For every unicorn founded a decade ago, 57 seed startups are dead
Published 23 September 2026
We took every startup in our Crunchbase copy that was founded between 2015 and 2017 and has a seed-stage funding history: 30,826 companies, the 2015 to 2017 seed class. Nine to eleven years on, we sorted each one into where it ended up: a later round, an acquisition, a shutdown, or none of those. For the companies with none of those, we checked whether the website is still there.
170 of the 30,826 carry Crunchbase's unicorn tag. About 9,710 are dead, which is 57 for every unicorn. Source: FundingWatcher, September 2026.
Key findings
- 170 companies became unicorns and about 9,710 are dead: 57 dead companies for every unicorn. The range, from the sample margin, is 54 to 60.
- 9,848 of the 30,826 (31.9%) raised a Series A or later. Mattermark reported about 31 in 100 for US software seed rounds from 2009 to 2012.
- 1,691 (5.5%) were acquired at some stage.
- 20,978 (68.1%) never raised another round. Crunchbase records no shutdown and no acquisition for 18,047 of them.
- We checked a random 1,500 of the 17,904 of those with a website. 620 (41.3%) are gone, down, parked, a placeholder or someone else's site now; another 144 (9.6%) still load but their copyright line stops in 2024 or earlier.
- 139 of the 1,500 (9.3%) now show someone else's website. 30 of those are gambling sites.
- The share reaching Series A falls with each younger cohort: 35.7% for companies founded 2012 to 2014, 31.9% for 2015 to 2017, 23.8% for 2018 to 2019.
The boxes fed by "No later round, no exit on file", apart from "No website on file", are estimates from the 1,500-site sample, rounded to 10. Source: FundingWatcher, September 2026.
57 seed startups from the 2015 to 2017 class are dead for every unicorn
We count a company as dead when Crunchbase marks it closed without an acquisition (2,310 companies), or when it never raised again, has no exit on file and its website is gone, down, parked, a placeholder or someone else's site. The second group comes from the 1,500-site sample: 41.3% of the 17,904, about 7,400 companies, with a margin of 2.5 points either way. Together that is about 9,710, between 9,260 and 10,160. We leave out the 144 sites with a stale copyright year from this count, because they still load.
170 companies in the class carry the unicorn tag. 9,710 divided by 170 is 57, and the margin puts it between 54 and 60.
| Founded | Seed-stage startups | Unicorns | Dead, estimated | Dead per unicorn |
|---|---|---|---|---|
| 2015 | 10,351 | 73 | about 3,270 | 45 |
| 2016 | 10,009 | 55 | about 3,440 | 63 |
| 2017 | 10,466 | 42 | about 2,990 | 71 |
Each single year rests on 432 to 555 sampled sites, so the per-year figures are looser than the pooled 57. The ratio rises for younger years mostly because they have had less time to reach a unicorn valuation.
31.9% of 2015 to 2017 seed-stage startups raised a Series A or later
9,848 of the 30,826 companies raised a Series A or a later round. Source: FundingWatcher, September 2026. 4,539 stopped at Series A, 2,830 at Series B and 1,689 at Series C or later. 790 have a post-IPO round as their last round, 222 of them a post-IPO debt round; this group includes small listings outside the US. 170 carry Crunchbase's unicorn tag.
The one public benchmark we found is close. Mattermark tracked 2,011 US software companies that raised seed rounds between 2009 and 2012 and wrote that "around 31 of them" in 100 go on to a Series A (Mattermark). Their base is narrower than ours: US only, software only, grouped by seed year.
86.0% of the companies that never raised again have no exit on file
20,978 companies have a seed-stage round as their last round. 1,110 of them (5.3%) have an acquisition on file. 1,821 (8.7%) are marked closed and have no acquisition; 268 more are marked closed and acquired, and we count those as acquired. The remaining 18,047 (86.0%) have neither, nine to eleven years after they were founded. 12,370 of them carry the status "active" and 5,677 carry no status. 70.3% of the 18,047 raised their last round in 2020 or earlier.
50.9% of the companies with no exit on file are dead or abandoned
143 of the 18,047 have no usable website on file, which leaves 17,904. We drew a random 1,500 and fetched each homepage. Here is what we found:
| What the website shows | Companies | Share of 1,500 |
|---|---|---|
| Same company, current or no copyright year | 599 | 39.9% |
| Domain has a DNS record but no working site | 223 | 14.9% |
| Domain no longer registered | 149 | 9.9% |
| Same company, copyright year 2024 or older | 144 | 9.6% |
| Someone else's website | 139 | 9.3% |
| Could not tell (bot wall, empty page) | 137 | 9.1% |
| Parked or for sale | 61 | 4.1% |
| Placeholder or empty app shell | 48 | 3.2% |
The six rows from "no working site" through "placeholder", leaving out "could not tell", add up to 764 companies, 50.9% of the sample. With 1,500 checked, the margin is about 2.5 points either way, so between 48.4% and 53.5%. Applied to the 17,904 that is about 9,100 companies, between 8,700 and 9,600. Source: FundingWatcher, September 2026.
144 of the 764 still load with the same company on them, so dead is too strong for those. Their copyright line has not moved since 2024 or earlier: 36 stop at 2024, 24 at 2023 and 21 at 2021, and eight still say 2015. We count 2024 as stale because we ran the check in September 2026.
139 of the 1,500 domains now belong to someone else, 30 of them to gambling sites
139 of the 1,500 (9.3%) show a different business, and 30 of those are casinos, slot sites, lotteries or betting sites. We think buyers pick these domains for the links that funding announcements and press left behind, but we did not measure that.
Four examples from the sample, with the total funding Crunchbase records:
- Arrivedo, a travel content startup that raised $3.88 million, now serves a page titled "Casino Review - Arrivedo".
- VOIR (voir.me), an app for animated stickers of your own face, backed by Bulba Ventures with $1 million raised, now lists online pokies for New Zealand.
- InSimu (insimu.com), a clinical practice simulator for doctors backed by Hiventures, now ranks foreign online casinos for Hungarian players.
- CareOnGo, an Indian pharmacy chain that raised $900,000, now serves "Top Online Casinos for Fast Cashouts".
Each younger cohort reaches Series A less often, from 35.7% to 23.8%
| Founded | Companies | Series A or later | Share |
|---|---|---|---|
| 2012 to 2014 | 21,921 | 7,825 | 35.7% |
| 2015 to 2017 | 30,826 | 9,848 | 31.9% |
| 2018 to 2019 | 22,278 | 5,292 | 23.8% |
| 2020 to 2021 | 27,915 | 4,191 | 15.0% |
| 2022 to 2023 | 22,010 | 2,122 | 9.6% |
The 2018 to 2019 cohort has had at least seven years. The 2020 to 2023 cohorts are still raising, so their shares will rise.
The funnel runs from 1,723,352 Crunchbase rows to 1,500 websites checked
- 1,723,352 organisation rows in our Crunchbase copy.
- 325,375 with any funding round.
- 30,826 founded 2015 to 2017 with a seed-stage history: a last round of pre-seed, seed or angel, or a Series A or later round with at least two rounds on file. Companies whose only round was a Series A are left out. So are companies whose last round is debt (other than post-IPO debt), a grant, private equity or a series Crunchbase did not label.
- 20,978 with no round after seed; 18,047 of them with no acquisition and no closed flag.
- 17,904 of those with a usable domain. 143 dropped: no domain, or a subdomain on a shared host where DNS says nothing about the company.
- 1,500 drawn at random and checked.
Method
Stage is the last round type in Crunchbase, because we have no round-by-round history. Pre-seed, seed and angel count as seed stage, Series A to J as Series A or later, and any post-IPO round as public. Two rounds on file does not prove one of them was a seed round, so a company that raised a Series A and then a Series B with no seed would still be in the 30,826. That can push the 31.9% up by an amount we cannot measure from this data.
Every one of the 1,500 homepages was requested over HTTPS, then HTTP, through a rotating proxy. No resolver finding the domain, confirmed on a second resolver, means "no longer registered". A domain that resolves and serves nothing after a retry means "no working site". The copyright year is the highest year next to a copyright sign or the word copyright; pages that print the year with JavaScript count as having none. Parked pages were caught by marketplace names (Sedo, Dan, Afternic, HugeDomains) and for-sale phrases.
Gemini 3.5 Flash-Lite, a small language model, read the company name and description next to the page title, final URL and the end of the page text, and said whether the site still belongs to the company. We read all 30 gambling titles and footers by hand and moved two back to "same company". We name no company from the model's other different-business verdicts, because that group mixes in rebrands and acquirers.
A site that was down on 23 September 2026, the day of the check, counts as not working.
Data
The files below are under CC BY 4.0.
- cohorts_by_founding_years.csv, the Series A share per founding cohort
- cohort_2015_2017_stage_by_outcome.csv, last stage by acquired, closed and no exit for the 30,826
- site_check_sample.csv, the 1,500 checked companies with domain, description, last round, website state, copyright year and page title
- site_check_summary.csv, the state counts with the estimate for the 17,904
- README.md and METHOD.md
Figures used: funded startups.