Explorer

Methodology

How every number on this site is collected, labelled and compared.

Sources: Public companies: revenue facts machine-read from each company's own SEC filings (XBRL companyfacts API for consolidated revenue; each 10-Q/10-K's XBRL instance for segments and product lines). Private companies and figures that are not in filings: company statements and reputable media, each with its URL, publication date and the source's own wording.
Data quality: Verified = read from the company's SEC filing by script. Reported = company or reputable media statement not machine-checked. Estimated = third-party or media estimate. Unavailable = the company is listed but nothing sourced exists; it is shown as missing, never filled.
As first reported: Each period's value is the one first filed. Later restatements (for example a business moved to discontinued operations) are flagged on the observation, not merged into the series.
Growth: Year-over-year growth is the company's own comparison: the current period versus the prior-year comparative printed in the same filing. Values from different filings are never paired, because a restatement can sit between them.
Fourth quarters: Most companies file only a fiscal-year figure for Q4. Q4 = fiscal-year revenue minus nine-month year-to-date revenue (or Q1+Q2+Q3 when no nine-month figure is tagged), same XBRL concept, marked “derived”. Derived quarters that would be zero or negative are rejected.
Tagging errors: A “quarter” larger than 60% of the fiscal year containing it is rejected as a tagging error (Oracle's 10-K tags full-year revenue on a three-month context). Pre-revenue zero values and predecessor-entity years (Nebius before 2024 was Yandex) are excluded.
Calendar alignment: Fiscal quarters are placed in the calendar quarter they mostly cover (period end minus about 45 days). Off-calendar fiscal years (Salesforce, Snowflake, NVIDIA, Oracle, Adobe) are marked. The comparison quarter is the latest one at least 80% of public companies have reported.
ARR vs revenue: ARR, run-rate and annualized figures are point-in-time and forward-looking; recognized revenue covers a period. They are kept as separate bases, labelled everywhere, plotted as separate series and never added, averaged or used in a denominator together.
Market totals: Matched panels: a market's growth compares the same companies in both periods, each with four filed quarters (trailing twelve months, TTM). Consolidated parents are replaced by their segments; multi-product companies count only in the market their revenue mostly belongs to; overlapping aggregates (Microsoft Cloud) are excluded.
Market share: Computed only for advertising, where a defensible denominator exists: TTM advertising revenue of a fixed panel of seven public platforms, shown only when all seven have data. It is share of that panel, not of the global ad market. No share is computed for cloud, software or AI, where vendors define revenue differently.
Causation: Correlated growth (one vendor speeding up while another slows) is never presented as displacement. Displacement is recorded only from sourced evidence: a customer statement, a company disclosure, or a survey, each labelled by evidence type. Everything else is a hypothesis.
Four mechanisms: Direct budget displacement, incremental AI spending, infrastructure expansion and venture-funded spending are kept apart. Filed revenue alone cannot separate them; each finding says what the data shows, what it suggests and what remains unknown.
Updates: The dataset is rebuilt from the original sources on every refresh and checked automatically before publishing: quarters must add up to the fiscal year, segments can never exceed their parent company, reported growth is reconciled against the stored figures, and every number must carry a source.