AI Economy Radar

AI ADOPTION, READ ACROSS THE SIGNALS

One metric can miss the bigger story.

See how demand, adoption, spending and funding fit together—and where the signals disagree.

authored by US Megafund Investment manager

See real examples How the method works

A SYSTEMATIC VIEW OF THE AI ECONOMY

Seven lenses. Different questions.

A usage measure cannot tell you whether capacity is arriving. A lease commitment cannot tell you whether compute is being used. Each view adds context; none is a proven causal chain.

Penetration

Paid use, adoption and monetization.

Supply–Demand

Demand, available capacity, prices and constraints.

Semi & ODM

Chip and manufacturing shipments and revenue.

Peripherals

Networking, optics, power and cooling around compute.

Data Center

Build, lease and energize physical capacity.

Upstream Supply

Equipment and materials needed for future chip supply.

Funding

Cash flow, capex, debt, leases and capital availability.

The same company can appear in more than one view when its disclosures answer different questions.

DATA CENTER COVERAGE Google Cloud One tracked cloud platform within the wider capacity view.

DATED COMPANY DISCLOSURES

What changes when you read across?

Three short examples show how a second signal changes the question—not what conclusion to jump to.

FY2025 Q3
31 Mar 2025
+33%Azure and other cloud growth, reported
16 ptsof growth attributed to AI services

On the same call, Microsoft’s CFO said most Azure upside versus expectations came from non-AI services.

Why read both? Year-over-year growth contribution and forecast surprise use different bases. They are not a direct contradiction.

The call does not specify the 16-point currency basis. Do not divide it by 33% or treat it as AI’s revenue share. One historical quarter; no inference about current usage or margins.

Microsoft FY2025 Q3 earnings-call transcript

Meta

Q2 2026
30 Jun + July events
≈$278.99bnuncommenced lease obligations at 30 June
≈$68bnseparate July data-center lease commitments

The July commitments were expected to commence in 2027–28; they are a later event, not a revision to the 30 June balance.

Why read both? A lease pipeline and a later commitment say something about planned capacity—not delivered compute or usage.

These disclosures are not a reconciled total. Neither gives facility-level megawatts, utilization, or AI allocation.

Meta Form 10-Q · Note 9

Why definitions matter

H1 2026 · six months ended 30 Jun
$98.411bnCash purchases of property and equipment
vs
$71.419bnOperating cash flow

Gross cash purchases show investment intensity; they do not show AI usage or delivered compute.

Amazon Form 10-Q · cash-flow statement

Sources are dated company disclosures and filings. Periods and definitions differ; the figures are not peer rankings. These historical examples do not prove causation or predict returns.

Amazon mark: © 1996-2026 Amazon.com, Inc. or its affiliates. Official US Press Center. No press accreditation or separate written permission is claimed.

THE METHOD, IN THREE STEPS

From observations to useful perspectives.

A repeatable structure keeps unlike signals in their proper units and makes coverage limits visible.

01

Source observations

Company filings, market series, usage evidence and supply data.

02

Comparable indicators

Each indicator keeps a fixed scope, native unit, period and documented inputs.

03

Seven perspectives

Signals are grouped by what they measure—not forced into one adoption score.

Under the hood: raw or atomic series are mapped at L4 for AI relevance, unit and signal-quality weight, then combined into L3 indicators and L2 subgroups before the seven views are presented. Estimated AI shares, missing observations and short histories remain explicit limitations.

Calculation and interpretation notesUnits, weights, coverage and missing data

AI share is an estimate

When a company or series spans several activities, the method estimates what portion is associated with AI. That assumption adds uncertainty and is recorded explicitly.

Weights describe signal quality

Weights reflect evidence coverage and reliability. They are not company market share or a measure of commercial importance.

Units and scope stay fixed

Inputs are normalized to comparable currency and units within a documented indicator. Different periods or definitions are not silently combined.

Missing data is not zero

A result is shown only when at least 80% of assigned weight is valid. An eligible late observation may carry forward for at most one period, with its age visible.

Seven groups organize signals

Penetration; Supply–Demand; Semi & ODM; Peripherals; Data Center; Upstream Supply; and Funding describe distinct questions, not a proven causal chain.

No single indicator proves adoption

Several observations can reveal disagreement, timing and coverage. The framework does not forecast outcomes or establish that one part of the AI economy caused another.

FULL TRACKER

Explore the interactive tracker.

A historical preview is available for 6 September 2026. Its charts, series and chart interactions are preserved; non-chart analysis and detail text is redacted. Later full editions remain protected.

HISTORICAL EDITION

Charts in context.

Choose 6 September 2026 to open the interactive preview. Later dates stay locked unless the server confirms release and member access.

Checking preview
Historical interactive preview · 6 September 2026Chart geometry, series and tooltips are shown. Analysis prose, source notes and detail tables are redacted.

Checking preview availability…

Preview date: 06 Sep 2026Other full editions remain protected

NOTES & EXPLAINERS

Ideas behind the evidence.

Writing on AI adoption, measurement and the infrastructure underneath it.

ARTICLE LIBRARY

No articles published yet.

The Substack/RSS feed is not connected. New writing will appear here after a source is configured and verified.

MEMBERSHIP

More detail, when it is ready.

Planned membership includes dated reports and the complete report archive.

Checkout is unavailable. Reports remain on hold while publication rights, archive access, email and billing are configured.

MONTHLYNOT CONNECTED

$9.90/ month

USD, billed monthly.

ANNUAL16.7% LOWER

$99/ year

USD · $8.25 monthly equivalent.

NOT CONNECTED