How we built the
Software Register.
What we count, what we exclude, where the data comes from, and where its limits lie.
What we count, what we exclude, where the data comes from, and where its limits lie.
The Software Register is Airframe's catalog of every enterprise software company that has reached at least $500M in enterprise value, organized by founding year. Together, these companies represent roughly $26T in cumulative value and define the modern software industry across the technological eras that produced them.
This page documents how the dataset is built. We update the methodology as the dataset evolves and as readers correct the record.
A company qualifies for the Software Register if it meets three criteria. First, it is an enterprise software company, meaning its primary business is selling software to other businesses, governments, or institutions, rather than consumers, hardware, services, or media. Companies that sell to both consumers and businesses (Microsoft, Adobe) qualify when their enterprise revenue dominates their economics. Second, it has reached at least $500M in enterprise value, drawn from the highest validated public market capitalization, the most recent priced private financing round, or a disclosed acquisition price. Where multiple data points exist, we use the most recent. Third, it has a verifiable founding year, defined as the year the company was incorporated under its current core identity. Spinouts are dated to the spinout, not the parent.
Companies that have crossed the $500M threshold and subsequently fallen below it remain in the dataset. Companies that have ceased operations also remain, marked accordingly. The dataset is cumulative, not point-in-time.
The Software Register deliberately excludes pure hardware businesses that wrap software around silicon, internet-native consumer companies (Meta, Netflix) where the software serves a non-software business model, and service-led firms where software is a delivery mechanism for human consulting. We also exclude companies that have not crossed the $500M threshold in any documented valuation. These exclusions are definitional, not editorial: a company either fits the criteria or it does not. Edge cases are documented in the per-company notes.
Our primary sources are SEC filings (10-Ks, S-1s, proxy statements) for public companies; Axios Pro Rata, Fortune Termsheets, Harmonic, PitchBook, and primary reporting for private financing rounds; press releases and acquisition announcements for documented sale prices; company-disclosed headcount on LinkedIn, careers pages, and annual reports; and the Wayback Machine for historical context where current sources are insufficient. When sources conflict, we default to the most recent number from the most authoritative source, with SEC filings outranking company disclosure, which outranks third-party databases, which outranks press accounts.
Value per employee is calculated as the company's most recent enterprise value divided by its most recent reported headcount. We use this ratio because it is the single most revealing metric for understanding what kind of business a software company actually is.
For most of the past two decades, the typical enterprise software company built between $1M and $10M of value per employee. That was the steady-state economics of the SaaS era, a sales-and-marketing-heavy organization with a moderate engineering team, scaling roughly linearly with headcount. Two recent inflections have broken the pattern. Google raised the ceiling, demonstrating that an algorithm-defined business with a high-leverage advertising model could compound value far faster than headcount, peaking at roughly $30M per employee. Anthropic redefined it. AI-native companies founded in the past five years are operating at $50M to $100M-plus per employee, an order of magnitude above the SaaS-era baseline.
This is not a quirk of valuation froth. It reflects a fundamental shift in how software businesses are built when the core economic engine is model intelligence rather than human-scaled go-to-market motion. The implication for the existing Software Register is direct: a substantial fraction of the list is operating at value-per-employee ratios that the next decade of competitive pressure will not sustain.
Each company is classified as Public, Private, PE-Owned, or Acquired. We treat PE-Owned as distinct from Private because the operating model, growth orientation, and strategic posture of a PE-controlled software business differs materially from a venture-backed or founder-led private company. PE-owned firms typically operate under leverage covenants, optimize for cash extraction over reinvestment, and constrain R&D spend in ways that meaningfully change their competitive trajectory. The classification matters for what the dataset reveals.
We organize the dataset around eight historical technological eras inside enterprise, each of which produced a distinct cohort of dominant software companies: Mainframe (1950s), Minicomputer (1970s), Desktop (1980s), Client-server (1990s), Internet/Web (1995), Cloud (2005), SaaS (2005), and Mobile Apps (2010s). The ninth, the AI era, began around 2020 and is the cohort the rest of this research walks.
Where founding cohorts span multiple eras, we assign each company to the era whose technology was most central to its initial product-market fit. This taxonomy is heuristic. Eras blur at their edges, and the most durable companies, Microsoft, Oracle, Salesforce, successfully traversed multiple transformations. The framework exists to make pattern recognition easier, not to enforce strict assignment.
We are transparent about three limitations. Private valuations are imperfect: last-round valuations reflect the price a single set of investors agreed to at a single point in time, and where a company has not raised in 18 months or more, the recorded value may be materially stale. Headcount data lags: self-reported headcount tends to update on a 30-to-90-day delay, particularly for private companies. And coverage, while comprehensive, is not exhaustive. We have prioritized companies above $1B and have less complete coverage in the $500M to $1B band, especially for international and PE-owned firms with limited disclosure.
We update the dataset monthly and accept corrections at hello@airframe.ai. Material corrections are reflected in the next publication and noted in the version history.
The software industry is the largest wealth-creation engine of the past sixty years, but no canonical, longitudinal dataset of its leading companies exists in the public domain. Analyst firms charge for fragmented coverage. Crunchbase and PitchBook capture financing events but not industry shape. Wikipedia is encyclopedic but unstructured.
We built the Software Register because Airframe needs this dataset to do its work, and because we believe the industry deserves a public record of what was built, by whom, and when. The data is free to use with attribution.
— Paul