus $98.2 billion of debt against $16 billion a year earlier. The cash return to shareholders has collapsed and the company is funding itself externally for the first time in two decades.

4. The Cloud backlog is concentrated, and the concentration is in customers financed by Alphabet. Analyst estimates put OpenAI and Anthropic at roughly a quarter of Google Cloud revenue in 2026 and potentially approaching half in 2027. Alphabet has committed up to $40 billion to Anthropic, which then buys TPUs from Alphabet. A backlog whose largest components are being funded by the supplier is not independent evidence of market-clearing demand.

5. The click economy is deteriorating and the company will not let anyone measure it. Zero-click searches estimated at 68%, publisher referral traffic in sustained decline, organic click-through rates on AI Overview queries down sharply. Alphabet declines to disclose query volumes, click volumes, revenue per query, or the cost of serving an AI answer. Advertising contracts are annual, so if the traffic deterioration is real the revenue effect could be deferred rather than absent. The company publishes the one number that supports the bull case and none of the numbers that would test it.

6. The regulatory cost is now continuous rather than terminal. Every divestiture has failed, and the market has read that as the risk going away. But behavioural remedies impose permanent compliance costs, the €890 million of DMA fines in July 2026 is the fourth material European penalty in four years, the appeal in the search case will run into 2027, and the default-payment prohibition — if it takes effect — attacks the distribution advantage that has underpinned Google's search share for two decades. Regulatory risk has become smaller in magnitude and larger in duration, which is the harder combination for a valuation to absorb.

23 — Risk matrix & catalystsRanked by probability times impact

Catalysts

24 — ConclusionConstructive at $385, on a machine that works and a mark that flatters it

Alphabet is running the best operating play in its history. Google Cloud grew 82% at a 35.6% margin, with $513.9 billion of contracted backlog behind it. Search grew 17% through the rollout of the technology that was supposed to destroy it. Gemini reaches users through every surface Google owns and now through Apple's as well. The company designs its own AI silicon, which almost nobody else at scale can do. On the evidence available, the operating business is more durable than the bear case allows.

And the reported earnings are not the operating business. Of $9.11 of quarterly earnings per share, $6.26 arrived because two companies Alphabet holds stakes in were marked up in value. Over twelve months, non-operating income exceeded operating income. The trailing multiple of 17.3 is an artefact of that accounting, and the clean multiple is closer to 32×.

What that means in practice is that the stock is not expensive because the market is wrong about the business. It is expensive in the sense that the price already reflects the operating excellence and a substantial part of the capital cycle working. Our base case of $385 gives 29.5× our FY2027 clean earnings estimate plus a discounted value for the equity portfolio. It is 11.6% above the current price and 10% below consensus. The gap to consensus is almost entirely a disagreement about whether unrealised marks on private companies should be capitalised at face value, and whether a business with negative free cash flow deserves a mid-thirties multiple.

We rate the shares Constructive. Constructive, not Conviction, because the margin of safety is thin, because the depreciation has not yet arrived, because the buyback that supported the shareholder return case has been switched off, and because the one question that decides the valuation — whether an AI answer monetises as well as a search result page — is one the company has the data to answer and has chosen not to disclose.

A shareholder buying Alphabet today is buying an excellent business at a fair price, with an unusually large amount of the return dependent on a question that will be answered by the operating margin over the next eight quarters. That is a reasonable thing to own. It is not a bargain, and the reported numbers suggesting it is one are the reason this report exists.

25 — AppendixFinancial summary tables

Quarterly revenue by reporting line ($ millions)

Quarterly operating income by segment ($ millions)

Reported vs clean earnings, trailing four quarters

"n/d" indicates the company did not disaggregate the equity securities gain from other income (expense), net in that quarter's release. Q3 2025 carried a $3.5 billion European Commission fine within Google Services and Q4 2025 carried a $2.1 billion Waymo compensation charge, both of which depress the operating line and neither of which is adjusted in the reported figures.

26 — Sources & disclosurePrimary references, limitations, and revision policy

Primary sources

Alphabet Inc. earnings release, second quarter 2026, 22 July 2026 · Alphabet Inc. Form 10-Q for the quarterly period ended 30 June 2026 · Alphabet Inc. earnings release, fourth quarter and fiscal year 2025, 4 February 2026 · Alphabet Inc. earnings release, third quarter 2025, 29 October 2025 (Form 8-K) · Alphabet Inc. earnings release, first quarter 2026 · Alphabet Inc. Q2 2026 earnings call transcript and management commentary · Alphabet Inc. Q2 2026 results announcement, Wiz acquisition completion, 11 March 2026 · European Commission, Digital Markets Act non-compliance decisions, 23 July 2026 · European Commission, advertising technology decision, AT.40670 · United States District Court for the District of Columbia, United States v. Google LLC, remedies opinion and judgment · United States District Court for the Eastern District of Virginia, United States v. Google LLC, ad technology remedies order, 2 September 2026 · UK Competition and Markets Authority, consultation on conduct requirements for Google search, 28 January 2026 · Anthropic, expanded Google and Broadcom compute partnership, 6 April 2026 · Broadcom disclosure of Anthropic TPU order, December 2025 · Alphabet Inc. blog, eighth-generation TPU announcement, April 2026 · Waymo operational disclosures, Q1 2026 and September 2026.

Secondary and third-party sources

Sensor Tower consumer AI assistant market share data, May 2026 (via TechCrunch) · eMarketer digital advertising revenue forecast, 13 April 2026 · Synergy Research-derived cloud market share estimates, Q2 2026 · SparkToro zero-click search estimates · US National Highway Traffic Safety Administration and National Transportation Safety Board filings relating to Waymo · Nasdaq and market data providers for price, market capitalisation and consensus estimates as of 16 September 2026 · contemporaneous financial press including Reuters, Bloomberg, CNBC, TechCrunch, USA Today and Financial Times for events cited with dates.

Limitations

Several figures in this report are estimates and are labelled as such. FY2026 and FY2027 revenue, operating margin and clean earnings per share are Farstar models, not guidance. The normalised tax rate of 19.5% applied to operating income is an assumption. The split of Q1 2026 earnings between operating and non-operating sources is a Farstar calculation, not a company disclosure. The identity of the private company in which Alphabet holds a measurement-alternative position is not confirmed by Alphabet; it is inferred from the arithmetic of the disclosure and third-party reporting. Alphabet's ownership percentage in that company has never been disclosed and no figure in this report depends on it. Cloud market share estimates are third-party and two versions circulate for Q2 2026; both are shown where they are used. The compliance dates for the search case remedies could not be verified against the docket and are flagged as unconfirmed. Where a figure could not be sourced, it is marked "n/d" rather than estimated.