Startup Failure Statistics 2026: Sources and Limits

A startup failure statistic can count a closed business, a lost investment or a missed forecast. Those are different outcomes. Before you use a percentage, check which one it measures and who is in the denominator.

I built Preuve AI to help you research an idea before building. For this reference, I separate original research from reporting based on provider interviews, and label the dates and limits. The September 12, 2026 revision corrects source attributions and removes claims I could not verify. It does not turn older observations into a 2026 cohort.

If you want the narrative on why startups fail (the CB Insights deep-dive, the PMF argument), I wrote that separately in why startups really fail in 2026. This page is the reference table. Bookmark it or argue with it.

Will your idea survive the market?

Preuve AI runs 10 agents against live market data and links every claim to a source. Free analysis in 60 seconds.

Validate My Idea FreeRated 5/5 · 14 reviews

What percentage of startups fail?

I would not put one universal percentage in a pitch deck. A survival study needs a defined cohort and a follow-up period; a list of closures cannot tell you what share of all startups failed.

In an HBS author interview dated March 7, 2011, Shikhar Ghosh described how estimates change with the definition. These are historical interview estimates, not a newly verified longitudinal dataset or current failure probabilities.

Missing a return target does not mean returning no money. Liquidation with substantial investor losses does not always mean total loss. Keep the definition attached when you quote a range.

What are the top reasons startups fail?

The CB Insights report published March 5, 2026 covers 431 VC-backed companies that publicly shut down since 2023, excluding companies with a prior exit. It draws on public post-mortems, founder interviews and shutdown announcements. The reason percentages use the subset of 385 companies with identifiable failure reasons. Categories overlap.

CB Insights interprets running out of capital as an endpoint and the other listed reasons as underlying problems. That is an interpretation of selected shutdown accounts, not a causal experiment. The full 431-company cohort raised $17.5 billion in equity funding before closure, with a median of $11 million. Funding raised is not a measurement of cash burned or investor losses.

The report gives a median of 22 months from last fundraise to shutdown. Nearly a quarter closed more than three years after their last raise. This interval does not measure remaining cash or when founders recognized a problem. I discuss the product-market-fit argument further in why startups really fail.

What is the startup failure rate by industry?

The BLS Business Employment Dynamics survival tables are the primary starting point for U.S. employer establishments by industry. An establishment is not interchangeable with a startup, a firm or every LLC registration. A NAICS sector also need not match a product category such as SaaS.

I could not retrieve the original BLS tables during this revision. The previous table relied on Commerce Institute's secondary reporting, which was also unavailable for verification. I have withdrawn the numerical table, its industry ranking and the claimed industry gap. None should be treated as citation-ready here.

For an industry comparison, retrieve a named birth cohort and compare the same follow-up horizon across sectors. Record the table version and distinguish survival from closure. Arithmetic on an unverified table would not validate the underlying figures.

My startup validation benchmarks answer a different question: how Preuve model assessment scores are distributed. They contain no observed startup survival rates.

How does the startup failure rate change by funding stage?

The sources checked here do not support a universal failure probability for each funding stage. Failing to raise the next round is not the same outcome as closing a business. A rate would need all companies entering that stage, a defined closure event and a follow-up period.

TechCrunch's January 26, 2025 provider interviews report a narrower finding from SimpleClosure: 74% of its observed shutdowns since 2023 were at pre-seed or seed, including 41% at seed. These are shares of closures, not stage-specific failure probabilities. Carta is a separate source in the same article.

The number of companies at each stage is missing from that comparison. A large share of shutdowns at seed could reflect how many companies are at seed, their risk, or both. I have removed the earlier stage-probability table because it lacked a primary longitudinal source.

Skip weeks of manual research

Get complete market research, sourced proof, competitor map, and pricing data for your idea instantly.

Run a Free Analysis

What shutdown counts were reported for 2023 and 2024?

The figures below come from Mary Ann Azevedo's January 26, 2025 TechCrunch reporting based on provider interviews, not provider-native raw datasets independently checked here. The two reported series increased between 2023 and 2024; they do not measure all U.S. startup closures.

Carta counted U.S.-based Carta customers that left because of bankruptcy or dissolution. Its head of insights, Peter Walker, acknowledged missing shutdowns. AngelList reported winddowns in its own tracked sample. Different coverage and possible overlap mean the totals should not be added together or treated as population-wide rates.

TechCrunch reported Carta sector shares of closures: enterprise SaaS 32%, consumer 11%, health tech 9%, fintech 8% and biotech 7%. These sector shares are not individual failure probabilities; Walker said they broadly aligned with initial funding shares. A larger sector can have more closures without a higher closure rate.

In CB Insights' separate since-2023 cohort, 62 healthcare/biotech companies had raised $5.1 billion in equity funding. That is capital raised before closure, not measured burn or losses.

Walker offered a working hypothesis: if more companies received funding without a better selection hit rate, more closures could follow. The article also carried a forecast for 2025, not an observed 2025 or 2026 result. Neither the hypothesis nor the forecast measures whether those founders skipped customer research.

What does "startup failure" actually mean?

The sources measure different events. Keep the source's population and endpoint attached to each figure.

If you need a probability for your own planning, first define the event you care about. Closing, losing investor capital and missing a growth target require different evidence.

How do first-time founders compare to repeat founders?

An HBS author interview dated February 2, 2009 reports results from research by Paul Gompers, Josh Lerner, David Scharfstein and Anna Kovner. The endpoint is an IPO of a venture-backed company, not generic startup survival or profitability. These figures belong to that historical interview and its definition.

Prior success is associated with a higher IPO success figure in this study. The comparison does not isolate preparation or show that using a validation product reproduces the difference. I would not apply these historical venture-backed figures to every first-time founder today.

How to read these statistics without getting paralyzed

These statistics describe different selected populations and historical periods. None provides a personal probability of failure, and the sources do not estimate how much a Preuve scan changes business outcomes.

The CB Insights categories give you questions to investigate, such as product-market fit, timing and unit economics. They do not prove that every company running out of cash ignored demand. For historical examples of timing and market changes, I wrote about dead startup ideas worth revisiting.

The historical Startup Genome premature-scaling report, version 1.2 edited March 2012, studied high-growth technology startups, including unfunded firms, and explicitly lacked longitudinal data. It reported 3.6x user growth among startups that pivoted once or twice compared with those that pivoted zero or more than two times. This is an association, not a causal estimate that pivoting prevents failure. It does not establish a current universal failure rate or a venture-return definition of failure.

For your own idea, test a specific assumption with potential buyers. My idea validation guide covers ways to do that. A free Preuve scan can help organize desk research, but customer commitments and business outcomes need separate evidence.