The State of Startup Accelerators in 2026
We track 3,133 accelerator programs in 108 countries — and we ping every program's website monthly. 637 are gone. This census covers how many accelerators actually exist, where they die, and what the survivors charge. Every number below is computed from the same public dataset that powers this site.
- 3,133 accelerator programs tracked across 108 countries — the largest openly browsable accelerator dataset.
- 20% of accelerator websites (637 of 3,123 checked) are dead — offline, parked, or gone.
- Accelerators founded 2015–2020 have died at 22%, vs 14% for programs founded 2000–2009.
- New-accelerator launches peaked around 2016 and have collapsed since.
- Median equity taken: 6% (of 344 programs that disclose terms). 49 programs are equity-free.
- Median disclosed investment: $50K. Median program length: 4 months.
- 59% of programs run hybrid, 35% in person, 7% fully remote.
1 in 5 accelerators is gone
On Jul 23, 2026 we requested every one of the 3,123 accelerator websites in our directory. 637 (20%) no longer resolve, return errors, or are parked domains. Another 156 block automated checks, so their status is unknown.
Most directories keep listing programs long after they shut down — a dead accelerator rarely announces it. This is why the widely repeated “7,000+ accelerators worldwide” figure should be treated with suspicion: no living public dataset backs it, and among programs we can actually verify, a fifth are already gone. Dead programs stay in our directory labeled as dead, because founders deserve to know both.
Where accelerators die
Share of accelerator websites now dead, for every country where we track at least 50 programs:
The spread is wide: Australia (35%) has roughly 3× the dead-rate of Japan (12%). Ecosystems that expanded fastest in the mid-2010s — often on time-limited government or corporate funding — are shedding the most programs now.
The launch boom is over
Founding year of the programs in our dataset. Launches accelerated through the 2010s, peaked around 2016, and fell off sharply after 2019:
Note: the most recent years are undercounted — new programs typically enter directories a year or two after launching. The post-2019 decline is real, but its exact depth is uncertain; see methodology.
Younger accelerators die faster
You might expect the oldest programs to have died at the highest rate — they've had the longest to fail. The opposite is true:
Programs that survived the 2000s were built to last — many are institutions today. The 2015–2020 launch wave, by contrast, includes hundreds of thinly funded corporate and government initiatives that quietly stopped operating within a few years. Survivorship, not safety, explains the veterans' better numbers.
What the survivors charge
Terms are the least-disclosed fact in the industry: only 344 of 3,133 programs (11%) publish an equity percentage at all. Among those that do, the median is 6%, and 49 programs take no equity — mostly university- and government-backed.
Investment amounts are similar: 121 programs disclose a cheque, with a median of $50K — far below the headline $500K deals of the top brand-name accelerators. The median program runs 4 months (1,134 disclose a duration), and of programs with a known format, 59% are hybrid, 35% in person, and 7% fully remote.
Who's actually accepting applications
Of the programs whose application status we've verified against their own websites this year: 180 open, 304 rolling admissions, and 159 closed between cohorts — 50 with a published deadline.
The live list, re-checked monthly: accelerators with open applications →
Methodology
The underlying directory has been compiled since 2023 from program websites, ecosystem lists, and submissions, and is continuously extended. On Jul 23, 2026 we ran an automated HTTP sweep of all 3,123 program websites (following redirects). “Dead” means the domain no longer resolves, returns a server error, or is parked; “blocked” means the site refused automated requests, so we make no claim about it. 1,790 program records additionally carry a human- or AI-verified review of their published details in 2026.
Disclosure-based stats (equity, cheque, duration) are computed only over programs that publish the figure, with the sample size stated inline — treat them as “among programs that disclose,” not industry-wide truths. Founding-year counts undercount recent years, since new programs enter directories with a lag. Spot an error? Tell us — we correct fast.
Cite this study
All figures are free to reuse with attribution:
Or explore the data yourself: the full directory of 3,133 programs, browse by country, or open applications.