Follow the signals from paid software use to chips, data centers, and the money financing the buildout.
Adoption has more than one signal. Read the chain, not a single headline.
One method. Multiple points of view.
Signals are grouped by what they tell us. The views connect, but each describes a different part of the landscape.
Who is using AI, and whether usage is becoming paid, persistent behavior.
17 indicatorsDemand signals alongside the supply, price, and capacity available to meet them.
17 indicatorsChip and manufacturing output that supports AI systems.
15 indicatorsDevices and components that connect people and systems to AI.
11 indicatorsBuildout in compute, power, networking, and physical capacity.
16 indicatorsMaterials and equipment needed to expand the AI supply chain.
17 indicatorsCapital flowing to AI companies and infrastructure, with financing treated as a signal rather than proof of adoption.
These diagrams explain the framework. Each is labeled illustrative or framework; none is a reported time series.
Usage can rise before recurring spend. Both matter, and they answer different questions.
Reach and monetization should be read side by side.
Price and capacity can reveal bottlenecks before they show up as end-user adoption.
Infrastructure signals describe constraints, not customer adoption.
New indicators enter as history accumulates. Coverage and confidence remain visible.
A newer indicator can be useful while its history is still short.
Company and market evidence flows through a consistent chain so readers can see how an observation becomes a group-level view.
Signals can move together without proving one caused another.
Each example is drawn from a tracked company and a dated primary source. The lesson is about measurement, not investment advice.
16 ptsAI services contribution inside +33% reported Azure and other cloud growth
On the same quarter’s call, Microsoft’s CFO said most of Azure’s upside versus expectations came from non-AI services.
One historical quarter. The 16-point FX basis was not specified; it does not reveal AI’s revenue share or prove usage, margins, or causality.
Microsoft earnings-call transcript123.36%quarterly cash capex divided by operating cash flow
Oracle reported $28.499bn in cash capex against $23.103bn in operating cash flow; the simple difference was −$5.396bn.
Whole-company flows for one quarter. This is not company-defined free cash flow, an AI-only measure, or a liquidity conclusion.
Oracle Form 10-Q, cash-flow statement$278.99bnuncommenced operating and finance lease obligations at quarter-end
A separate July disclosure added about $68bn in data-center lease obligations, expected to commence in 2027–28.
The two dated disclosures are not one reconciled balance. Neither gives facility-level MW, utilization, or AI allocation.
Meta Form 10-Q, Note 9
137.79%cash purchases of property and equipment divided by operating cash flow
Amazon reported $98.411bn in cash purchases against $71.419bn in operating cash flow. Their simple difference was −$26.992bn.
Whole-company H1 flows, not AI-only spending. Amazon’s defined free cash flow nets proceeds and incentives; this gross-purchase calculation is not that measure.
Amazon Form 10-Q, cash-flow statementSources are first-party company or SEC filings. Periods differ; the figures are not directly comparable. Accessed 05 Oct 2026.
Amazon mark: © 1996-2026 Amazon.com, Inc. or its affiliates. Official US Press Center.
Atomic series are adjusted for AI relevance, rolled into indicators, then grouped into subgroups and the seven views. Combination rules preserve consistent units and show how much input coverage supports a result.
Explore the calculation notesWhere a company or series spans multiple activities, the method estimates the portion associated with AI. This is the largest source of uncertainty and is reviewed as an explicit assumption.
Weights reflect coverage and weighted signal-to-noise reliability. They are not company market share or a measure of commercial importance.
Series are brought to consistent currency and units before each valid input contributes value × estimated AI share to a sum. At least 80% of the assigned weight must be valid for a combined result to be shown.
A late series may carry forward for at most one period. Results are grouped across related signals; they do not establish causation or create a single score that proves adoption.
Method snapshot: 05 Oct 2026. Source outputs are dated separately; report distribution remains on hold pending publication clearance.
Choose a publication date. Full reports stay protected behind membership access and rights clearance.
The original report is served in its native format inside this protected viewer when the report and your access are cleared.
Membership detailsChecking report availability…
Short context on adoption, measurement, and the infrastructure underneath it.
The publication feed is not connected yet. When a real Substack or RSS feed is configured, new writing will appear here.
Membership is planned to include dated reports and the complete report archive, with notes behind each signal.
Membership checkout is currently unavailable. Reports remain on hold while access, rights, email, and billing are configured.
$9.90/ month
Full access, billed monthly.
$99/ year
$8.25 per month, billed yearly.
Secure-link sign-in is not connected yet. No sign-in link will be shown or claimed until email delivery is verified.