How we built the
Software Innovators Leaderboard
What we count, what we exclude, where the data comes from, and where its limits lie.
The Software Leaderboard catalogs enterprise software companies that have reached at least $500M in enterprise value, organized by founding year. Together, they represent roughly $28.2T in cumulative value across the technology eras that shaped the software industry.
The dataset powers three views at airframe.ai/leaders : a ranked company list, a year-by-year timeline, and an interactive map.
This page explains how we build the dataset. We update it as coverage changes and readers submit corrections.
All three criteria apply.
A company must meet three criteria. Its primary business must be selling software to businesses, governments, or institutions. Consumer products, hardware, services, and media do not qualify; mixed businesses such as Microsoft and Adobe qualify when enterprise revenue dominates their economics. The company must have reached $500M in enterprise value, supported by a validated public market capitalization, a priced private financing round, or a disclosed acquisition price. We use the most recent validated value when several are available. Finally, its founding year must be verifiable: the year it incorporated under its current core identity. We date spinouts to the spinout, not the parent.
Companies remain in the dataset after falling below $500M or ceasing operations, with their status marked accordingly. The record is cumulative.
Exclusions follow the inclusion criteria.
We exclude pure hardware businesses with supporting software, consumer internet companies such as Meta and Netflix whose software supports another business model, and service-led consulting firms. Companies also need a documented valuation above $500M. Per-company notes explain edge cases.
SEC filings outrank company disclosure, which outranks third-party databases, which outranks press accounts.
For public companies, we use SEC filings including 10-Ks, S-1s, and proxy statements. Private financing sources include Axios Pro Rata, Fortune Termsheets, Harmonic, PitchBook, and primary reporting. Press releases and acquisition announcements establish disclosed sale prices. Headcount comes from LinkedIn, careers pages, and annual reports; the Wayback Machine supplies historical context when current sources are insufficient. For conflicting values, we choose the most recent number from the most authoritative source: SEC filings, then company disclosures, third-party databases, and press accounts.
A way to compare the economics of software companies.
We divide a company's most recent enterprise value by its most recent reported headcount. Value per employee helps compare the economics of different software businesses.
For most of the past two decades, typical enterprise software companies built $1M to $10M of value per employee. SaaS businesses commonly combined a large sales and marketing organization with a moderate engineering team, growing roughly with headcount. Google's advertising model reached roughly $30M per employee. Anthropic operates at roughly $193M per employee, while AI-native companies founded in the past five years range from $50M to more than $200M per employee, around an order of magnitude above the SaaS baseline.
In our reading, this reflects a change in how model intelligence generates value relative to businesses that scale through sales and marketing headcount. Competitive pressure over the next decade may make some existing leaderboard companies' value-per-employee ratios difficult to sustain.
PE-Owned is a distinct category.
We classify companies as Public, Private, PE-Owned, or Acquired. PE-Owned is separate because leverage covenants, cash distributions, and constraints on R&D can materially change a company's operations and competitive position compared with venture-backed or founder-led businesses.
Era boundaries are approximate.
Six historical enterprise software eras produced distinct company cohorts: Mainframe (1950s), Minicomputer (1970s), Desktop (1980s), Client-server (1990s), Internet/Web (1995), and Cloud (2005). The seventh, AI, began around 2020 and is the subject of this series.
We assign companies to the technology era central to their initial product-market fit. Boundaries overlap, and companies such as Microsoft, Oracle, and Salesforce have crossed several eras. The taxonomy helps readers identify patterns.
Private valuations, headcount, and coverage have limits.
A private company's last financing price reflects one set of investors at one point in time. After 18 months without a new round, that value may be stale. Reported headcount often lags by 30 to 90 days, especially at private companies. Coverage is also incomplete: we have prioritized companies above $1B, leaving more gaps between $500M and $1B, particularly among international and PE-owned firms with limited disclosure.
We update the dataset monthly and accept corrections at research@airframe.ai. We include material corrections in the next publication and record them in the version history.
A public record across the industry's history.
Software has created substantial wealth over sixty years, but public information about leading companies is fragmented. Analyst coverage is paid. Crunchbase and PitchBook focus on financing events, while Wikipedia lacks a consistent dataset structure.
Airframe needs this dataset for its work. We also want a public record of what was built, by whom, and when. The data is free to use with attribution.
Paul