Skip to content
The Narraitive

Big Tech Capex Is Now a Macro Force, Not a Line Item

Four companies' data-center spending now rivals national infrastructure programs. The depreciation wave it creates is the most predictable earnings story of 2027.

Published Apr 3, 2026Updated Jun 3, 2026Data refreshed Jun 11, 20262 min read
capexhyperscalersdata centersearnings
Share
◆ AI Pulse · Proupdated Jun 11, 2026Contested

The AI Pulse is a Pro feature

Machine-synthesized latest developments, market read, and watch list — plus an embeddable widget for your own site.

Upgrade to Pro

AI-readable summary

Combined capital expenditure by the four largest cloud/AI companies is running at an estimated $480B annualized in 2026 — roughly 1.6% of US GDP and larger than most national infrastructure programs. This spending now measurably moves aggregate investment statistics, power markets, and construction employment. Because data-center assets depreciate over 5–6 years, the 2024–2026 spending surge produces a predictable depreciation wave that compresses reported margins beginning in 2027, independent of AI revenue outcomes.

TL;DR

Four companies' capex (~$480B/yr) is now macro-scale: it moves GDP investment lines, power markets, and construction labor. The mechanical consequence is a 2027 depreciation wave — the most predictable margin headwind in large-cap tech.

Key facts

  • Top-4 combined capex: ~$480B annualized in 2026 (The Narraitive estimate), ~1.6% of US GDP.
  • Data-center construction now employs more workers than several traditional heavy industries.
  • Server/accelerator assets depreciate over roughly 5–6 years.
  • Depreciation from the 2024–2026 surge begins compressing margins in 2027 regardless of AI revenue.

Key metrics

Top-4 capex

$480B

annualized 2026

Share of US GDP

1.6%

+0.5pp YoY

Asset life

5–6 yrs

being extended

Margin impact

2027

depreciation wave

Main thesis

The market debates whether AI revenue justifies the capex. The more useful observation is mechanical: depreciation is revenue-independent. The 2024–2026 build creates a margin headwind that arrives on a schedule, and companies are already responding the predictable way — extending useful-life assumptions. Watch the accounting footnotes, not the keynotes.

The scale, in macro terms

Combined capex for the four largest cloud/AI companies reached an estimated $480B annualized in 2026. For context, that is roughly 1.6% of US GDP, concentrated in data centers, accelerators, and power infrastructure.

At this scale, the spending stops being a corporate-finance story. It shows up in national investment statistics, regional power prices, transformer lead times, and construction wages. Economists are now adjusting GDP nowcasts for hyperscaler guidance revisions — a sentence that would have been absurd in 2022.

Top-4 combined capex, annualized$B
Combined capexSource: Company filings, The Narraitive annualization (illustrative preview data)

The depreciation wave is arithmetic, not opinion

Accelerators and servers depreciate over five to six years. Spending that tripled between 2023 and 2026 therefore produces a depreciation expense that roughly triples on a lag — beginning to bite reported operating margins around 2027.

This is the rare earnings headwind that requires no forecast: the assets are already purchased and the schedules are already set. The open variable is whether AI revenue grows into the cost base before the wave crests.

Modeled incremental depreciation from the 2024–2026 build
YearIncremental D&A ($B)Margin impact (pp, top-4 avg)
202522−0.9
202648−1.8
202776−2.6
202895−3.1

Source: The Narraitive model assuming 5.5-year average life (illustrative preview data)

Watch the useful-life footnotes

Our opinion: the most informative disclosures this year are not AI revenue claims but useful-life extensions. Several companies have lengthened server depreciation schedules since 2023, each extension deferring expense into later years. Sometimes justified by genuine hardware longevity — and always convenient.

When capex growth decelerates while useful lives extend simultaneously, margins are being managed through the denominator. That pattern deserves more scrutiny than it gets.

Related markets via TradingView

Methodology

Capex is annualized from the latest two reported quarters. The depreciation model assumes a 5.5-year average life and straight-line schedules, with sensitivity bands in the artifact data. Preview note: this starter article ships with illustrative mock data generated by The Narraitive's refresh pipeline; live data connections replace it at launch.

Data sources

  • Quarterly filings of the four largest cloud/AI companies
  • BEA investment statistics
  • The Narraitive depreciation model

Data freshness

Published Apr 3, 2026. Narrative last updated Jun 3, 2026. Underlying data last refreshed Jun 11, 2026 by the automated pipeline; charts and tables on this page render from those artifacts. If a refresh fails, the previous good data remains live.

What changed since last refresh

  • Jun 3: 2026 capex estimate raised to $480B from $455B after Q1 reports.
  • May 1: Added margin-impact column to the depreciation table.

Risks and limitations

  • Useful-life extensions could defer the modeled margin impact by 1–2 years.
  • A capex pause would change the 2028 figures materially; 2026–2027 are largely locked in.

Frequently asked questions

How much are big tech companies spending on data centers in 2026?
The Narraitive estimates the four largest cloud/AI companies are spending about $480B annualized on capital expenditure in 2026, roughly 1.6% of US GDP.
What is the depreciation wave?
Data-center assets bought in the 2024–2026 capex surge depreciate over 5–6 years, mechanically raising expense and compressing reported margins from about 2027 onward, independent of AI revenue.

Related briefings