indiconomics.Research library

Research article · 7 September 2026

The Server’s Second Life

Microsoft and Alphabet each reported a $3 billion net-income benefit from longer equipment lives in 2023. Their growing infrastructure makes those estimates more consequential, but the old effects cannot simply be scaled up.

The life of a server can change on a financial statement before anything changes in the building that contains it. A company revises how long it expects to use equipment, the annual charge allocated to that equipment changes, and reported profit changes with it. Whether the revision reflects better use of the machine or an estimate that later proves optimistic is a separate question.

Microsoft extended the estimated lives of server and network equipment from four to six years in FY2023. It reported a $3.7 billion increase in operating income and a $3 billion increase in net income. Alphabet extended servers from four to six years and certain network equipment from five to six in 2023, reporting $3.9 billion less depreciation and $3 billion more net income. [1] [2]

The identical rounded net-income effects make a striking comparison. What matters more is what sits beneath them: different reporting periods and asset populations, an allocation of past spending rather than a new cash receipt, and an estimate whose relevance changes as more infrastructure enters service.

Four per cent of which earnings?

Microsoft’s $3 billion effect was about 4.1% of its $72.361 billion reported net income for the year ended June 2023. Alphabet’s separate $3 billion effect was also about 4.1% of $73.795 billion for the year ended December 2023. Microsoft’s pre-tax effect was about 4.2% of $88.523 billion in operating income. [1] [2]

Company and reporting periodDisclosed net-income effectReported net incomeEffect / reported income
Microsoft FY2023, ended June$3.0bn$72.361bn4.1%
Alphabet 2023, ended December$3.0bn$73.795bn4.1%

Indiconomics calculations from company disclosures. Each denominator includes the effect. Numerators are rounded; these ratios are not shares of earnings growth, estimates of AI profit or findings of manipulation.

The similarity does not establish a common industry-sized effect. Microsoft described its FY2023 estimate using equipment carried at June 2022; Alphabet’s calculation also included assets placed in service during 2023. The firms’ different fiscal years further limit the comparison. Neither a matching rounded ratio nor these differences can establish whether their operational reasons for longer lives were the same.

The pre-tax and after-tax numbers also cannot settle that question. The gap between each pre-tax and net-income effect is about 18.9% of the pre-tax effect for Microsoft and 23.1% for Alphabet. A company-wide effective tax rate need not equal the tax rate applicable to a particular expense change. These ratios therefore do not demonstrate that Alphabet’s disclosed net effect had some non-tax explanation.

The spending grew; the measures still differ

Microsoft’s cash additions to property and equipment rose from $28.1 billion in FY2023 to $115.9 billion in FY2026, about 4.1 times as much. Total depreciation expense rose from $11 billion to $34.3 billion, about 3.1 times as much. Finance-lease additions sit outside that cash-spending line. [1] [3]

Alphabet’s purchases of property and equipment were $32.3 billion in 2023, $91.4 billion in 2025 and $80.6 billion in the first half of 2026. Its net property and equipment rose from $134.3 billion at December 2023 to $321.2 billion at June 2026, about 2.4 times as much. [2] [4] [5]

These figures describe cash investment, annual depreciation and an asset stock after accumulated depreciation. None measures a uniform stock of servers governed by one life. Buildings, land, equipment and unfinished construction can all sit inside broader property accounts. Multiplying the historical $3 billion effect by any of these growth rates would manufacture a current estimate the companies have not supplied.

The scale still has an economic consequence. As more assets enter use, errors in estimated service lives can affect a larger stream of reported expenses. Assessing that exposure requires the mix, age and deployment of the equipment, not just the total capital bill.

A range is not a reversal

Microsoft’s FY2026 policy note describes servers and network equipment as generally having lives of two to six years. It is tempting to treat the two-year end as a retreat from the six-year extension. The disclosure history does not support that inference.

In FY2022, the broader category of computer equipment carried a two-to-four-year range. FY2023 changed that range to two to six years, alongside the specific server and network extension. FY2024 and FY2025 retained the broader range; FY2026 attached it to the narrower label. [1] [3] [6]

The two-year floor therefore predates the extension. The later table is not evidence of a universal reversal, but neither does it prove that every individual server estimate remained unchanged. Alphabet’s 2025 policy describes servers and network equipment as generally lasting six years; its June 2026 policy note identifies no new server-life revision. [4] [5] These are bounded readings of policy disclosures, not observations of how long every machine actually runs.

The shell and the silicon

Microsoft did announce another estimate change on its July 2026 earnings call: from the start of FY2027, estimated lives for data centres and office buildings would increase from fifteen to twenty-five years. Management expected a minimal FY2027 operating-income benefit, while the larger effect would involve lease classification and reported capital expenditure. On the same call, it said roughly two-thirds of the quarter’s capex, including finance leases, had gone to short-lived assets, primarily CPUs and GPUs. [7]

A building and a processor can remain useful for very different periods. The new building estimate does not establish another server-life extension, and the description of recent spending does not reveal the retirement age of each accelerator. It does explain why one depreciation headline is an inadequate account of the whole infrastructure investment.

Alphabet supplies another distinction: $122.8 billion of its property and equipment was not yet in service at June 2026, compared with $50.6 billion at December 2024. Depreciation begins when assets are ready for their intended use. [4] [5] Moving depreciable assets into service will add expense, other things equal, but not all are servers and not all will enter service together. Their eventual charges depend on the asset mix, applicable lives, residual values and timing.

One purchase, two allocations

Consider a hypothetical machine costing 120 units. With straight-line depreciation, no residual value and a full year of use, a four-year life produces an annual charge of 30 and a six-year life produces 20. The outlay remains 120. Holding everything else constant, the longer schedule raises early accounting profit and spreads the same assumed cost over more years. This illustration excludes tax, impairment, disposal and financing; it is not company data.

Revising the estimate for equipment already in service differs from selecting a life for a new purchase. Reconstructing either company’s revision needs carrying amounts, asset cohorts and timing. Alphabet states that useful-life estimate changes are recognised prospectively. [4] The simple machine example explains the allocation mechanism, not the firms’ reported effects.

Microsoft attributed its extension to software efficiencies and technological advances. [1] That explanation deserves a fair test. If older equipment performs useful work for longer, extending its service period can describe its use more accurately. A processor displaced from the hardest workload may still serve another. Technical survival alone, however, does not establish economical operation: power, maintenance and replacement costs matter too.

The test must work in both directions. Better use of old equipment supports longer lives; earlier retirement or unexpectedly costly operation prompts scrutiny. A later disappointment does not automatically prove the original estimate was unreasonable, just as a favourable outcome does not validate every assumption made at the outset.

The 2023 disclosures quantify a historical reporting effect. The subsequent filings show larger investment and asset totals, a continuing range of equipment lives, and a distinct change for buildings. To establish better returns on capital, the missing evidence is what the equipment cost to keep running, what replacement spending it avoided and what useful work it delivered. A longer life on the page is the start of that inquiry.

Sources and calculation notes

  1. Microsoft FY2023 annual report, Note 1, income statement and cash flows. Disclosed effect divided by same-period reported earnings: 3,000/72,361 = 4.1%; 3,700/88,523 = 4.2%. Effects are rounded.
  2. Alphabet 2023 Form 10-K, Note 1, financial statements. 3,000/73,795 = 4.1%; policy effective in fiscal 2023.
  3. Microsoft FY2026 Form 10-K, property and equipment, policy and cash-flow disclosures. The $115.948bn is cash additions to property and equipment, not capex inclusive of finance leases; $34.3bn is total depreciation.
  4. Alphabet 2025 Form 10-K, Note 1 and cash flows. General six-year policy, prospective estimate changes, cash purchases.
  5. Alphabet June 2026 Form 10-Q, Note 1, property and equipment, and cash flows. Interim figures are unaudited; unfinished assets are not all server equipment.
  6. Microsoft FY2022, FY2024 and FY2025 annual reports, Note 1. Category labels and ranges compared across years.
  7. Microsoft FY2026 fourth-quarter earnings call, 29 July 2026, CFO remarks and Q&A. Management expectations for FY2027 are prospective; no FY2027 realised effect is asserted.