How much venture capital funding actually goes to AI companies?

Key takeaways

• By count, AI-core startups (core product is AI) made 22.2% of 2025 Form D offerings in this sample (370 of 1,668).
• By dollars, they raised 37.3% of the 2025 Form D amount (likely range 23.4% to 52.8%), or 25.3% (17.8% to 34.4%) with Databricks counted like Snowflake, as a data platform with AI features.
• The count share was 16.6% in 2023; about half of the gap to 2025 comes from more offerings by startups founded since 2022, so the direction is more reliable than the size.
• Among startups incorporated in 2022 or later, AI-core startups made about a third of offerings in every year (30.9% in 2023, 34.2% in 2025), with no clear change.
• The provisional first half of 2026 (22.0%, 106 of 481 offerings) is not clearly different from the first half of 2025 (22.8%).
• Pooled over 2023-2025, the median equity offering was $7.35M for AI-core startups and $5.84M for others, a gap found only in second and later offerings.

This article is for founders and investors who want AI's real share of fundraising by VC-listed startups from 2023 to mid-2026, by count and by dollars, and where the numbers stop being reliable.

How many funding rounds by VC-listed startups go to AI startups?

By count, startups whose core product is AI made 22.2% of 2025 Form D offerings in this sample (370 of 1,668; likely range 19.8% to 24.5%). That is a share of offerings, not of dollars; the dollar share is in the next section.

Form D is the short notice a company files with the SEC, the securities regulator in the United States, when it sells securities privately; it shows how much was sold and when, but names no investors. Each offering counts as one round. The startups are VC-listed, which does not show that a VC invested in a given round.

What counts as an AI startup in this data?

The headline counts AI-core startups: AI is the product, or the core mechanism that does its main job. Remove the AI and the product stops working; MatX, which designs chips built for large language models, is an example.

AI-as-a-feature companies add AI to a product that worked before it, such as Clear Skye's identity-governance software with an AI add-on. Counting them as well, the 2025 share is 32.7% of offerings (likely range 30.1% to 35.3%).

Did the AI share of offerings rise from 2023 to 2025?

By today's labels, the AI-core share of offerings was lower in 2023 (16.6%, likely range 14.8% to 18.5%) than in 2025 (22.2%), a difference of 5.6 points. About half of that difference comes from later years' offerings coming more from startups founded since 2022. Because labels come from companies' current websites, the direction is more reliable than the size of the change.

Among startups incorporated in 2022 or later, about a third of offerings went to AI-core startups in every year: 30.9% in 2023 (likely range 26.4% to 35.8%) and 34.2% in 2025 (30.6% to 38.0%), with no clear change.

Is the AI share of startup rounds still rising in 2026?

Not clearly. In the provisional first half of 2026, AI-core startups made 22.0% of offerings (106 of 481; likely range 18.1% to 25.9%), not clearly different from 22.8% in the first half of 2025.

Two things keep 2026 provisional. Only about 70% of second-quarter offerings had probably been filed by the June 30 cutoff, and startups raising for the first time in 2026 may not yet appear on VC portfolio pages, which were read in September 2026.

What share of first funding rounds goes to AI startups?

AI-core startups are more common at the first raise: they made 30.2% of first equity Form D offerings in 2025 (likely range 26.2% to 34.4%), about 10 points more than their share of later offerings.

What share of venture dollars goes to AI, and why is it hard to pin down?

By dollars, AI-core startups raised 37.3% of the 2025 Form D amount in this sample (likely range 23.4% to 52.8%). Counting Databricks as a data platform with AI features, as Snowflake is counted, the share is 25.3% (17.8% to 34.4%).

Databricks supplies $8.59B of 2024 dollars and $5.73B of 2025 dollars, so its label alone moves the 2024 share from 37.8% to 19.4%.

As labelled, the AI-core share of Form D dollars rose from 21.7% in 2023 to 37.3% in 2025 (+15.6 points, likely range +1.5 to +26.5). Without each year's five largest offerings it rose by 8.3 points (+1.2 to +15.4). Counting Databricks as a data platform with AI features, the change (+9.5 points, -1.2 to +20.4) is not clear, so the dollar rise depends on how that one company is labelled.

The first half of 2026 shows 49.3%, but that rests on five offerings, all AI-core, in an incomplete half-year.

Offerings above $2B were kept only when they were venture equity rounds (Stripe in 2023, Databricks in 2024 and 2025); a debt filing by a listed company's lending arm and one anomalous amount were dropped. Offerings by issuers that were already public at the time, 252 by 123 issuers, were excluded too, because they are not startup venture rounds.

OpenAI, Anthropic and xAI have no Form D offerings in the sample, so this dollar share describes the sample, not AI's share of all venture dollars.

Do AI startups raise bigger rounds than other startups?

Only after the first round. Pooled over 2023-2025, the median equity Form D offering by an AI-core startup was $7.35M, against $5.84M for other startups (ratio 1.26, likely range 1.11 to 1.45); at the first offering there is no clear difference.

The median is the middle value: half of offerings are bigger and half smaller, so a few huge raises cannot distort it. At the first equity offering the medians are $5.70M and $5.06M, a ratio of 1.13 whose range (0.97 to 1.29) includes 1.

To be fair, the gap in later rounds is real: $8.28M against $5.21M at the second offering and $13.70M against $7.17M at the third and later, and both hold after correcting for the number of comparisons. Year by year the evidence is borderline, with only 2024 passing both tests, narrowly.

Which sectors have the highest share of AI funding rounds?

Software. In 2023-2025, AI-core startups made 46.3% of software & infrastructure offerings in this sample of VC-listed startups, against 10.1% of health & life sciences offerings.

Part of the software figure is circular, because companies that build AI models and AI infrastructure sit inside that group and are AI by definition. Without them, software & infrastructure is 36.8% AI-core.

No sector's change from 2023 to 2025 survives correction for testing 11 groups at once.

What share of the startups VC firms list are AI startups?

Of the 15,418 links between VC firms and the startups in this sample, 23.5% point to AI-core startups (likely range 21.5% to 25.6%). A link is a portfolio-page listing, undated and including exits, so it shows that a firm lists a startup, not that it invested in a given offering.

Firm counts depend on how same-name records are matched: between 3,573 and 3,680 VC firms list at least one startup in the sample, and between 1,712 and 1,757 list at least one AI-core startup. The upper counts treat a same-name record with no other website as the same startup.

How was this measured?

The analysis counts SEC Form D offerings with a first sale from January 2023 to June 2026 by VC-listed startups, and labels each startup from its own website. The source is SEC filings (Form D notices) for VC-listed startups, compiled by VC Catalog.

How were companies classified as AI?

Each of the 4,022 startups with an offering in the window was labelled with one written rule, mostly from its own website text as of 2025-26. AI language models applied the rule, and for 67 startups whose text was thin the label used general knowledge instead.

Two independent relabellings, of 300 and 200 random startups, agreed with the labels on AI-core or not for 96% of startups. Most disagreements sit on the line between AI-core and AI-as-a-feature, and depending on where that line is drawn, the 2023-2026 H1 share ranges from 17.2% to 19.8% of offerings. For that range, each offering counts with the chance that the second labeller calls its startup AI-core, given the startup's own label.

How do labels from today's websites affect 2023?

The labels describe what each company says it is now, and they also apply to its 2023 offerings. 60% of 2023 AI-core offerings came from startups incorporated before 2022, whose current website may describe a later pivot to AI, so the 2023 level and the size of the change are uncertain beyond the ranges shown.

A check of archived mid-2023 homepages found AI wording for 16 of 26 sampled AI-core startups. It reads homepages only and is small, so it gives no corrected 2023 level.

Ranges are 95% confidence intervals. For the overall AI shares of offerings and dollars and their changes, they come from re-drawing whole startups, not single offerings, many times, because many startups raise more than once; other differences use tests of whether a difference could be chance.

What are the limits of this data?

The figures describe Form D offerings by startups that VC firms list today, labelled from today's websites, so they are shares of a sample and not totals for the venture market.

Bottom line

By count, about one Form D offering in five by VC-listed startups goes to an AI-core company; by dollars the share is higher but hinges on how a few large companies are labelled.

• By count, AI-core startups made 22.2% of 2025 offerings; by dollars, 37.3%, or 25.3% with Databricks counted like Snowflake.
• The 2023 share was lower, but about half of the gap is composition; read the direction, not the size.
• Among startups incorporated since 2022, about a third of offerings went to AI-core startups each year.
• The provisional first half of 2026 neither shows a slowdown nor rules one out.
• AI-core offerings are larger only from the second offering on.

Where to go next