Open dataset · 2026

AI capex, stripped of false precision.

Download the evidence behind the $720B–$745B hyperscaler spending envelope. Every row preserves the company's definition, period, source, and comparison limits.
Companies
4
MSFT · AMZN · GOOGL · META
Guidance low
$720B
Guidance high
$745B
Reviewed
2026-09-25

Download the complete comparison

CSV is ready for spreadsheets and models. JSON preserves nested watch items, limitations, and source metadata. Both formats are free to download and subject to our terms of use.

What the total means

This is a capital-spending envelope, not an audited AI-only total. Amazon is companywide, Meta includes finance-lease principal, Microsoft uses calendar guidance across a June fiscal year, and none of the four isolates every AI dollar.

See how to normalize the numbers
Company2026 guidanceLatest signalPrimary comparison limitSources
Microsoft

MSFT

About $175B

Calendar 2026

$41B

Roughly two-thirds short-lived assets, primarily CPUs and GPUs; finance leases were $5.6B

The company does not disclose a standalone audited AI-capex figure.
Amazon

AMZN

About $220B

Calendar 2026

~$200B to ~$220B

Raised July 30, mainly due to memory costs; capacity still short of 2026 demand

The $220 billion plan is companywide and cannot be treated as pure AI or AWS capex.
Alphabet

GOOGL

$195B–$205B

Calendar 2026

$44.9B

About 60% servers and 40% data centers and networking

Alphabet does not publish a clean AI-only capital-expenditure measure.
Meta

META

$130B–$145B

Calendar 2026

$31.08B

Includes principal payments on finance leases

The capex outlook includes infrastructure used across Meta's core business and research, not only generative AI.

Method and update policy

Yield Theory records the latest explicit annual guidance from each company, preserves its accounting scope, and links the primary earnings or investor-relations source. We update the dataset after a material guidance change rather than manufacturing a daily estimate from market commentary.

Read the research methodology

What this dataset cannot tell you

  • It does not isolate a comparable audited AI-only capex line.
  • It does not predict a supplier's revenue from the spending total.
  • It does not prove the new capacity will reach profitable utilization.
  • It does not replace current company filings or personalized advice.

The capex number is the start of the thesis.

hyperscaler capex guidance stays in the issuer’s own definition. Members get the investment case: catalysts, risks, and the evidence that would change the view. $15/month or $150/year.

$15/mo or $150/yr · cancel future renewals anytime · sources included