Alphabet (GOOGL) — The Backlog Came With Its Bill

In July we named two showings that would close most of the gap between our value and Alphabet's price. The third quarter delivered half of one: a $561 billion Cloud backlog, with just over 55% of the total backlog due within 24 months; whether the 36% segment margin holds once rented capacity is in use cannot yet be read. It also delivered the bill: $51.4 billion of capital spending, 40% of revenue, and a 2027 the company says will be "significantly" higher. We value Alphabet at $223.64, against the $324.89 close before the release. The largest step, −$33.13 (−$53.32 if applied last, about half the gap to the price), is a choice of method, not a fact of the quarter: fair value is now the company's own cash flows, with peer and segment multiples as cross-checks. The quarter's data, with our reading of them, add +$6.04. Our bull case, $338.17, sits above that close.

TON618 Capital Research · As of 28 October 2026 · Download PDF · Download model

On 28 October Alphabet reported the order book our July note was waiting for. Cloud revenue doubled to $30.7 billion, its margin rose to 36.2%, and the backlog grew $47 billion in a quarter to $561 billion, with "just over 55%" expected as revenue within 24 months. July had named two showings that would close most of the gap: a 2027 capital plan below 35% of revenue with a path down, or the backlog converting at least half within 24 months with the segment margin holding above 30% after capacity eases. The same release showed what the order book costs. Capital spending rose 114% to $51.4 billion, more than the quarter's $42.3 billion of operating cash. This note asks what the order book is worth once its bill is counted.

Terms used throughout, defined once. Capex: purchases of property and equipment, the line Alphabet guides. Backlog: Cloud revenue under contract but not yet recognized (accountants call it remaining performance obligations). TPU systems: Alphabet's own AI chips (tensor processing units) sold as complete systems to outside customers; the release reports them as Cloud product revenue. Equity-security marks: gains and losses on the shares Alphabet owns in other companies (SpaceX the largest), most of them unrealized; ex-marks removes the after-tax mark, using the company's own footnote. Free cash flow to equity (FCFE): net income plus depreciation, less capex and working capital, plus net new borrowing; stock-based compensation stays an expense. Comps and SOTP (sum of the parts) value Alphabet at peer multiples on its own figures and at segment-by-segment multiples. Beta: how far a stock moves with the market, which sets its cost of equity. Growth stages: the forecast split into periods, each lasting as long as the thing that drives it. ROIC: after-tax operating income over invested capital (equity plus debt less cash), against the weighted average cost of capital (WACC). Terminal growth: the rate the business grows at forever after the forecast ends.

Revenue was $127.58 billion, up 25% in both reported and constant currency. Operating income was $40.58 billion, a 31.8% margin, up 30%. That growth rate flatters the quarter: a year ago Alphabet booked a $3.5 billion European Commission fine in its operating expenses, and without it operating income rose 17%. Google Services earned a 40.7% margin, against 42.5% a year earlier with the fine removed. Other cost of revenues rose 32%, which the CFO attributed mainly to hardware costs for TPU system sales, depreciation, content and data-center operations; research and development rose 36% on AI hiring and depreciation. Alphabet-level costs, mostly shared AI research, rose to $7.7 billion from $4.5 billion.

The miss, in two parts. Diluted EPS was $2.12 against consensus of $3.01 (MarketBeat) or $3.02 (Investing.com), a gap of about 89 cents. The release's footnote sizes the mark at 55 cents a share: an $8.7 billion loss on equity securities, which the CFO attributed primarily to SpaceX's share price, cut net income by $6.7 billion after tax. Without it EPS was $2.67, against $2.19 a year earlier on the same basis, up 22%; the year-ago quarter also carried the EC fine, about 28–29 cents a share before tax, and without both, growth is roughly 8–10%. If consensus excluded marks, the remaining gap of roughly 34 cents is not explained by them; we cannot tell from the compilations whether analysts modeled a mark at all.

The reported line misleads in both directions (Exhibit 1). In the first half, gains on the same holdings added $2.35 and $6.26 a share; on the July anchor date the reported trailing price-to-earnings ratio was 17 times, against 33 on clean earnings. Excluding the marks, quarterly EPS has run between $2.67 and $2.85 for a year. The reported figure has run between $2.12 and $9.11.

What we do with the stakes. The model counts them. Alphabet's $90.4 billion of marketable equity securities sit inside its $265.1 billion of cash and marketable securities, at the 30 September price, and we add that pile at book, as July did. The $135.2 billion of non-marketable stakes is added after a 22% tax, the rate Alphabet itself applied to the June-quarter gain. Together they are worth about $15.77 a share of fair value, so the quarter's mark moved value by about 54 cents after tax, not by the dollars it moved earnings. Carrying the non-marketable stakes at zero would lower fair value to $215.15. SpaceX has traded since 30 September; we take the filed mark and do not re-price it daily.

Google Cloud revenue was $30.71 billion, up 103%, and its operating income $11.12 billion, a 36.2% margin against 23.7% a year earlier (Exhibit 2). The release says the growth was led by Google Cloud Platform and that "TPU systems sales were also a meaningful contributor." Neither gives a figure; the CFO said Google Cloud Platform accounted for the vast majority of the quarter's Cloud revenue. The backlog was $561 billion at 30 September, against $514 billion at 30 June. The CFO said "the majority" relates to "typical GCP contracts," and that Alphabet expects "to recognize just over 55% of the total backlog as revenues over the next 24 months." In June the figure was just over 50%. If the same share holds for Cloud, that is about $309 billion of Cloud revenue scheduled for the next eight quarters, against $93 billion recognized in the last four.

Two cautions sit beside the number. First, the margin faces headwinds. Alphabet began renting third-party computing capacity at the end of the quarter to meet demand it cannot yet serve itself; the CFO said "we expect margin headwinds in the near term as we utilize this capacity." Second, part of the growth is hardware. TPU systems are sold, not rented: the company says it builds them into inventory only after a contract is signed, and it bills customers as it delivers them. Inventory rose to $13.3 billion from $2.4 billion at the start of the year. Unless TPU systems earn chip-company margins, we assume a hardware sale is worth less per dollar of revenue than a cloud contract that renews. We do not know the share, so the sum-of-the-parts cross-check values TPU systems as their own line at an assumed 12% of 2027 Cloud revenue.

We built the 2027 base case from the backlog. Cloud revenue reaches $180 billion in 2027, against $160.5 billion in July's model, and grows 35% in 2028. Over the 24 months to September 2028 that puts about $391 billion through the segment, about 79% of it already under contract on the company's 55% statement; the rest is consumption outside the backlog. Search grows more slowly than in July's base case: 10% in 2027, not 13%. Search & other grew 19%, 17% and 12% in the three quarters of 2026, with currency and tougher comparisons named on the call, and the CFO warned that the fourth quarter laps the acceleration that began a year ago. On the final model, Cloud's path alone is worth about +$6.21: with July's $160.5 billion in the base case, fair value is $217.43. In the bridge, together with the slower Search, the revenue changes net +$0.53.

Capex was $51.36 billion, up 114%, or 40.3% of revenue. Operating cash flow was $42.29 billion, so free cash flow was −$9.07 billion, the second negative quarter in a row (Exhibit 3). The CFO noted that operating cash also carried the TPU inventory build and a $5.2 billion payment of a 2018 European Commission fine. For the twelve months, free cash flow was $19.7 billion against ex-marks net income of about $135 billion, so 15% of earnings reached shareholders as free cash. Capex ran 6.1 times depreciation in the quarter.

The guide fixes most of the fourth quarter. The CFO left 2026 capex at $195–205 billion, which after $131.95 billion in nine months leaves $63–73 billion for the fourth quarter, an annual pace of $252–292 billion. She repeated that capex will "increase significantly in 2027," with details on the fourth-quarter call, and gave no figure; press figures near $300 billion are not company guidance. July's engine carried $233 billion for 2027 even though management had already said "significantly higher." We now carry $285 billion in the base case, about 43% above 2026 and close to the fourth quarter's own pace, at a cost of $8.64 a share of fair value.

The bull carries $302 billion, the base's share of revenue on its larger revenue, because more demand needs more capacity. The bear carries the base's dollars on smaller revenue, because a build committed through suppliers and leases does not shrink within the year demand disappoints; the 10-Q's $213.4 billion of purchase commitments due within twelve months is one sign, though it includes items other than capex and the company calls the amounts maximums. A bear that cut faster, reaching terminal intensity over four years rather than six, would add about $2.64 to fair value.

The funding is now visible on the balance sheet. Long-term debt is $125.1 billion, from $46.5 billion at the start of the year, after $28.7 billion of new notes in the quarter. Alphabet sold $12.4 billion of new shares through its at-the-market program (36 million shares) and bought back none, against $39.8 billion in the first nine months of 2025. The CFO also gave the mix: about 65% of the quarter's technical-infrastructure spending was "servers and networking equipment, which generally have useful lives of six years." July assumed 55%. A larger server share means faster depreciation, a larger tax shield and, later, a larger replacement bill.

The commitments run further out still. The 10-Q, filed the day after the call, puts material purchase commitments at $924.4 billion at the end of September, plus $130.5 billion of data-center and energy leases that begin between 2026 and 2031 and $21.9 billion of funding commitments to equity investments. The purchase commitments are more than four times the 2026 capex guide but spread over several years: $213.4 billion falls due within twelve months, and most of the multi-year part by 2031. They cover technical infrastructure, inventory, energy and content licenses as well as open purchase orders, and the inventory part funds TPU systems, which the company says are built only against committed contracts. The CFO called them maximum contractual amounts, a number of them with "variable pricing and volume flexibility, which could result in lower spend."

Why the gap did not close. In the bridge, the new revenue path nets only +$0.53 (faster Cloud, slower Search; Cloud's path alone is worth about +$6.21 on the final model, §2), while the higher 2027 figure takes $8.64 and a slower fall from it takes $23.45 (about $13.53 on the final model), the second our reading of the commitment schedule rather than a fact of the quarter (§4). July's margin leg is still unread. Meeting it at 30% would add nothing to fair value (§7, tripwire 6), because our 44% terminal margin already carries Cloud at about its current 36% segment margin (§5); only two readings at or above 40% raise the number, and two below 30% cut it, so on this model July's backlog showing, with both legs met, no longer closes most of the gap, a change from July's statement. The gap widened mainly on the method: fair value is now the cash-flow value alone (−$33.13). Growth landing on the terminal rate (−$14.21) also lowered value; terminal growth from long-run GDP (+$9.22), the bull's longer Cloud stage (+$7.60), the filed balance sheet (+$7.50), the argued cost of equity (+$7.36), Other Bets carried as an asset (+$7.17) and the share count (+$4.12) went the other way.

The engine is July's: free cash flow to equity, stock compensation expensed, a cost of equity built on the capital-asset pricing model, depreciation vintage by vintage from capex, debt drawn to 0.75 times EBITDA. Before changing anything we re-ran July's configuration. It reproduces $256.42 to the cent: FCFE $223.29, comps $250.84 and SOTP $328.27, blended 50/25/25. The table gives today's scenarios; the bridge walks from July's number to today's, and the paragraphs after it argue each block.

How the methods are combined: −$33.13. The methods answer different questions. The cash-flow model asks what Alphabet's own future cash is worth at a cost of capital argued for this company. Comps and the sum of the parts ask what the market pays for peers and for businesses like Alphabet's segments; they inherit the market's discount rates and, in this cycle, a market pricing the same capital boom, so they cannot test whether the build earns its cost, which is the question this note is about. Damodaran treats intrinsic value and relative pricing as answers to different questions; a composite is a choice some practitioners make, and July made it, capping the relative methods at half. Cash conversion weighs against it now: free cash flow was 15% of ex-marks earnings over the last year, about 19% without the $5.2 billion fine payment, and about 27% with the nine-month TPU inventory build of $11.0 billion also added back, though that build is working capital for a product line, not a one-off.

The strongest case against the choice is that 59% of the base's operating value is terminal value, resting on a margin, a capex glide and a replacement cycle Alphabet does not date; our scenarios show that risk but do not hedge it, and averaging with the market's multiples is the usual hedge. We accept the cost because the question is whether the build earns its cost. We would revisit the weighting, as a judgment rather than a rule, once capex runs no more than 3 points of revenue above depreciation for four straight quarters (around 2033 on our base path), when the multiples are no longer pricing a build in progress. July's weights would give $276.96 today, so the choice, applied last, is −$53.32, about half of the $101.25 gap to the price. Dropping only the comps gives $272.27; an even split with the sum of the parts, $296.58.

The quarter's data and our reading of them: +$6.04, to $229.33. Rolling the valuation date to 28 October and booking nine months of filed results, with the larger cash pile and debt, adds +$7.50. The 2026 year and the higher server share move value +$2.05. Shares add +$4.12: we count the third quarter's diluted figure plus the convertible preferred as converted, at the maximum rate that applies below its initial price, 12,422 million in every scenario. July's projected counts added about 59 million at-the-market shares without their proceeds and about 258 million for suspended buybacks on top of stock compensation already expensed. The revenue path adds +$0.53, the 2027 capex figure takes $8.64, and a larger Cloud share of terminal capital intensity moves value +$0.48: Services at 10% of revenue, near Alphabet's whole-company level of 9.6–11.1% in 2021–23, and Cloud at 26%, a house estimate of a cloud provider at steady state.

The cost of equity: +$7.36. We argue the rate from Alphabet's situation. The 10-year Treasury was 5.05% on 28 October (4.67% in July), and we use a 4.5% equity risk premium. For the forecast years we use Alphabet's own 60-month beta against the S&P 500, 1.194, with a standard error of 0.20: a cost of equity of 10.4%. Its 30-month beta rose from 1.12 before the capex surge to 1.34 during it, and the same-cycle peers rose the same way (Microsoft 0.99 to 1.36; Meta 1.01 to 1.56; Amazon 1.39 to 1.55), but on windows that short the difference is within statistical noise (standard error about 0.5) and the build's expected cost is already in the cash flows, so we add no premium for it. Financial leverage adds little: debt is 3% of the equity's market value, and Alphabet's cash and debt securities exceed it by about $50 billion.

For the terminal value we use a beta of 1.10, a cost of equity of 10.0%, reached in a straight line across the 2029–33 fade: a mature advertising-and-cloud business, set from Alphabet's own pre-build 1.12. That treats the pre-build business as the mature one, a judgment the noise finding cannot test. Microsoft's, Meta's and Amazon's 60-month betas describe businesses still mid-build. Their pre-build reads are the strongest case for a lower terminal beta (median 1.01, Meta's 1.01), though their mean, 1.13, sits near Alphabet's own, and cyclical advertising keeps the rate above the market's. The higher Treasury alone would take $12.76 a share; the beta judgments add back $20.12. Those three peers' own 60-month betas would give 9.9%, 10.3% and 11.5%, unused. The other reads:

What changed in the cash-flow model, and why: −$13.05 in total. Six changes, in the order of the bridge.

1. Terminal growth from long-run GDP, +$9.22. The Federal Reserve's projections of 16 September 2026 give 2.0% real growth and 2.0% inflation in the longer run, so 4.0% in the base and bull cases and 3.0% in the bear (a share-loss case), against July's 3.5%, 4.0% and 2.75%. July listed 4% growth forever among the Street's conventions, labelled there as nominal GDP-plus; this note sets 4.0% from the Federal Reserve's longer-run medians instead, and the Congressional Budget Office's lower 1.8% potential growth (about 3.8% nominal at 2% inflation) is the cross-check.
2. Growth lands on the terminal rate, −$14.21. July's forecast ended in 2032 with base growth at 7.6% and then dropped to 3.5% in one step. Each business line now falls in a straight line from its 2028 rate to terminal growth by 2033.
3. A slower fall in capex intensity, −$23.45. Intensity now falls in a straight line from its 2027 peak to terminal over six years, to 2033, rather than July's four, to 2031. The length comes from the company's commitment schedule: the 10-Q says the significant majority of $802 billion of multi-year commitments is fulfilled through 2031, so spending stays high for years after 2027. Six years is also one life of the servers that make up 65% of technical-infrastructure spend, and a server's life does not depend on the scenario, so the bull carries the same shape. Revenue keeps growing, so capex in dollars still rises after 2027, peaking at about $301 billion in 2029 in the base and $334 billion in 2029 in the bull, then falling to about $193 billion in the base by 2033. Management has dated neither a decline after 2027 nor the 2028–29 spending we carry. A reader who expects a faster fall sits below our base on capex; July's four-year shape on today's model gives $237.17.
4. The bull's Cloud stage runs a year longer, +$7.60. With terminal growth capped at long-run GDP, stage length is where the bull's growth upside lives, and the backlog supports the year for Cloud: about 45% of the $561 billion, some $252 billion, is due beyond 24 months. The other lines, and the base and bear, keep 2033.
5. Other Bets is carried as an asset, +$7.17. Its revenue and its operating loss, 1.7% of revenue over nine months, leave the consolidated paths, and the business is added at house estimates of $75 billion in the base ($20 billion bear, $175 billion bull), so it is counted once. Charging the losses through the margin path while valuing the assets at zero was inconsistent.
6. The forecast runs to steady state, +$0.62. Ending in 2033 left the terminal year carrying the build's depreciation, 17.8% of base revenue and above the 17.3% the revenue mix implies, which forced terminal capex up to that level. Seven more years at terminal growth and intensity let the peak vintages finish depreciating; depreciation settles at 12.5% of revenue by 2040, and terminal capex is the mix level (Exhibit 5). The terminal year then reinvests 19% of after-tax operating profit to grow 4%, an implied 21% return on new capital (24% bull, 10% bear) against a trailing 30.1%.

Capex is still the input with the least guidance behind it. Two points lower or higher at terminal moves fair value to $241 or $206, and a 2027 figure of $260 billion or $320 billion moves it to $231 or $216.

The cross-checks, re-marked. July's peer multiples were captured on 24 July. On the six peers' 28 October closes and their filings available by then, the median EV/EBITDA is 18.2 times (16.1 in July) and EV/revenue 9.1 times (8.1). The forward P/E method drops out: we hold no dated consensus earnings capture for it. Comps now give $291.02. The sum of the parts gives $369.53, with July's segment multiples scaled by the change in the peer EV/EBITDA median (×1.13), TPU systems valued as hardware at two to four times sales instead of at Cloud's five to eleven, and the Alphabet-level AI research cost re-scaled to its trailing loss. Neither moves fair value.

How long each part of the business grows faster than normal. Each stage's length comes from what drives it, not from a default.

Return on capital. Trailing return on invested capital is 30.1% against the 10.2% WACC, three times the hurdle, and the regime is fading: the spread fell 6.6 points in four quarters and 12.2 over the twelve quarters shown (Exhibit 6). Average invested capital grew 55% in a year on the debt-funded build, the new shares and the stake gains kept in equity, while operating income grew 27%. The trend matters more than the level. Capital still under construction is already in the denominator, but neither its depreciation nor the revenue it is built to serve has reached the numerator. Which arrives first sets the next turn of the trend.

Terminal value is 59% of the base case's operating value (57% in the bull, 66% in the bear), so the price, like our value, is mostly a claim about the years after the build. Each line below moves one lever and holds the rest of our inputs; the fair-value column moves it in all three scenarios and starts from a lower value than the base case ($223.64 against $231.73), which is why it asks more of the margin.

The margin is the lever we can least see. The base carries a 44% terminal EBITDA margin, against 40.0% over the first nine months of 2026. We set it from the mix: Cloud is 46% of base revenue by 2033 and earns a 36% segment operating margin after depreciation, so its EBITDA margin is higher still, less the Alphabet-level AI research cost. Alphabet does not disclose depreciation by segment, so this level is a judgment, and what we cannot see favors the price: the more of the build's depreciation sits in Cloud, the further Cloud's EBITDA margin sits above its operating margin. Even so, our base ends 2040 at a 31.5% operating margin, below the 31.8% of the third quarter. The 42–46% sensitivity ($215 to $233) is where our judgment sits. At the base case's 54.4% alone, fair value would be $270.23; that margin becomes 48.1% at a cost of equity one point lower and 60.9% at one point higher (explicit and terminal rates moved together). A segment depreciation disclosure would settle the level, and we would re-set the margin from it.

Two bets. The price is therefore close to one of two bets. One is that our inputs for the build understate value by enough to reach the price: the terminal margin or the cost of equity, alone or with a faster capex glide. The sum of the parts says so: it is above the price, though comps, at $291.02, are below it. The other is that our own bull path is close to certain: with no weight on the bear the price needs about 88% on the bull, and any bear weight above about 5.4% puts it out of reach on our three cases.

Three depend on the 2027 capex guide due in early February and are pending. The backlog test is a waypoint: its conversion leg reads as met on the company's own statement, its margin leg is not yet readable. The other four did not fire.

Each reading rule is decided now, so the next update scores it rather than reinterprets it. Effects are on fair value at today's other inputs.

1. Backlog and the TPU split (each 10-Q; the FY2026 10-K, about February 2027). The Q3 10-Q, filed 29 October, confirmed the call: $560.8 billion of Cloud remaining performance obligations, and just over 55% of the $567.7 billion total backlog due within 24 months. It gave no dollar split of Cloud product revenue. A 24-month share below 50% in any later filing is the first sign the backlog stage is shorter than argued. If a filing separates Cloud product revenue (TPU systems), the sum-of-the-parts cross-check is re-sized ($359 at 25% of Cloud, $379 at zero); fair value moves only if the split changes our terminal margin. No split leaves the house 12% in place.
2. Fourth-quarter capex (Q4 release, about 3 February 2027). Inside $63–73 billion holds the base. Above $73 billion means the 2027 build is starting early; the base 2027 figure moves to $300 billion.
3. The 2027 capex guide (same call). At or below $260 billion: base, bear and bull all re-run at $260 billion, fair value about $231, and July's "above $260 billion" test does not fire. Between $260 billion and $320 billion: the base holds. At $320 billion: about $216, and lower above it. No number on the call: the base holds, and the next update says so.
4. Cloud revenue in 2027 (FY2027 release, February 2028; waypoints every quarter). The base carries $180 billion. Below $155 billion, the backlog stage was shorter than argued, and fair value moves to about $216. Above $205 billion, about $232. Each quarter's Cloud revenue and 24-month share are recorded against the path; the fourth quarter's house figure is $35.3 billion.
5. Search pace (Q4-26 and Q1-27 releases, by 30 April 2027). Both quarters below 8%: base Search moves to the bear's 7% and 6%, about $219. Both above 12%: base Search moves to the bull's 13% and 11%, about $227. July's below-10% test stays armed alongside.
6. Cloud margin after the bridging period. Margins in the bridging quarters (Q4-26 and Q1-27) are recorded against the CFO's guided headwind, not scored. Every margin reading uses the first two quarters after the rented capacity is fully in use, as the CFO states it, or Q2-27 and Q3-27 if no statement has been made by the Q1-27 call. In those two quarters, both below 30% fails the July margin leg and moves the base terminal EBITDA margin from 44% to 42% (about $215). Both at or above 30% meets the July leg. Both at or above 40% moves it to 46% (about $233).
7. The 2028 capex guide (FY2027 release, about February 2028). The first read on the glide, the largest judgment after the weighting. We score capex's share of revenue, not its dollars, because every scenario carries the same shape on different revenue: the guide's midpoint over consensus 2028 revenue on that day, never our own path, against 2027's reported share; reported 2028 figures settle it in February 2029. On its own revenue the base falls 5.0 points in 2028, from 47.4% to 42.4%, about $297 billion (bull 5.2 points, bear 4.9). A fall of about 7.5 points or more is July's four-year pace, and the glide re-runs on that shape, about $237. A fall of about 3.8 points or less is an eight-year pace, about $211. In between, or with no number on the call, the base holds.
8. Price. A close at or below $223.64 resolves the gap. $168 is a 25% margin of safety to fair value. Replaces July's $192.

We value Alphabet at $223.64 a share, from $256.42. The quarter brought half of what our July note was waiting for, and the backlog lifted value. The same quarter set a capital bill that leaves free cash flow negative and has the company borrowing and selling stock. That bill is larger than July's model carried, and we now assume capex intensity falls over six years, in line with the company's own commitment schedule (−$23.45), a decline management has not dated. Our argued cost of equity moved value +$7.36. The largest change is how the methods are combined (−$33.13): we value the company on its own cash flows and show the market's multiples beside them. To reach $324.89 on one lever alone, fair value needs a cost of equity near 8.6% or terminal margins about 12 points above ours in every scenario (the base case alone needs 8.7% or 54.4%); smaller moves in both together reach it too. The other routes are that the sum of the parts, which is above the price, is the better measure, or that our own bull path, worth $338.17, is close to certain. The fourth-quarter call, with the 2027 capex number, is the next test. The Fund holds no position in GOOGL.