capital-markets
AI Companies' Share of U.S. Private Venture Money (SEC Form D, trailing four quarters)
Latest
44.04%
Methodology
## What this measures
The percentage of venture-like money raised over the trailing four quarters that was raised by companies Palanor classes as AI companies.
## How it is made
Every company that sells securities under Rule 506 of Regulation D files a notice, Form D, with the SEC within 15 days of the first sale, and amends it at least once a year while the offering runs. The SEC's Division of Economic and Risk Analysis publishes every notice as filed in quarterly data sets (2009 onward). Palanor loads all of them.
**Offerings and amendments.** An offering is its SEC file number: an amendment (Form D/A) keeps the number of the notice it amends. Each filing reports the total sold so far, so the money a filing adds is its total minus the previous filing's. Money raised in a quarter is the sum of what the filings dated in that quarter add. An amendment that raises the total adds the new money in the quarter it is reported; an offering's latest filing decides whether it counts.
**Venture-like offerings.** Equity (alone or with debt, options or other securities) sold under Rule 506 by a U.S. private operating company: no pooled-fund interests, no funds or deal vehicles (by industry group, entity type or name, such as "Fund", "SPV" or "a Series of"), no banks, investment vehicles, oil and gas, coal or real estate, no limited partnerships or business trusts, no registered investment companies, no business combinations, and no company that was already an SEC reporting company (a 10-K, 10-Q, 20-F, 40-F or 8-K in the 15 months before), so public-company private placements are out. Every filing that adds $1 billion or more was checked by hand; typing errors (one notice reported $105 billion for a round of about $105 million) and vehicles the rules missed are dropped. The full rule list is in the Palanor code (src/lib/form-d/venture.ts).
**AI companies.** Palanor's own classification of each issuer, not the SEC's and not a data vendor's. Every issuer that raised $30 million or more from 2015 on, and every issuer whose name mentions AI (or hints at it) that raised $1 million or more, was checked by a search model that read the company's website or reliable reporting and answered whether AI or machine learning was the core of what the company sold in the years it raised the money; its evidence page must be among the pages the search actually retrieved, and the page is stored. Smaller issuers are classed from their names alone. An AI company here builds AI models or AI-native software, AI developer tools or data infrastructure for AI, chips or compute built for AI, autonomous vehicles or robots driven by machine learning, or AI-first products where the model is the product; a business where AI is a feature or a marketing word is not one. A hand check of 60 issuers drawn at random (30 classed AI, 30 not) found 22 of the 30 AI calls clearly right and 5 more defensible, and 3 AI companies among the 30 not-AI calls, all smaller raisers classed from their names.
## How to read it
The share of private venture money going to AI companies. Four quarters smooth single mega-rounds: one AI round of $10 billion or more still moves it by several points. By calendar year it rose from about 4% in 2015 to about 47% in the first three quarters of 2026. Palanor checked the yearly shares for 2015–2026 against an industry tally built from deal reports: the levels differ, but the years rank in almost exactly the same order.
## What it does not cover
Form D is a notice of an exemption, filed by companies that choose Rule 506. Some of the largest AI developers raise without filing one (in these data sets OpenAI and Anthropic appear only through investor vehicles, which are left out, never for their own rounds), and many companies file late or never, so the totals undercount private money and the AI share runs below industry tallies built from deal reports. The money counts what companies report as sold, not when it was wired. A company's class is set once for the whole period from what it sold while raising; companies that pivoted to AI late can be misclassed. Industry tallies such as PitchBook's count deals, use their own company classification and include rounds Form D never sees, so the levels differ; this series is built to follow the direction, not to match the level.
## Source
Palanor calculation from SEC Form D filings (U.S. Securities and Exchange Commission, Form D data sets; public domain) and EDGAR filing indexes. The AI classification is Palanor's.
Read this signal inside the lattice.
Palanor weighs every signal against the world your organization is watching. Methodology is public; integration is the platform.
See it in Palanor →