AI Startup Funding Rounds 2021–2026

Key findings

1. 5,412 companies raised 8,122 rounds between 2021 and mid-2026. 6,974 rounds disclosed an amount, totalling $301.5B.
2. 2024 was the busiest year, with 1,768 rounds. Disclosed funding in 2025 was $83.2B, 1.8× the 2021 level ($47.3B).
3. Mega-rounds dominate the dollars. Rounds of US$100M or more are 11% of rounds with a disclosed amount, but 67% of disclosed dollars.
4. Early rounds got bigger. The median Seed round went from $3.0M in 2021 to $4.8M in 2025; the median Series A from $13.0M to $16.5M.
5. Funding is concentrated. The top three countries (United States, United Kingdom, Canada) hold 79% of disclosed dollars.
6. The fastest-rising sectors (share of rounds, 2021–2022 vs 2024–2025): Business Services (3.8% → 6.0%), Enterprise Infrastructure (6.3% → 8.5%), Consumer (2.6% → 3.9%).

Funding by year

Full data. 2026 covers January to June.

Round size by stage

Most active investors (top 50)

Rounds each investor joined, full data. Organisations only; aliases such as "Accel" and "Accel Partners" are merged.

Largest disclosed rounds (top 50)

Most funded companies (top 50)

Sum of disclosed round amounts, full data. Amounts that failed the source-sentence check are excluded, so totals for the very largest companies are understated.

Top 30 countries by disclosed funding

Sectors (top 40 by rounds)

Column dictionary

19 columns in the sample CSV.

Method

1. Extraction. Radar's pipeline reads each funding article and extracts structured fields from it, including the source sentence.
2. Back-fill days removed. Two days on which archived articles were loaded with the load date as their publication date (2024-04-22 and 2025-02-13) are excluded, because those rounds carry no reliable date.
3. Scope. Venture-style instruments only (pre-seed to Series H, growth/PE, bridge, convertible notes, venture debt, grants, strategic). IPOs, QIPs, private placements, rights issues, bank loans, bonds and acquisitions are dropped. Companies founded in 2000 or later; rounds after a company's IPO and group subsidiaries of large conglomerates are excluded.
4. Deduplication. Reports about the same company and stage within 45 days are one round. A round repeating an earlier amount within 180 days is dropped, as are lettered stages below one already announced (recaps). One round filed under two domains of the same company is merged.
5. Amount check. Each amount is compared with the money figures parsed from its source sentence (tolerance 0.6–1.7×). verified matched; unchecked had no parseable figure; failed_check is blank because it did not match, exceeded a stage ceiling, or was a $500M+ figure reported by one article only; not_disclosed had no amount.
6. Currency. US dollars, converted at the time of extraction and rounded to three significant figures.

7. News columns. The excerpt is the earliest reporting sentence that does not name a data provider. Source and URL come from that article when it can be retrieved.

8. Sample. Each company key is ranked by sha1(key). Keys are taken in rank order until the file holds max(25%, 5,000) rows (capped at 100,000).

9. Profile links. Links are built from Fliar's company index and checked against it. A row whose company has no public profile has an empty link.

Limitations

• Derived from news: rounds that were never reported are missing, and coverage is stronger for larger and English-reported deals.
• Amounts are as reported; fixed-rate conversion to US dollars makes local-currency rounds drift slightly over time.
• Stage labels mean different things in different markets.
• 2026 covers January to June only.
• Sector labels come from a company classifier, so some companies are mislabelled.
• news_source and news_url are blank when the article could not be retrieved at release time (about 40% of rows).

Cite this dataset

@misc{fliarbi_ai_startup_funding_2026,
  author    = {{Fliar BI}},
  title     = {{AI Startup Funding Rounds 2021–2026}},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.23431429},
  url       = {https://doi.org/10.5281/zenodo.23431429},
  note      = {Sample release. Licence CC BY-NC 4.0}
}