From Ritu Raj | Product & Market Analysis

Alphabet Capex 2026: $200B of Spend Becomes a $49B Depreciation Bill in 2027

On this page

Alphabet will spend $195 billion to $205 billion on capital expenditure in 2026, and almost none of it has reached the income statement yet. Using only the asset lives in Alphabet's own 10-K, the 2026 build adds about $23.8 billion of depreciation a year once in service. Our model puts total Alphabet depreciation near $49 billion in 2027, against $21.1 billion in 2025.

Key takeaways

  • Every $100 billion of Alphabet capex adds about $11.9 billion of yearly depreciation. That rate comes from the 6-year server life and the 60% to 65% server share Alphabet discloses, with buildings at a 25-year midpoint of its 7 to 40 year range.
  • Depreciation is already accelerating. Alphabet's Q3 2026 depreciation was $8.4 billion, up 49% on Q3 2025, derived from its 10-Q cash flow statements.
  • Our 2027 depreciation estimate is about $49 billion, 2.3 times the 2025 figure. The year-on-year increase of roughly $18 billion equals about 11% of Alphabet's annualised Q3 2026 operating income.
  • Server life is the assumption that matters. Moving servers from 6 to 5 years adds about $4.2 billion a year to the 2026 vintage alone, while the building life range moves it by less than $3.1 billion.
$195B-$205BAlphabet 2026 capex guidance, raised in July and held in October. Source: Q2 and Q3 2026 earnings calls, via CNBC.
+49%Q3 2026 depreciation versus Q3 2025, $8.4B against $5.6B. Source: Alphabet Q2 and Q3 2026 10-Q filings, our subtraction.
$11.9BAnnual depreciation added per $100B of Alphabet capex, once in service. Source: our model from Alphabet's disclosed useful lives.

The Alphabet capex and depreciation inputs that are actually disclosed

A depreciation model is only as good as its inputs, so this one uses nothing Alphabet has not published. Every figure in this section comes from a filing or from management's own statements on an earnings call.

Depreciation is the accounting process that spreads the cost of a long-lived asset across the years it is expected to be used. A server bought for $6 million and given a 6-year life costs $1 million a year in reported expense, regardless of when the cash left.

The capex guidance path through 2026

Alphabet entered 2026 with a capex plan that kept growing. The $180 billion to $190 billion range was given on the Q1 call, as restated in the June 2026 equity offering release. On 22 July, after reporting $44.9 billion of Q2 capex, CFO Anat Ashkenazi raised the range to $195 billion to $205 billion.

The Q3 report on 28 October left that range unchanged after Q3 capex of $51.4 billion. Nine-month cash capex reached $132.0 billion. Subtract that from the guided range and the implied Q4 capex is $63 billion to $73 billion, the largest quarter of the year by a wide margin.

Management has also said 2027 capex will "increase significantly", with detail promised on the Q4 call. No number exists yet, which is why the 2027 vintage appears as two scenarios below rather than one figure.

Asset lives and the mix of spend

Alphabet's FY2025 10-K depreciates servers and network equipment over six years and buildings over 7 to 40 years. Land is not depreciated. Depreciation starts only when an asset is ready for its intended use, so construction in progress sits on the balance sheet at zero expense.

The mix comes from the calls. In Q2 2026 about 60% of technical infrastructure investment went to servers and 40% to data centres and networking. In Q3 the split was about 65% to servers and networking and 35% to data centres. The categories differ slightly between quarters, so we use 62.5% as the short-lived share and test both ends.

Depreciation itself is disclosed annually and in cumulative quarterly cash flow lines. It was $15.3 billion in 2024 and $21.1 billion in 2025. For the first nine months of 2026 it was $22.0 billion, up 45% on the same period of 2025.

Disclosed inputs used in the model
InputFigureWhere it comes from
Capex 2024 / 2025$52.5B / $91.4BFY2025 10-K, MD&A
Capex guidance 2026$195B to $205B (model uses $200B)Q2 call, reaffirmed on the Q3 call
Depreciation 2024 / 2025$15.3B / $21.1BFY2025 10-K, MD&A
Depreciation, nine months 2026$22.0BQ3 2026 10-Q, cash flow statement
Server and network useful life6 yearsFY2025 10-K, accounting policies
Building useful life7 to 40 years (model uses 25)FY2025 10-K, accounting policies
Short-lived share of spend60% to 65% (model uses 62.5%)Q2 and Q3 2026 earnings calls

The 25-year building life and the $200B midpoint are our choices, not Alphabet disclosures. Both are tested in the sensitivity table below.

Every $100B of Alphabet capex adds about $11.9B of annual depreciation

The core of the model is one rate: how much yearly expense a dollar of capex creates once it is in service. Everything else is timing.

The arithmetic is short. Servers and network gear take 62.5% of spend and depreciate over 6 years, which is 10.42 cents per dollar per year. Data centres take 37.5% and, at a 25-year life, add 1.50 cents. Together that is 11.92 cents, or about $11.9 billion a year for every $100 billion spent.

Apply that to each year of spend and the vintages stack up fast. The 2024 vintage of $52.5 billion carries about $6.3 billion a year. The 2025 vintage of $91.4 billion carries $10.9 billion. The 2026 vintage at $200 billion carries $23.8 billion, which on its own exceeds Alphabet's entire 2025 depreciation charge of $21.1 billion.

That last comparison is the single most useful sentence in this post. One year of the current build adds more annual cost than every asset Alphabet owned at the end of 2025 was generating.

Annual depreciation each year of Alphabet capex adds, once in service $ billions per year. 11.9 cents per capex dollar: 62.5% servers over 6 years, 37.5% buildings over 25. 2024 vintage ($52.5B)$6.3B 2025 vintage ($91.4B)$10.9B 2026 vintage ($200B)$23.8B 2027 vintage, flat$23.8B 2027 vintage, +25%$29.8B Total 2025 depreciation: $21.1B 2027 vintages are illustrative scenarios. Alphabet has said only that 2027 capex will rise significantly. Source: Alphabet FY2025 10-K, Q2 and Q3 2026 calls. Zan Digital calculation.
The dashed line is everything Alphabet depreciated in 2025. The 2026 vintage crosses it alone, before any 2027 spend is counted.

Why server life does most of the work

The short-lived share dominates the rate. Servers contribute 10.42 of the 11.92 cents, roughly 87% of the total. So the sensitivity of the whole model sits almost entirely in one accounting estimate: the six-year server life.

Alphabet has moved that estimate before. In January 2023 it extended servers and certain network equipment to six years, which it said would cut 2023 depreciation by about $3.4 billion for assets already in service, per its Q1 2023 disclosure. How that choice compares across the other hyperscalers is laid out in the comparison of server useful lives in big tech 10-Ks.

Sensitivity: depreciation per $100B of capex and for the 2026 vintage
Assumption changedCents per capex dollar2026 vintage, $ per yearChange vs base
Base: 62.5% servers at 6 years, buildings at 2511.9$23.8BNone
Servers at 5 years14.0$28.0B+$4.2B
Servers at 4 years17.1$34.3B+$10.5B
Server share 65%12.2$24.5B+$0.7B
Server share 60%11.6$23.2B-$0.6B
Buildings at 15 years12.9$25.8B+$2.0B
Buildings at 40 years11.4$22.7B-$1.1B

All rows are Zan Digital arithmetic on Alphabet's disclosed lives and mix. A 4-year server life is the estimate Alphabet used before 2023, shown to size the stake in that judgement, not as a prediction.

When Alphabet capex reaches the income statement

The rate tells you how big the bill is. Timing tells you when it arrives, and timing is where most public estimates go wrong.

Cash leaves when Alphabet pays for equipment or construction. Depreciation starts only when the asset is placed in service. For a GPU or TPU rack that gap can be weeks. For a new data centre campus it can be years, because the building and its power connection must be finished first.

The ceiling case: in service the year it is bought

Assume every dollar enters service in the year it is spent, on average at midyear. Then each vintage contributes half its annual rate in year one and the full rate after. On that basis 2026 depreciation would be $21.1 billion plus half of the 2025 vintage plus half of the 2026 vintage, or about $38.5 billion.

That is too high, and the filings already prove it. Nine-month depreciation was $22.0 billion. Even if Q4 doubled Q3's $8.4 billion, the year would close near $39 billion, and nothing in the quarterly trend points to a doubling.

The lagged case: in service a year after the cash

Now assume each vintage enters service around the middle of the following year. On that basis 2026 depreciation is $21.1 billion plus the second half of the 2024 vintage plus the first half of the 2025 vintage, or about $29.7 billion.

That is close to what is happening. If Q4 2026 depreciation merely matches Q3's $8.4 billion, the year closes at $30.3 billion. If it keeps rising at its recent pace, our estimate is nearer $31.4 billion. Either way the actual sits just above the lagged case and far below the ceiling.

The finding is worth stating plainly. On Alphabet's own numbers, capex is reaching the income statement roughly a year after the cash leaves. That lag is the reason 2026 margins held up while capex more than doubled, and it is also why 2027 is where the bill lands.

The Alphabet depreciation bill for 2027 and 2028

Roll both cases forward and the range is wide but bounded. For 2027 we hold capex flat at $200 billion in one scenario and raise it 25% to $250 billion in the other, since management has said only that it will rise significantly.

Modelled Alphabet depreciation, $ billions
YearLagged case (floor)Calibrated centralCeiling caseBasis
202521.121.121.1Reported, FY2025 10-K
202629.730.3 to 31.438.5$22.0B reported for nine months
2027, capex flat at $200B47.048 to 5062.3Model
2027, capex at $250B47.048 to 5065.3Model
2028, capex flat at $200B70.9About 7386.1Model, 2028 capex equal to 2027

Model rows exclude assets that finish depreciating and drop out, which slightly overstates each figure. In the lagged case 2027 capex has not entered service by the end of 2027, so both 2027 scenarios give the same floor.

The calibrated central case

The lagged case slightly understates reality, so we calibrate it. Actual 2026 depreciation of $30.3 billion to $31.4 billion closes 7% to 19% of the $8.8 billion gap between the lagged and ceiling cases. Applying the same share to the 2027 gap of $15.3 billion adds $1.0 billion to $2.9 billion to the floor.

That gives 2027 depreciation of $48 billion to $50 billion. We use $49 billion as the central figure. It is 2.3 times the $21.1 billion of 2025, and it does not depend on the size of 2027 capex, because most 2027 spend will not be in service until 2028.

The 2028 picture is harder still. If 2027 capex simply equals 2026, depreciation reaches roughly $73 billion in 2028 on the same calibration. That is more than three times 2025, two years after the spend that caused it.

Alphabet depreciation: reported, then modelled $ billions. Light bars reported; dark bars Zan Digital central case. Whiskers show floor to ceiling. 15.32024 21.12025 ~312026 est. 492027 model ~732028 model Whisker tops: 2027 ceiling $62B (flat capex), 2028 ceiling $86B. Source: Alphabet 10-K and 10-Qs; Zan Digital model.
2026 looks manageable because most of the build has not entered service. The 2027 column is where the July guidance raise shows up in reported profit.

What $49B of depreciation does to Alphabet margins

A larger depreciation line is not a problem in itself. It becomes one only if revenue does not grow fast enough to absorb it, so the useful comparison is against Alphabet's current earning power.

Alphabet reported Q3 2026 revenue of $127.6 billion and operating income of $40.6 billion, a 31.8% margin. Annualised, that is about $510 billion of revenue and $162 billion of operating income. Our modelled step-up from 2026 to 2027 is about $18 billion.

At flat revenue that increase would take roughly 3.5 points off the operating margin, and it equals about 11% of annualised operating income. Revenue will not be flat, since Q3 revenue grew 25% year on year. The point is the size of the hurdle: Alphabet has to add roughly $18 billion of gross profit in 2027 before reported operating income rises at all.

That step is also bigger than the one just absorbed. Depreciation rose about $10 billion from 2025 to 2026. The 2027 increase is nearly double that, and the 2028 increase in our central case is larger again at roughly $24 billion.

We think the market is still reading Alphabet's margins off 2026, which is the wrong year. The 2026 operating margin reflects capex decisions made in 2024 and 2025, when annual spend was a quarter to a half of today's level.

Google Cloud carries part of the load

Depreciation is spread across cost of revenue, research and development and other lines, so no segment wears all of it. Google Cloud is the segment selling the capacity directly. It reported Q3 2026 revenue of $30.7 billion, roughly double a year earlier, and operating income of $11.1 billion.

Cloud's margin expansion is the strongest counterweight to this model. Whether it survives the 2027 depreciation step is the subject of the analysis of Google Cloud's operating margin against rising depreciation. The wider cash side of the same build, including the June equity raise, is covered in the breakdown of Alphabet's $80 billion equity raise.

What the October 28 Alphabet report already settled

This model was planned as a preview of the 28 October report. That report is now public, and it answered three of the questions a preview would have asked.

First, guidance held. Alphabet left the $195 billion to $205 billion range unchanged and repeated that 2027 capex will rise significantly. The implied Q4 of $63 billion to $73 billion means the 2026 vintage is back-weighted, which pushes more of its depreciation into 2027 and 2028.

Second, depreciation accelerated. Q3 depreciation grew 49% year on year, faster than the 45% for the nine months. The trend is the one the lagged model predicts as the 2025 vintage enters service.

Third, cash turned negative. CNBC reported Q3 free cash flow of minus $9.07 billion. Part of that was a $5.2 billion cash payment for a 2018 European Commission fine, so the underlying figure was less negative. The direction is still clear, and it matches the pattern traced in the comparison of hyperscaler capex against operating cash flow.

The profit miss itself came mostly from an $8.2 billion unrealised loss on Alphabet's SpaceX stake in other income, according to Investing.com's call coverage. That line is below operating income and says nothing about depreciation. Do not read the share price reaction as a verdict on the build.

Where this depreciation model is weakest

Every number above is arithmetic on disclosed inputs, and the arithmetic is the easy part. The judgement calls are where the model can fail.

It ignores roll-off and treats land as depreciable

Assets bought six years ago finish depreciating and stop adding expense. We do not subtract them, because Alphabet's useful lives changed in 2021 and 2023 and the vintage-by-vintage roll-off cannot be rebuilt from public filings. Spend from that period was a fraction of today's, so the overstatement is small, but it is real.

The model also applies the building rate to the full 37.5% non-server share. Some of that is land, which is never depreciated. Both errors push our figures up, which is why we would treat the central case as an upper-middle estimate rather than a midpoint.

The useful life is a choice Alphabet can change

The largest risk to the model is not arithmetic. It is a change to the 6-year server life. A shorter life, if AI accelerators wear out or obsolesce faster than general servers, would add about $4.2 billion a year per year of spend at 5 years. A longer life would do the opposite.

The case against this model is strong here. Alphabet has extended server lives twice in recent years and has not shortened them. Nothing in its 2026 filings signals a change. The question of whether accelerators deserve a different life from general servers is examined in the analysis of GPU depreciation and AI earnings. We think the six-year estimate is generous for accelerators, but we cannot prove it, and the model does not assume it changes.

Finally, capex is not all depreciation. Alphabet disclosed $58.5 billion of data centre leases not yet commenced at the end of 2025, starting between 2026 and 2031. Those costs arrive as lease expense, not depreciation, so this model understates total infrastructure cost even as it may overstate depreciation alone.

Lines to read in Alphabet's Q4 report and 2026 10-K

The Q4 report and the 10-K that follows will either confirm this model or break it. These are the specific lines that decide which.

Where to look, and what would change the estimate
LineWhereWhat would move the 2027 estimate
Full-year 2026 depreciation10-K, MD&A and cash flow statementAbove $32B means a shorter lag; raise the 2027 figure toward $52B or more
2027 capex guidanceQ4 earnings callMatters mainly for 2028; a figure above $250B lifts the 2028 ceiling past $95B
Server useful life10-K, property and equipment policyAny move from 6 years reprices every vintage; 5 years adds about $4.2B per $200B
Assets not yet in service10-K, property and equipment noteA large balance is depreciation queued for 2027 and 2028
Capex in accrued liabilitiesCash flow supplemental disclosure$35.9B at 30 September 2026, up from $14.6B a year earlier; equipment received but not yet paid for
Leases not yet commenced10-K, MD&AGrowth above the $58.5B at end-2025 adds fixed cost this model excludes

The accrued capex line deserves more attention than it gets. Net property and equipment rose $126.2 billion in nine months, while cash capex was $132.0 billion and depreciation $22.0 billion. Assets are arriving faster than the cash line shows, and every unpaid but delivered asset is one step closer to depreciation.

Frequently asked questions

What is Alphabet's capex guidance for 2026?

Alphabet guides 2026 capital expenditure to $195 billion to $205 billion. It raised the range from $180 billion to $190 billion on its Q2 call on 22 July 2026 and left it unchanged on its Q3 call on 28 October. Capex was $44.9 billion in Q2 and $51.4 billion in Q3, which implies Q4 spending of roughly $63 billion to $73 billion.

How much will Alphabet's depreciation increase in 2027?

Our model puts Alphabet's 2027 depreciation near $49 billion, against about $31 billion estimated for 2026 and $21.1 billion reported for 2025. The estimate uses Alphabet's disclosed 6-year server life, a 25-year building life within its 7 to 40 year range, and a one-year lag between spending and service, calibrated to 2026 actuals.

What useful life does Alphabet use for servers?

Alphabet depreciates servers and network equipment over six years, according to its FY2025 10-K. It extended that estimate in January 2023, which it said would reduce 2023 depreciation by about $3.4 billion. Data centre buildings are depreciated over 7 to 40 years, and land is not depreciated at all.

Why does capex take time to hit Alphabet's earnings?

Depreciation starts only when an asset is ready for use, not when it is paid for. Data centres need construction and power connections before servers inside them run. Alphabet's 2026 depreciation of about $30 billion fits a model where spending reaches the income statement roughly a year after the cash leaves, which pushes the 2026 build into 2027 and 2028 profit.

How does Alphabet capex compare to its operating cash flow?

Alphabet's capex now exceeds its operating cash flow in some quarters. CNBC reported Q3 2026 free cash flow of minus $9.07 billion, which included a $5.2 billion payment for a 2018 European Commission fine. Alphabet also announced an $80 billion equity raise in June 2026 to help fund infrastructure commitments.

Is hyperscaler AI infrastructure spending going to hurt margins?

Only if revenue grows more slowly than depreciation. For Alphabet, our modelled 2027 depreciation increase of about $18 billion equals roughly 3.5 points of operating margin at Q3 2026's annualised revenue. Revenue grew 25% in Q3, so the margin outcome depends on whether that pace holds while the 2026 build enters service.

How to use this model before the Q4 report

If you own, cover or sell to Alphabet, rebuild this in a spreadsheet with four inputs: annual capex, server share, server life and lag. It takes 20 minutes and it will tell you more about 2027 earnings than any capex headline.

Then set one test for the 10-K. If full-year 2026 depreciation comes in above $32 billion, the lag is shorter than our model assumes and the 2027 bill is closer to $52 billion than $49 billion. Write the number down now, so you are checking a prediction rather than rationalising a result.

Related in this series

The sector-wide version of this question starts with the payback math missing from 2026 AI capex. The lease side that this model excludes is covered in the breakdown of leases not yet commenced.

References

  1. Alphabet Inc., Form 10-K for fiscal year 2025, filed 2026. Used for 2024 and 2025 capex and depreciation, useful-life policy, server share of technical infrastructure, and leases not yet commenced.
  2. Alphabet Inc., Form 10-Q for the quarter ended 30 September 2026. Used for nine-month capex, depreciation, net property and equipment, accrued capex, Q3 revenue and operating income.
  3. Alphabet Inc., Form 10-Q for the quarter ended 30 June 2026. Used for six-month capex and depreciation, from which Q3 figures were derived.
  4. Alphabet Inc., Equity offering release, exhibit 99.1, 1 June 2026. Used for the $180B to $190B range given on the Q1 call.
  5. The Next Web, Alphabet lifts capex guide to $205bn as Google Cloud jumps 82%, July 2026. Used for Q2 capex and the guidance raise. Secondary coverage of the Q2 call.
  6. CNBC, Alphabet Q3 earnings miss estimates. What analysts say, 29 October 2026. Used for unchanged guidance and Q3 free cash flow.
  7. Investing.com, Alphabet Q3 2026 earnings call transcript, 28 October 2026. Used for Q3 capex, the server and data centre split, Cloud results and the SpaceX loss.
  8. Alphabet Inc., Q1 2023 exhibit on useful-life change, April 2023. Used for the 2023 server life extension and its $3.4B effect.

Weakest point in the source base: the server and data centre split comes from call commentary relayed by secondary coverage, not from a filing, and the categories differ between Q2 and Q3. All model outputs are Zan Digital calculations, not company guidance. Figures last verified 12 November 2026.

RR
Ritu Raj
Writes for Zan Digital about AI infrastructure economics, hyperscaler filings and what the numbers behind capex announcements actually imply.

Related reading