Alphabet

IThe business

Alphabet has two kinds of customer. Advertisers and subscribers pay for Google Services: Search, YouTube, Android, Chrome, Play and devices. Companies pay for Google Cloud: computing, data, security and AI. In the twelve months to September 2026 revenue was $471.1B, up 22%, and operating income was $157.0B, a margin of 33.3%.C (calculated)⁴¹⁵⁰ Revenue has grown 18.5% a year since 2016.C (calculated)⁶

Google Services earned a 42.4% operating margin on $376.3B of revenue. Google Cloud earned 34.2% on $93.2B, and 36.2% in the September quarter against 23.7% a year earlier. Other Bets, which holds Waymo, lost $9.6B on $1.5B of revenue. Alphabet-level activities, which hold shared AI model research, corporate costs and fines, lost $24.7B.C (calculated)^4131542

How it makes money

• Search and other. $250.1B in the twelve months, 53% of revenue.C (calculated) Revenue grew 12% in the September quarter, down from 17% in the June quarter and in line with 13% in both 2024 and 2025.C (calculated)⁴¹⁴²⁵³
• YouTube ads. $43.5B, up 9% in the September quarter; Alphabet says YouTube’s 2025 revenue, ads and subscriptions together, passed $60B.C (calculated)⁴¹¹⁶
• Network. $29.6B of ads on partner sites. It is the one advertising line that shrinks: down 1.9% in 2025 and 1.1% in the first nine months of 2026, though it grew 1.4% in the September quarter.C (calculated)⁴¹³
• Subscriptions, platforms and devices. $53.2B, up 11% in the September quarter.C (calculated)⁴¹
• Google Cloud. $93.2B, up 76% in the twelve months and 103% in the September quarter. It includes Google Cloud Platform, Workspace and, since the second quarter, sales of TPU systems, the chips Alphabet designs, to customers for their own data centers. Alphabet signed a limited number of such agreements, expects the significant majority of their revenue in 2027 and does not say how much of the quarter’s growth they supplied.C (calculated)⁴¹⁵
• Other Bets. Waymo, Wing and others. Waymo’s February 2026 round of $16B valued it at $126B.¹⁷

Advertising was 73% of 2025 revenue.C (calculated) The 10-K says the margins on cloud, subscriptions, platforms and devices “are generally lower” than on advertising, and that these lines are growing faster, so the mix shift lowers the margin.³

What changed after 2022

Capital spending went from $31.5B in 2022, 11% of revenue, to $159.8B in the twelve months to September 2026, 34% of revenue.C (calculated) Net property and equipment went from $112.7B at the end of 2022 to $372.8B, of which $145.3B was not yet in service.⁴¹⁶ Management raised its 2026 capital spending guide from $180B to $190B in June to $195B to $205B in July, repeated it with the September results, and says 2027 will be a significant increase on 2026.¹⁶¹⁸⁴⁶

Operating income more than doubled from 2022 while free cash flow fell 67%.

Operating income rose from $74.8B to $157.0B. Free cash flow, operating cash flow less capital spending, fell from $60.0B to $19.7B, and it has been negative for two quarters running: minus $5.9B in June and minus $9.1B in September.C (calculated)^4142619

IIThe moat

Alphabet has two moats of different types. In Search the moat is distribution and habit: a default position in browsers and phones that Alphabet pays partners to keep, and advertisers who bid where the people are. It is holding. The second is a cost advantage in computing for AI, built on Alphabet’s own chips and network and a quarter-century of running data centers. I cannot yet tell whether it is widening: the evidence is Alphabet’s own, and Cloud’s margin rose before the depreciation on $145.3B of unfinished equipment began.⁴¹⁴²

StatCounter puts Google at 90.38% of worldwide queries on average over the twelve months to September 2026, against 89.77% a year earlier, and Bing at 4.68% against 3.99%. September alone was weaker: Google fell to 89.94% from 91.02% in August, and Bing’s 5.29% is its highest in 25 months. StatCounter counts queries on search sites, so it misses questions asked inside AI assistants.C (calculated)⁸

Search revenue grew 13% in 2024 and 2025, 17% in the June quarter and 12% in the September quarter. AI Mode passed one billion monthly users in June, and the Gemini app reached one billion users in October.^4132047 If AI assistants take queries, it shows first as Search growth falling below 8% while Cloud grows. Payments to partners were 20.0 cents of each dollar of revenue subject to them in the September quarter, against 20.1 a year earlier; the 10-Q attributes the fall to mix, as Network, which pays out the most, shrinks, and says the rate on Search and other was substantially consistent.C (calculated)⁴¹⁴²

Alphabet’s cost claims are its own and I could not verify them: it says Gemini serving costs fell 78% in 2025.R (reported)¹⁶ The visible result is Cloud’s operating margin, 11.3% in the June 2024 quarter and 36.2% in the September 2026 quarter, with growth up from 32% in the June 2025 quarter to 103%.C (calculated)^4142514 Three things limit it. The margin is before the depreciation on the $145.3B of equipment not yet in service. It leaves out shared AI costs held at Alphabet level: with its 24% share of the September quarter’s $7.7B, Cloud’s margin would be 30.2%. And the 103% includes TPU system sales, which Alphabet does not break out; inventory, which is mainly TPU systems and consumer devices, rose from $2.4B at the end of 2025 to $13.3B.⁴¹⁴²

The backlog shows companies committing years ahead: $567.7B at 30 September 2026, $560.8B of it Cloud, against $242.8B at the end of 2025, with just over 55% to become revenue in 24 months.⁴¹³ Alphabet does not report the customer mix or its TPU system revenue, so I cannot say how much of Cloud’s growth is hardware or how much comes from companies Alphabet funds. The 10-Q discloses one piece: $16.8B of the backlog is from Crux AI, a joint venture with Blackstone in which Alphabet holds a minority interest and which it treats as a related party.⁴¹ The first-quarter 10-Q says the backlog includes agreements for multiple gigawatts of TPU hardware, mostly recognized in 2027, and that Alphabet agreed to back credit obligations of third-party data centers for some of them.²¹ Fortune names Anthropic as an Alphabet investment and reports that it committed to at least five gigawatts of Google Cloud capacity.²²

… levels that are meaningfully exceeding our available supply.

The moat has limits. The December 2025 Search judgment restricts how Google distributes its services and requires it to share certain search data with competitors; both sides have appealed.³ Returns have fallen with the build: the after-tax operating profit gained per dollar of capital added was 56 cents for 2019 to 2022, 49 for 2020 to 2023, 26 for 2021 to 2024 and 30.5 for 2022 to 2025.C (calculated) All are well above my 10.0% cost of capital, but the $74.6B added in the first half of 2026 has not yet reported a return; I have not recomputed that figure for the September balance sheet.²⁶

Search has held 90% of queries on search sites through two years of AI; Cloud’s margin tripled before its new equipment started depreciating.

IIIThe competitive advantage

Alphabet’s advantage over Amazon, Microsoft and Meta is breadth. It runs an advertising business larger than Meta’s or Amazon’s and the fastest-growing of the three big clouds, with its own chips and models connecting them. Google advertising was $81.6B in the June quarter, 1.4 times Meta’s $59.4B and 4.1 times Amazon’s $19.8B.⁵⁹²³ The same quarter shows the weakness. Alphabet’s advertising grew 14%, against 27% at Meta and 26% at Amazon, and Meta added more advertising dollars in the year than Google did.C (calculated)⁵⁹²³

Meta’s margin includes $2.4B of legal charges and $1.2B of severance.²³

In cloud the picture is the reverse. Google Cloud is growing twice as fast as AWS from a smaller base. The peer tables stay on the June quarter so that the three companies are compared on the same dates.

Microsoft’s Intelligent Cloud adds server software and GitHub to Azure, whose growth its June release puts at 43%; the Microsoft page on this site uses the basis Microsoft adopted in September 2026, $29.4B and 42%.⁹²⁴²⁶

Where Alphabet leads

Alphabet designs the chips and trains the models, and puts them in front of 13 products with more than a billion users each.²¹⁶ Google Cloud’s growth has accelerated for six quarters running: 32%, 34%, 48%, 63%, 82% and 103%. AWS also accelerated, from 17% to 37% through the June quarter.C (calculated)^414259141521 The second lead was funding capacity: in the twelve months to June, operating cash flow of $185.7B was the largest of the three, and free cash flow of $53.3B was positive where Amazon’s was negative. Alphabet’s own free cash flow has since fallen to $19.7B.⁶⁹⁴¹

Where it trails

AWS has nearly twice Google Cloud’s share of the market, 28% to 15%, and a higher operating margin.⁵⁷⁹ Meta and Amazon are growing their advertising faster.⁵⁹²³ No verified third-party share exists for AI assistants; the only usage figure is Alphabet’s own, one billion Gemini app users.R (reported)⁴⁷

IVManagement

Sundar Pichai has run Google since October 2015 and Alphabet since December 2019. Since 2015 revenue has grown 18% a year, operating income 21% and free cash flow 16%.C (calculated)⁶²⁷ The stock returned 22% a year from October 2015 to Q4 2026, with dividends reinvested, against 15% for the S&P 500, and 24% a year against 16% since he became Alphabet’s chief executive.C (calculated)⁵¹

Two founders control the vote. Larry Page and Sergey Brin own Class B stock with ten votes a share and hold 52.7% of the vote with about 12% of the shares; I added their non-voting Class C stock from Form 4 filings, Page’s as of April 2022.³²⁷²⁸ The chair, John Hennessy, is independent, and 7 of the 10 nominees were. Say on pay received 81% of the votes for and against in June 2026, and a proposal for equal voting 31%. Page and Brin hold about 7.5 billion votes. If they voted with the board, outside holders supported pay with about 52%; if they voted against the proposal, they supported equal voting with about 79%.E (estimate)²⁷²⁹

Capital allocation

I weigh the 2026 change in capital allocation most.

From 2021 to 2025 Alphabet spent $279.2B on 2,084M shares, an average of $134, which would be worth $591B at $283.82. The share count fell only 10.5%, from 13,504M at the end of 2020 to 12,088M at the end of 2025, largely because employee stock awards issued new shares.³⁶¹⁹³⁰ Alphabet paid an average of $189 in 2025. Then it bought none in the first nine months of 2026, and in June it sold about 86M new shares at $348 to $355, including $10B to an affiliate of Berkshire Hathaway, plus $19.25B of mandatory convertible preferred stock that converts into up to 54.5M more. Net proceeds were $49.6B.^2313241 In the September quarter it sold a further 36M shares through its at-the-market program for $12.4B.⁴¹⁴²

I read the June sale two ways. A sale at 79% above my base value helps the holders who kept their shares, as the buybacks at $134 did. But management also chose to raise cash at $355 rather than wait. In the first nine months operating cash flow covered 96% of capital spending, and then $34.1B of acquisitions, $22.3B of net stake purchases and $17.4B of net payments on employee stock awards came on top.⁴¹ The ten years from 2016 to 2025 saw $20.6B of cash acquisitions in all.³⁶

Incentives

Pichai’s salary is $2.0M and he has no annual bonus. Every three years he gets a stock award, and the March 2026 award was $385M at target: $126M of performance shares measured only by Alphabet’s total shareholder return against the S&P 100 companies, $84M of units that vest with time, and $175M of Waymo and Wing units that depend on those companies’ values. His 2025 reported pay was $10.9M; compensation actually paid, which marks stock to market, was $213.9M.¹¹²⁷

No part of the award measures return on capital, free cash flow or the cost of the build. Relative total shareholder return is what holders feel, and it has been strong, but it does not test whether capital earned its cost. $175M of the $385M award, 45%, pays on the rise in Waymo’s and Wing’s unit values, which each company measures by its own valuation method; the 8-K does not describe it. Alphabet funded most of the round that valued Waymo at $126B.¹¹¹⁷²¹ Operating record: strong. Revenue per employee was $2.1M in 2025, and headcount rose 8% in the year to September while trailing revenue rose 22%.C (calculated)³⁴²⁵⁰ Pichai directly holds about 1.9M shares, $0.54B at the Q4 2026 closes, plus 1.1M in two 2026 annuity trusts; the direct holding is about 4.2 years of his March 2026 award spread over three years ($128M a year).C (calculated)¹¹²⁸

Pichai is paid in stock on relative share return, and he owns about $0.54B of it directly. That pays him to beat other large companies on the stock, and it does not charge him for the build. On capital over a cycle, Alphabet spent $279.2B on its own shares from 2021 to 2025 at an average of $134, spent nothing in the first nine months of 2026, and sold new shares at $348 to $355 in June.¹¹³⁴¹ No pay measure tests whether the $259B of Year 1 spending earns its cost.

VFinancials

Operating income rose 6.6 times from 2016 to the twelve months to September 2026; free cash flow fell 24%, because capital spending now takes 89% of operating cash flow.⁶⁴¹⁵⁰

Free cash flow is operating cash flow less purchases of property and equipment; free cash flow after stock pay also subtracts stock-based compensation, which treats stock pay as the cost it is.³⁶

The share of after-tax operating profit that Alphabet reinvested went from 20% in 2021 to 101% in the twelve months to September: it already puts back more than all it earns. In my base case it is 133% in Year 1.C (calculated)⁶⁴¹⁵⁰

Operating margin rose from 26.3% in 2016 to 33.3%, through the early part of the build. The gain came from costs before depreciation. The margin before depreciation went from 31.3% in 2023 to 37.3% in 2025 and 39.3% in the twelve months to September, and depreciation rose only from 3.9% to 5.9% of revenue, because $145.3B of equipment is not yet in service. The September quarter alone was 38.4% before depreciation, down from 40.0% in June, as shared AI research and the cost of TPU systems sold rose.⁴¹⁴²⁶

Depreciation now catches up. In my base case it rises from 5.9% of revenue to 9.3% in Year 2, 14.7% in Year 5 and 16.7% by Year 7, and the operating margin falls to 27.3%. I apply the 10-Q’s 60% share of technical infrastructure assets that are servers and network equipment to each year’s spending, written off over six years.⁴¹³ The 10-K itself says that the costs of operating its technical infrastructure, depreciation among them, “are expected to significantly increase”.³

Accounting to watch

1. Investment gains. GAAP net income was $235.7B in the twelve months to September, but $129.6B of the $287.6B of pre-tax income was net gains on equity securities, mostly unrealized marks on SpaceX and one private company.⁴¹⁵⁰ Earnings per share were $19.15, and about $10.8 without the gains, on my 21% tax estimate. I do not use GAAP price-to-earnings. The marks also run the other way: in the September quarter Alphabet booked an $8.7B net loss on equity securities, which cut diluted earnings per share by $0.55, and the unrealized loss on marketable and other equity was $12.4B, against the roughly $12.9B I had estimated from SpaceX’s price.⁴²⁴¹ SpaceX closed at $170.86 on 30 June, $150.86 on 30 September and $143.89 on 29 October, a further mark-down of about $3.5B on the $75.2B of shares restricted at 30 September.³³⁴⁹⁴¹
2. Asset lives. Alphabet lengthened the life of its servers from four years to six in January 2023, which cut 2023 depreciation by $3.9B.¹⁹³⁴ A server life of five years in my model lowers my base value to $189, and four years to $178; seven years raises it to $207. A server share of 70% of each year’s spending lowers it to $176; 50% raises it to $218.
3. Stock pay. Stock-based compensation was $30.4B, 6.5% of revenue, and Alphabet also paid $22.6B in cash for taxes on vesting stock in the twelve months. The 284M unvested units were valued at $227.39 on average when granted; at $283.82 they are worth $16.0B more, about $1.3 a share, which the expense does not show.C (calculated)⁴¹⁴²⁵⁰

VIValuation

How this goes wrong

Before any numbers I ask what would most likely ruin this, and work back. I see three ways. The build earns less than it costs: assistants take queries or compute prices fall, and the roughly $245B of my bear case arrives anyway, so depreciation nearly triples to 16% of revenue. Cloud growth fades faster than the 21% a year my base needs, while $260B to $320B a year of spending arrives either way. Or advertising, 73% of revenue, turns down while cash is already negative, which on my base takes $22.8B of net cash to about $48B of net debt by September 2028.⁴¹⁶

What an owner can take out is far from what is reported. In the twelve months to September, net income was $235.7B, but $129.6B of the pre-tax income was gains on equity stakes. Operating income was $157.0B. After $159.8B of capital spending and $30.4B of stock pay, free cash flow was minus $10.7B.⁴¹⁵⁰ Depreciation of $28.0B is my closest measure of standing still, so about $132B of the spending is growth that still has to earn its cost C (calculated).⁴¹

What I can’t judge is how much of each new dollar of capital comes back as revenue. It was $1.15 in 2024 and $0.65 in the last twelve months, and I use $0.51 rising to $0.90. That one input moves my base from $198 to $284; read $198 with that in mind.

What the price implies

The price values the operating business at $3,292B once the stakes and net cash are taken out. That is 21.0 times trailing operating income, 25.3 times after-tax operating profit and 7.0 times revenue.⁴³⁴¹⁵⁰ On consensus earnings per share it is 18.8 times 2027 ($15.07, 18 analysts) and 15.3 times 2028 ($18.49, 11 analysts).⁴⁵ Nasdaq does not say what these include; the $20.40 for 2026 carries the first nine months’ equity gains. My base earnings, after-tax operating profit per share, are 9% below consensus for 2027 and 16% below for 2028, because I assume slower growth and a lower margin before depreciation. Free cash flow after stock pay was negative, minus $10.7B, and after-tax operating profit yields 4.0% of the operating business, below the 4.98% ten-year Treasury.⁴¹⁴⁸

Run my model backwards, holding its margins and capital needs. The price then needs revenue to grow 16.1% a year for ten years, to about $2.1 trillion in Year 10, against $471B now. Alphabet grew 18.3% a year from 2015 to 2025, from a base one-sixth of today’s.³⁶ Holding the margin at today’s 33.3% with depreciation never catching up, the price still needs 13.6%.

The price needs revenue to grow 16% a year for ten years, to $2.1 trillion, on my margins and capital needs.

Each needed value comes from solving my model for the price with every other input unchanged.⁴¹⁵⁰ Alphabet does not disclose capital by segment, so these tests charge growth outside Cloud the same capital as Cloud. If growth outside Cloud cost capital at 2023’s reported $1.21 per $1 and Cloud took the rest of my base capital, everything else would need 13.4% and Cloud 31.7%.E (estimate)

Capital explains most of the gap between my value and the price. With no new capital needed for growth, my base revenue would be worth about $410 a share; if each new dollar of capital keeps bringing only $0.51 of revenue, $125.E (estimate) Swapping one bull input at a time into my base adds $37 for growth, $36 for margins, $27 for capital needs and $8 for growth after Year 15; together, in the chart’s order, they reach my bull case at $335.

Reported revenue added per $1 of net new capital was $1.40 in 2022, $1.21 in 2023, $1.15 in 2024, $0.75 in 2025 and $0.65 in the twelve months to September. The price needs the efficiency of the years before the build to return in Year 1, at $0.86, and to nearly double by Year 10. The guide rules that out for Year 1: $0.86 means capital spending of $170B, while the guide leaves at least $63B for the fourth quarter of 2026 and a 2027 no lower than 2026’s $132.0B pace for its first three quarters, about $195B or more.⁴¹⁶⁴⁶

Only the first three rows fit the guide. The floor, the guide’s low end with no growth in the first three quarters of 2027, is worth $230, $32 above my base. The price is almost exactly what 2024’s efficiency held flat would be worth, and the guide rules that out. I keep $0.51 because spending arrives before its revenue, and because my base takes the fourth quarter of 2026 from the guide’s $63B to $73B and sets the first three quarters of 2027 at $186B to $196B, so that 2027 as a whole lands near $250B or more, a rise of about a quarter on 2026, which fits a significant increase; the floor does not.⁴¹⁴⁶

What the price pays for the cloud gives the same answer from the other side. I value everything outside Cloud, including Search, YouTube, subscriptions, Other Bets and the shared AI research, at a multiple of its $103.9B of after-tax operating profit and read off what is left for Cloud. I charge Other Bets and the Alphabet-level costs to the rest, which raises what is left for Cloud; charging the $24.7B of Alphabet-level costs to Cloud instead leaves $804B at 20 times, 22 times Cloud’s annualized profit.⁴¹⁴²

Backlog is $560.8B, and just over 55% becomes revenue in 24 months. At the middle row the price pays $1.21 trillion for Cloud, 33 times its annualized September-quarter after-tax profit.⁴¹⁴²

Bear, base and bull

The model runs fifteen years, October 2026 to September 2041: ten explicit years of growth in Google Cloud and everything else, then five in which growth fades to the terminal rate. The operating margin is a margin before depreciation, which I estimate, less depreciation built from the capital the growth needs: 60% of spending on servers and network equipment over six years and 40% on buildings and other assets over 25, starting ten months after the cash goes out.⁴¹³ I rolled my June cases forward one quarter: each case keeps its calendar path of growth, margin and capital efficiency, and Cloud and everything else start from the reported September figures. The model is close on 2025 depreciation ($22.1B against $21.1B reported) and the twelve months to September ($26.4B against $28.0B), and carries $141B of equipment not yet in service against $145.3B reported; a 70% server share would overshoot 2025, the nine months and the twelve months, so I keep 60%.⁴¹³

Each year’s cash flow is after-tax operating profit less net new capital: revenue growth divided by the new revenue each $1 of net new capital brings. Tax is 17%, close to 2025’s 16.8%, and stock pay is a cost inside the margin. Cash, debt securities and the stakes after 21% tax, less debt, finance leases, accrued fines and noncontrolling interests, come to a net $213.2B, or $17.26 a share; the preferred is counted as 54.5M shares, so I divide by 12.35B shares.⁴¹³²

I discount at 10.0%. The capital asset pricing model gives 10.04%: the 5.29% ten-year Treasury yield against which Damodaran computed his 4.20% implied equity premium of 1 October 2026, plus that premium times Alphabet’s ten-year beta of 1.13.C (calculated) At the Q4 2026 yield of 4.98% the formula gives 9.73%; I keep the yield that goes with the premium. The five-year beta of 1.21 gives 10.4%, and the ten-year beta adjusted toward 1 gives 9.9%. The 8.27% that the price needs implies an equity premium of 2.6 points.^10121348

My base Year 1 capital spending of $259B implies $186B to $196B in the first three quarters of 2027 on top of $63B to $73B in the fourth quarter of 2026, which is the $195B to $205B guide less the $132.0B already spent; Alphabet repeated that guide with the September results.⁴¹⁴⁶ My base Cloud revenue for Years 1 and 2 is $157B and $222B, $379B in all, against the $310B or so that the whole backlog supplies in 24 months.⁴¹ The return my base earns on new capital after tax is 14% in Year 1 and 21% in Year 10, below the 30.5% of the last three reported years and above the 10.0% cost of capital.

The bear case is stranded capital. AI assistants take queries and compute prices fall: Cloud growth drops from 47% to 14% by Year 3 and 3% by Year 10, everything else grows 9% in Year 1 and fades to 3%, and the margin before depreciation stays at today’s 39%. The roughly $245B build arrives anyway, so depreciation nearly triples to 16% of revenue and the operating margin falls to about 22%.⁴¹³

My bull case matches consensus for 2027 and 2028 and extends it with my assumptions. Its after-tax operating profit per share lands within 3% of consensus earnings per share for both years, $15.21 against $15.07 and $18.30 against $18.49, a comparison that sets operating profit against earnings, which also carry interest and other income. Years 3 to 15 are mine: Cloud compounds 26% a year to $962B and each new dollar of capital buys $0.56 of revenue rising to $1.10. At my 10.0% rate the bull case is worth $335; the price is 15% below it, so the market is paying less than my bull case and 43% more than my base. Consensus 2026 earnings of $20.40 include the first nine months’ equity gains and are not comparable. Consensus for 2027 and 2028 barely moved with the report: $15.03 and $18.45 before it.⁴⁵⁴⁴²

The price earns a return if each case plays out: the discount rate at which the case’s cash flows equal $283.82.⁴³

That compares with the 4.98% ten-year Treasury yield and my 10.0% discount rate.⁴⁸ The value after Year 15 is 57% of my base, 49% of the bear and 60% of the bull.⁴¹³

The cells at an 8.0% discount rate reach the price; none of the others do. One point on the discount rate matters: $241 at 9.0% against $167 at 11.0%. A tax rate of 21% lowers the base to $185.¹²¹³

I discount from the end of each year and date the values to 30 September, the last balance sheet. Mid-year timing and dating the values to Q4 2026 would lift all three cases by about 6%, my base to $210. SpaceX is the one item I date to Q4 2026, because it is listed and its price is known.⁴³⁴¹

What is in the price besides Search and Cloud

The stakes and net cash are $17.26 a share, 6.1% of the price, and are counted in every case; stakes are taxed at 21%, operations at 17%. Two items sit outside the cases. Waymo is consolidated, so its losses are inside Other Bets’ $9.6B annual loss, a cost of about $9 a share in my base. The February round valued all of Waymo at $126B, or $10 a share before tax and before outside investors’ share, so the two roughly offset and I give no net credit; Alphabet does not disclose its share.¹⁷

Fortune reports that Alphabet’s initial $300M investment in Anthropic has grown to $13.3B, with up to $30B more committed, and that Anthropic’s private valuation rose from $350B to $965B. Anthropic’s S-1 puts what Google can still fund at $20.0B at 30 June, after a $10.0B purchase in May.⁵² The 10-Q carries all private holdings at $128.2B, re-marked only on observed transactions ($8.8B of them in the September quarter), and discloses $21.9B of future funding commitments related to equity investments, which I do not deduct. I cannot reconcile the figures, the carrying value may lag a later valuation, and I give no credit for that.⁴¹²²

The margin of safety

• Price$283.82Close, Class A (GOOGL), Q4 2026 [43]
• Base value$198My estimate E (estimate)
• Margin of safety below$13930% under my base value E (estimate)

I set the margin of safety at 30% below my base value, which gives $139. My rule is 25% for a business with net cash and no customer above a tenth of revenue, 30% when one cycle or two customers drive results, and 35% when fixed charges such as leases sit ahead of the shares. No Alphabet customer exceeded a tenth of revenue in 2023 to 2025, and net cash is $22.8B before the stakes, which would fit 25%.⁴¹³ But my base case has free cash flow after stock pay of −$55B in Year 1 and −$16B in Year 2, which turns that net cash into net debt of about $48B by September 2028, before dividends, interest and the stakes, and advertising, 73% of revenue, follows the cycle, though the 2020 test was mild: revenue in the June 2020 quarter was 1.7% below a year earlier and 7.0% below the March quarter.C (calculated)⁶ That puts Alphabet at 30%. I do not use 35%, which gives $129: the $130.5B of leases not yet started is 4% of market value, and the $924.4B of purchase agreements, mostly equipment, are a risk my bear case prices, not a charge ahead of the shares.⁴¹ The price is 104% above that line, so the stock would have to fall 51% to reach it.

The line is strict. The base is dated 30 September; rolled to Q4 2026 it is $210, and the same 30% gives $147, 48% below the price.E (estimate) The lowest close of the past year was $273.50 on 30 March 2026, 97% above the line; the stock has fallen 30% from its $402.62 close of 13 May, including a 12.6% fall the day after the September-quarter report.C (calculated)¹⁴³

Where this could be in 5 and 10 years

These are my model’s values at later dates, per today’s 12.35B shares, in each year’s dollars. Each is the present value at that date, at 10.0%, of the cash flow after it, plus $17.26 a share of net cash and stakes held flat. Cash flow is free cash flow after stock pay. All columns are per share. The return starts from the $283.82 price.

Owner cash is the model’s cash flow added up over ten years. It is negative at first, −$55B in Year 1 in the base, so the $86 is net of that. The return counts the price growth and that cash, and assumes the price meets my value in 2036.

For the base to hold at five years, revenue has to reach about $1,023B at a 29% operating margin, with $101B of cash flow, and each new dollar of capital has to buy $0.69 of revenue. At ten years it needs $1,431B and a 28% margin. Growth, margin and capital drive the cases. Revenue grows 6.3%, 11.8% or 14.9% a year. The Year 10 operating margin is 22%, 28% or 33%. New revenue per $1 of capital reaches $0.70, $0.90 or $1.10. The 10.0% rate is the same in all three; at 9.0% today’s base is $241, at 11.0% $167.

The likeliest way my base is wrong is Cloud growth. It has to be 41% in Year 2 and 21% a year over ten years, while $260B to $320B a year of spending arrives either way. If growth fades faster, depreciation lands on less revenue and the result moves toward the bear.

These are estimates of value, not forecasts of the share price. The market price can sit above or below value for years.

The next best use of the same dollar

The closest comparable here is Amazon, whose base case returns 6.1% a year from its price, against 5.4% for Alphabet’s and 4.98% for the ten-year Treasury.⁴⁸ How good the business is and how good the price is are two questions. The first is strong: operating income up 6.6 times since 2016, a 33.3% margin. The second asks for 16% revenue growth a year for ten years.

My base needs three things together: Cloud growth near 21% a year, new revenue per dollar of capital rising from $0.65 to $0.90, and a margin before depreciation of 40.7% to 44.0%. If I give each a 70% chance, all three come to about one in three.

VIIThe case for the price

What most people miss

Alphabet does not say how much of its backlog is Anthropic. Anthropic’s S-1, filed on 26 October, gives a floor on the contract. It must spend at least about $114.5B with Google from April 2026 to July 2033, used or not, with some exceptions, and can cancel only over Google’s uncured material breach.⁵² That equals 20% of the $560.8B Cloud backlog C (calculated).⁴¹ But some of it has been spent already, and neither company says how much of the rest sits there. Anthropic expects to pay $125B to $150B for more TPU computing delivered from 2028 to 2031, by suppliers including Google. At 30 June, Alphabet could still fund up to $20.0B of Anthropic notes or stock, so some of this money is its own.⁵²

Sold TPU systems skip the capital spending line. Anat Ashkenazi, the CFO, called them “not the traditional kind of CapEx”. Cash goes into inventory, mainly TPU systems and phones, up from $2.4B to $13.3B this year, and comes back on delivery, with a “meaningful revenue ramp in 2027”.⁴⁶⁴¹ I read these sales as revenue with no new plant of Alphabet’s own. Alphabet gives no margin on them, and $6.6B of repurchase guarantees sit behind them.⁴¹

Google Cloud’s backlog is close to AWS’s: $561B against $578B, on quarterly revenue of $30.7B against $46B.⁴⁶⁵³

The bull case

Alphabet makes its own chips and models, and Search helps pay for the build. Gavin Baker said in December 2025 that Google has been the lowest-cost producer of tokens, the units of AI output, but expected Nvidia’s Blackwell chips to end that.⁵⁵

Warren Buffett told CNBC on 15 July that he started Berkshire’s position and that waiting was a mistake. He also said he likes it less than “at least four or five other businesses” Berkshire owns, and that Google and its rivals are “laying out hundreds of billions”.⁶⁰ Greg Abel told CNBC on 2 September that Berkshire “saw Google as a significant player” in AI.⁵⁶ Berkshire’s 13F shows 106.0M shares, worth $37.8B, at 30 June.⁵⁴ Yahoo Finance reports that Pivotal Research kept a $475 target on 30 October.⁵⁹

“We are very bullish long term on Alphabet,” Bill Ackman wrote on X on 16 May, as Reuters reported, but at current valuations he used the stake to fund Microsoft.⁵⁷ Pershing Square held 344,102 shares at 31 March, down from 6.84M at 31 December.⁵⁸

Against reality

The price sits about three-fifths of the way from my base to my bull. Bull margins lift $198 to $235, and bull capital needs lift that to $261. By my estimate, with growth after Year 15 at my base 3%, revenue then has to grow about 13% a year for ten years, against my 11.8% E (estimate). So if Alphabet takes its margin before depreciation to 48% by Year 5, gets $1.10 of revenue from each new dollar of capital by Year 10, and grows at that rate, the price is possible.

Cloud is ahead: 103% growth in the September quarter, against 79% in my bull for Year 1. The rest grew 11.1%, under the 12.6% in my base. Margin is behind: 38.4% that quarter, down from 42.0% in March, against 42.4% in my bull for Year 1.⁴¹⁴²⁵⁰ Each new dollar brought $0.65 over the last twelve months. That clears the Year 3 line, but the ratio has fallen four years running, and capital spending is set to “increase significantly” in 2027.⁶⁴¹⁴⁶

What they have to hit

Rows from my Risks signposts sit halfway between base and bull, so they are a minimum. Every filing is free on SEC EDGAR (opens in a new tab).

Where I come down

The S-1 moved me most: part of the backlog is firmer than I assumed, though concentrated and partly funded by Alphabet. It did not move my values, because a contract does not lift margins or cut the capital behind each dollar of revenue. A margin before depreciation back above 40.7%, and disclosed TPU revenue that shows growth arriving without capital spending, would raise my values.

VIIIRisks, and what would change my view

The risk Alphabet names in its own filing is the one I weight most:

… have excess capacity that we cannot easily redeploy.

1. The build outruns demand. Alphabet has $924.4B of purchase commitments, $130.5B of leases not yet started, $22.8B of financial guarantees and $75.7B of backstops for data centers, $48.0B of it executed with third-party providers.⁴¹ Capital spending was $132.0B in the first nine months against $127.1B of operating cash flow, and management expects 2027 to rise significantly.⁴¹⁴⁶ My bear case prices that in.
2. Funding and dilution. Debt rose from $49.1B at the end of 2025 to $127.5B at face value on 30 September. The preferred adds up to 54.5M shares by May 2029, and the $40B at-the-market program has $27.6B left after selling 36M shares for $12.4B in the September quarter. Buybacks stopped; $69.5B of the authorization is unused.⁴¹³²⁴²
3. Marks that reverse. Equity gains were 45% of pre-tax income in the last twelve months, and the September quarter showed the other side: an $8.7B loss. A lower SpaceX price or private-company mark flows straight through earnings. My values count SpaceX at its Q4 2026 close and the private holdings at their September carrying value, each after 21% tax, so a further fall in SpaceX lowers them. The SpaceX shares were restricted at 30 September; a 10% discount for that would take $0.46 a share.⁴¹⁴²⁴⁹
4. Antitrust and fines. The Search judgment of December 2025 is under appeal by both sides. In the ad tech case the court issued its remedies decision in September 2026, rejecting the sale of AdX and ordering behavioral remedies; the 10-Q filed on 29 October still treats the final judgment as to come, after which Alphabet plans to appeal. The European Commission’s EUR 3.0B ad tech fine is under appeal, and $13.8B of accrued fines and settlements is counted as a claim.⁴¹³³⁶
5. AI changes how people search and shop. The 10-K says AI is quickly reshaping advertising and that Alphabet may not adapt effectively.³
6. Control. Two founders hold 52.7% of the vote; outside holders cannot change the capital plan, the pay design or the board.³²⁷²⁹
7. Higher rates. The ten-year Treasury yield was 4.98% in Q4 2026, down from 5.29% on 30 September. Each point on the discount rate moves my base by 16% to 22%.⁴⁸¹²
8. The customer is also an investment. Alphabet carries private stakes at $128.2B, mostly one private AI company, and has $21.9B of future funding commitments for equity investments. Fortune names Anthropic and reports that it committed to at least five gigawatts of Cloud capacity. Anthropic’s S-1 shows at least about $114.5B committed to Google, and that $10.0B purchase of its preferred stock.⁵² If Anthropic is that private company, the backlog and the mark depend on the same balance sheet. Alphabet does not disclose the backlog’s concentration beyond $16.8B from Crux AI, so I cannot tell how much of my Cloud Year 1 of $157B depends on companies it funds. Halving the private holdings’ carrying value takes $4.10 a share, 2.1% of my base.⁴¹²²

What would move me

Where a row comes from my cases, the thresholds sit halfway between my base and each case. Year 1 now runs October to September, so those four rows are re-derived from the rolled cases; the thresholds I set before the quarter are in brackets, and the other four rows are unchanged.

The first test: the September quarter

Alphabet reported the September quarter on 28 October, and the stock fell 12.6% the next day, from $324.89 to $283.82.⁴³⁴⁴ Before the report I set four tests for it. Here is how each came in, with two of the signposts from the table above.

My base value moved from $192 to $198. Rolling the model forward a quarter added about $6, Cloud’s beat added $2.5, a slower rest of the business took off $0.7 and the larger share count took off $1.1; net cash was $213.2B against $213.9B.E (estimate) The price is now 43% above my base value, against 67% before, and 15% below my bull value. It needs revenue to grow 16% a year for ten years, against 19%. The quarter confirmed the Cloud path and the funding strain, and it did not break the thesis: the price still pays for more than my base case. Consensus earnings for 2027 barely moved, from $15.03 to $15.07.⁴⁵⁴

Three things in the quarter count against my base. The margin before depreciation was 38.4%, 2.3 points under the 40.7% my path needs for Year 1, and each point of it in every year is worth about $10 a share. Revenue outside Cloud grew 11.1% against the 12.6% my path needs in Year 1.C (calculated)⁴¹⁴² Management also expects margin headwinds in Cloud in the fourth quarter as it ramps third-party computing capacity.⁴⁶