Startup Failure Rate Statistics 2026
By Axis Intelligence Research
Co-author: Mia Scarlett (Business & Filings) | Last updated: August 8, 2026 | License: CC BY 4.0
22.1% of new U.S. businesses close within one year, 48.6% within five years and 65.3% within ten, according to Bureau of Labor Statistics cohort data published through March 2025. Among venture-backed companies the picture is harsher: 431 tracked shutdowns since 2023 destroyed $17.5 billion in equity, with poor product-market fit cited in 43% of post-mortems.
Quick Answer: What Is the Startup Failure Rate?
There is no single startup failure rate, and the two numbers that circulate most are measuring different populations. For all U.S. private-sector establishments, the failure rate is 22.1% at one year and 48.6% at five years (BLS, cohorts tracked to March 2025). For venture-backed startups, the outcome distribution is far worse, because the bar is a return multiple rather than survival.
According to Axis Intelligence Research, the gap between those two populations is measurable and it narrows sharply with age. Our Startup Mortality Divergence Index (SMDI™) puts the Information sector’s first-year failure rate at 1.29x the all-industry baseline, but only 1.07x by year ten.
Key Findings
- According to Axis Intelligence Research, 22.1% of U.S. establishments born in the year to March 2024 failed within twelve months, based on Bureau of Labor Statistics cohort survival data released through March 2025.
- The record 1,065,228-establishment cohort born in the year to March 2022 recorded 76.3% first-year survival — the weakest of any cohort since 2008, per Axis Intelligence Research analysis of BLS Table 7.
- Information-sector establishments born in the year to March 2024 posted 71.6% first-year survival, the lowest reading in that sector since the 2001 dot-com cohort, according to Axis Intelligence Research.
- Axis Intelligence Research derives that roughly 270 of 385 classified venture-backed post-mortems cited running out of capital, against 166 citing poor product-market fit, from percentages published by CB Insights in March 2026.
- The Axis Startup Mortality Divergence Index (SMDI™) reads 1.29 at year one and 1.07 at year ten, indicating the tech failure premium is concentrated almost entirely in a company’s first twelve months.
What Is the Startup Failure Rate in 2026?
The answer depends entirely on which population you count, and most published figures never say. The U.S. Bureau of Labor Statistics tracks every private-sector establishment born in a given year and reports how many survive each subsequent year. That series, Table 7 of the Business Employment Dynamics establishment age data, is the only continuous, census-grade cohort record of American business survival, running from 1994 to March 2025.
First-Year, Five-Year and Ten-Year Failure Rates
| Tenure | Birth cohort (year ended) | Survival rate | Failure rate | Source |
|---|---|---|---|---|
| 1 year | March 2024 | 77.9% | 22.1% | BLS BDM Table 7 |
| 2 years | March 2023 | 65.9% | 34.1% | BLS BDM Table 7 |
| 3 years | March 2022 | 56.3% | 43.7% | BLS BDM Table 7 |
| 4 years | March 2021 | 52.4% | 47.6% | BLS BDM Table 7 |
| 5 years | March 2020 | 51.4% | 48.6% | BLS BDM Table 7 |
| 6 years | March 2019 | 46.8% | 53.2% | BLS BDM Table 7 |
| 10 years | March 2015 | 34.7% | 65.3% | BLS BDM Table 7 |
Source: U.S. Bureau of Labor Statistics, Business Employment Dynamics, Table 7 (Total Private), data through March 2025. Retrieved August 8, 2026. Each row is a distinct birth cohort observed at the stated tenure.
Read the table as a set of cohorts rather than a survival curve for one group of companies. Each line follows a different vintage, which is why the ten-year figure comes from businesses founded a decade ago in a completely different capital environment.
The Cohort Nobody Is Talking About
The most interesting number in the BLS series is not a rate. It is a count. The cohort born in the year ended March 2022 contained 1,065,228 establishments — the largest single birth cohort in the thirty-one-year history of the series, and the only one to clear a million.
That cohort then recorded 76.3% first-year survival. According to Axis Intelligence Research, only two cohorts in the entire series survived their first year worse: 2001 (75.7%) and 2008 (75.2%). Both were recession vintages. The 2022 vintage was not born into a recession; it was born into the tail of the cheapest capital in modern history, and it still underperformed every non-recession cohort on record.
Mia Scarlett: A record birth cohort with bottom-quartile survival is not a paradox, it is a denominator problem. When formation runs that far above trend, the marginal new establishment is by definition weaker than the median one — thinner capitalisation, less committed founders, more speculative demand assumptions. The 2022 cohort is now three years old and 43.7% of it is gone. The interesting question for anyone reading filings is not whether that cohort underperformed, but whether the 2023 and 2024 vintages, which are only slightly smaller, are tracking the same path. On the evidence to March 2025, they are: 78.2% and 77.9% first-year survival, both below the series median.
Why Both the 90% Figure and the 20% Figure Are Real
The “90% of startups fail” line and the “80% survive year one” line are not in conflict. They count different things:
- BLS counts establishments. A dentist’s office, a food truck and a seed-stage SaaS company all count once. The formation side of the same series — how many businesses are being started in the first place — is tracked in our startup statistics report. Closure includes voluntary wind-downs, retirements and relocations, not only insolvency.
- Venture datasets count outcomes against a return threshold. A company that reaches $5 million in profitable revenue after raising a Series A can be classified as a failure by an investor whose fund needs a 10x.
Neither definition is wrong. Quoting one while implying the other is. Any figure above roughly 65% lifetime failure is describing venture-backed companies or applying a returns-based definition, and should say so.
Why Do Startups Fail? Post-Mortems Counted by Cause
Most articles on this topic still cite a CB Insights study of 110-plus post-mortems in which “no market need” led at 42%. That study has been superseded. In March 2026, CB Insights published a rebuilt analysis of 431 venture-backed companies that shut down since 2023, of which 385 had failure reasons that could be classified. The cause ranking changed.
The 385 Classified Post-Mortems, by Cause
CB Insights published shares, not counts. According to Axis Intelligence Research, applying each published share to the disclosed classified denominator of 385 companies yields the following counts:
| Cause cited | Share of post-mortems | Axis-derived count | Calculation | Source |
|---|---|---|---|---|
| Ran out of capital | 70% | ~270 companies | 0.70 x 385 = 269.5 | CB Insights, Mar 2026 (share); Axis (count) |
| Poor product-market fit | 43% | ~166 companies | 0.43 x 385 = 165.6 | CB Insights, Mar 2026 (share); Axis (count) |
| Bad timing / macro conditions | 29% | ~112 companies | 0.29 x 385 = 111.7 | CB Insights, Mar 2026 (share); Axis (count) |
| Unsustainable unit economics | 19% | ~73 companies | 0.19 x 385 = 73.2 | CB Insights, Mar 2026 (share); Axis (count) |
Source: shares from CB Insights, “The top 9 reasons startups fail,” March 5, 2026. Counts calculated by Axis Intelligence Research from the published shares and the stated classified denominator (385). Companies cited multiple causes, so shares sum above 100%. Counts are rounded to whole companies and are estimates, not observed tallies.
Running out of capital tops the list and explains almost nothing. It is the terminal event, not the diagnosis. Two-thirds of the product-market-fit failures were early-stage companies that never found a market at all — but twenty Series B or later companies also cited it, meaning they raised on early traction that never widened.
Bad timing clustered by sector rather than by stage. Climate and energy, food and agriculture, and blockchain absorbed a disproportionate share, all three having attracted heavy capital in 2021 and 2022 against trends that did not arrive.
What the Final Twelve Months Look Like
CB Insights also published deterioration signals drawn from its own platform data, which are more useful to an operator than the cause taxonomy:
| Signal | Reading | Population | Source |
|---|---|---|---|
| Health score declined in year before death | 72% | Companies with 12-month score data | CB Insights, Mar 2026 |
| Average decline in health score | 15% | Same | CB Insights, Mar 2026 |
| Tracked partnership activity drop | 44% | 206 companies (48% of dataset) | CB Insights, Mar 2026 |
| Headcount shrinking in final 6 months | Two-thirds | Companies with headcount data | CB Insights, Mar 2026 |
| Died with 10 or fewer employees | Nearly one-third | Full dataset | CB Insights, Mar 2026 |
| Died with more than 100 employees | About 15% | Full dataset | CB Insights, Mar 2026 |
Mia Scarlett: The partnership signal is the one a credit analyst would reach for first, because it is the hardest to manage. Headcount can be cut deliberately and reads as discipline. A health score is a third party’s opinion. But business-development relationships thinning by 44% is counterparties voting with their calendars, and counterparties see the operating reality before the cap table does. If you want an early read on a private company you cannot get financials from, count who is still willing to co-sell with it.
How Much Riskier Is a Tech Startup? The Axis SMDI™ Reading
Everyone assumes technology startups fail more often than businesses generally. Almost nobody has measured the gap against a like-for-like cohort, because it requires reading two BLS cohort tables side by side. Axis Intelligence Research built the Startup Mortality Divergence Index (SMDI™) to do exactly that.
SMDI™ stands for Startup Mortality Divergence Index. It expresses how much more likely a technology-sector establishment is to fail than an average U.S. establishment born in the same year and observed at the same age.
How SMDI™ Is Calculated
SMDI(t) = Information-sector cumulative failure rate at tenure t
------------------------------------------------------
Total-private cumulative failure rate at tenure t
Where, for each birth cohort, cumulative failure rate = 100 − (BLS “survival rate since birth”). Both numerator and denominator are drawn from the same cohort year and the same observation date, which removes vintage effects and macro conditions from the comparison. Inputs come from BLS Business Employment Dynamics Table 7, Total Private and NAICS 51 (Information).
A reading of 1.00 would mean tech startups fail at exactly the national rate. Above 1.00 means they fail faster.
SMDI™ Readings by Tenure (Baseline, as of March 2025)
| Tenure | Cohort (year ended) | Information failure | All-industry failure | SMDI™ |
|---|---|---|---|---|
| Year 1 | March 2024 | 28.4% | 22.1% | 1.29 |
| Year 3 | March 2022 | 48.8% | 43.7% | 1.12 |
| Year 5 | March 2020 | 54.3% | 48.6% | 1.12 |
| Year 10 | March 2015 | 70.0% | 65.3% | 1.07 |
Source: Axis Intelligence Research calculation from U.S. Bureau of Labor Statistics Business Employment Dynamics Table 7 (Total Private and NAICS 51 Information), data through March 2025. Snapshot date: March 2025. This is the baseline reading of the index.
According to Axis Intelligence Research, the finding worth citing is the shape of that curve, not its level. Technology startups do not fail at a constant premium to the wider economy. They fail at a 29% premium in year one, and by year ten that premium has decayed to 7%. The Information sector’s danger is concentrated almost entirely in the first twelve months. A tech company that reaches its second birthday is running at close to national baseline risk thereafter.
That reframes a common piece of founder advice. The extra risk in building a software company is not distributed across a decade of competitive pressure; it is front-loaded into the validation window, where the cost of being wrong is lowest and the ability to change course is highest.
Startup Failure Rate by Industry
The Information Sector, Cohort by Cohort
| Cohort (year ended) | Births | Yr 1 survival | Yr 3 survival | Yr 5 survival | Source |
|---|---|---|---|---|---|
| March 2019 | 21,223 | 75.6% | 56.8% | 45.4% | BLS Table 7, NAICS 51 |
| March 2020 | 22,098 | 79.6% | 59.0% | 45.7% | BLS Table 7, NAICS 51 |
| March 2021 | 30,925 | 74.2% | 54.1% | — | BLS Table 7, NAICS 51 |
| March 2022 | 45,393 | 75.6% | 51.2% | — | BLS Table 7, NAICS 51 |
| March 2023 | 42,098 | 75.1% | — | — | BLS Table 7, NAICS 51 |
| March 2024 | 32,515 | 71.6% | — | — | BLS Table 7, NAICS 51 |
| March 2025 | 30,594 | — | — | — | BLS Table 7, NAICS 51 |
Source: U.S. Bureau of Labor Statistics, Business Employment Dynamics Table 7, NAICS 51 (Information), data through March 2025. Retrieved August 8, 2026. Blank cells indicate tenure not yet reached.
Two readings stand out. Information-sector births ran from 21,223 in the year to March 2019 to 45,393 in the year to March 2022, a 114% increase in three years, then fell back to 30,594 by March 2025. And the cohort born in the year to March 2024 survived its first year at 71.6% — a level the sector has not seen since the 2001 cohort’s 65.2%, according to Axis Intelligence Research analysis of the full BLS series.
Which Sectors Dominate Shutdown Counts
Establishment data and venture shutdown data disagree about which sectors are dangerous, because they are counting different animals. SimpleClosure’s 2025 State of Startup Shutdowns, built from several hundred structured dissolutions, tracks the venture side:
| Sector | Share of shutdowns 2024 | Share of shutdowns 2025 | Change (pp) | Source |
|---|---|---|---|---|
| AI | 17.7% | 15.9% | −1.8 | SimpleClosure, Dec 2025 |
| B2B SaaS | 5.2% | 7.7% | +2.5 | SimpleClosure, Dec 2025 |
| Developer tools | 6.0% | 6.4% | +0.4 | SimpleClosure, Dec 2025 |
| CPG | 8.5% | 6.1% | −2.4 | SimpleClosure, Dec 2025 |
| Fintech | 5.6% | 4.3% | −1.3 | SimpleClosure, Dec 2025 |
| Biotech and diagnostics | 1.2% | 3.6% | +2.4 | SimpleClosure, Dec 2025 |
| Healthcare | 0.8% | 2.8% | +2.0 | SimpleClosure, Dec 2025 |
| Climate and energy | 2.0% | 2.8% | +0.8 | SimpleClosure, Dec 2025 |
Source: SimpleClosure, State of Startup Shutdowns 2025, published December 3, 2025. Shares are of all shutdowns processed in each year.
By count of failed companies rather than share of dissolutions, CB Insights ranks healthcare and biotech first (62 companies, 14% of its 431, and $5.1 billion in equity destroyed), fintech second (57 companies, 13%) and food and agriculture third (54 companies, 13%).
The fintech entry carries the most information. Its median equity raised was $4 million against a dataset-wide median of $11 million and a healthcare median of $47 million, and 60% of the fintech failures were based outside the United States versus 47% of the dataset overall. Fintech is failing early, cheap and internationally — a very different pattern from biotech, which fails late and expensively after clinical readouts. Our fintech funding statistics cluster tracks the capital-formation side of that same story.
How Much Capital Dies With a Failed Startup?
The 431 companies in the CB Insights dataset raised $17.5 billion in equity before shutting down. The report states a median raise of $11 million and an average of $48 million.
Those figures do not reconcile against the headline total, and the discrepancy is worth publishing rather than smoothing over. According to Axis Intelligence Research, $17.5 billion divided across all 431 companies is $40.6 million per company, not $48 million. Recovering the stated $48 million average requires a denominator of roughly 365 companies — implying the average was computed only on the subset with disclosed funding, while the $17.5 billion total is attributed to the full 431. Anyone citing “average capital destroyed per failed startup” should specify which denominator they mean. We publish both:
| Measure | Value | Denominator | Basis |
|---|---|---|---|
| Total equity raised before death | $17.5B | 431 companies | CB Insights, Mar 2026 |
| Median raise per failed company | $11M | Companies with funding data | CB Insights, Mar 2026 |
| Average raise, as published | $48M | ~365 companies (implied) | CB Insights, Mar 2026 |
| Capital per company across full dataset | $40.6M | 431 companies | Axis calculation |
The gap between an $11 million median and a $40.6 million mean is the real finding. Capital destruction in venture failure is not evenly spread; it is dominated by a thin tail of very heavily funded companies. Two of them, a healthcare AI automation business and a digital freight brokerage, each raised close to $1 billion, both reached roughly $4 billion valuations in the pandemic-era boom, and both shut down within two weeks of each other in October 2023.
Mia Scarlett: A median of $11 million tells you what a typical failure costs. A mean of $40.6 million tells you what the portfolio costs. Those are different questions with different audiences, and conflating them is how a shutdown wave gets described either as a rounding error or as a catastrophe depending on which number the writer picked up. For anyone modelling reserve requirements, the median is the wrong statistic entirely — the tail is where the fund’s capital actually goes to die.
When Do Startups Fail? The Funding Cliff
Median time from a company’s last fundraise to its death is 22 months, per CB Insights. Over half the dataset died within two years of the last money in.
Nearly a quarter had gone more than three years since their last raise before formally shutting down — companies operating without new capital and without a formal wind-down, running on residual cash, revenue or founder subsidy. CB Insights counts roughly 50,000 venture-backed startups that have not raised since the start of 2023.
That population is the forward indicator for the next several years of shutdown counts, though the share of it that will fail rather than reach profitability or sell is not knowable from the published data. Shutdown volumes on Carta’s platform, which are drawn from U.S. companies leaving the platform through bankruptcy or dissolution, rose through 2024 to 966 for the year against 769 in 2023. Carta’s own head of insights has cautioned that the platform undercounts, because companies leave without stating a reason.
Our AI investment statistics report covers the capital going in; this is the capital coming out. SimpleClosure’s 2025 data shows the wave moving up-market: Series A shutdowns rose from roughly 6% to roughly 14% of all closures in a single year, a 2.5x increase. Companies shutting down at Series A now average around seven years old, and Series B and later closures are closer to a decade old. The correction has moved past failed ideas to failed models.
Startup Failure Rates Outside the United States
Comparable cohort survival data outside the U.S. is thinner than most cross-country comparisons admit. Eurostat’s business demography statistics, the closest European equivalent to the BLS series, tracks enterprise births, deaths and survival across the EU business economy.
| Indicator | EU, 2023 | Source |
|---|---|---|
| Active enterprises | Over 33 million | Eurostat (bd_size) |
| Enterprise births | 3.5 million | Eurostat (bd_size) |
| Enterprise birth rate | 10.5% | Eurostat (bd_size) |
| Enterprise death rate (preliminary) | 8.5% | Eurostat (bd_size) |
| Highest birth rate | Lithuania, 19.6% | Eurostat (bd_size) |
| Lowest birth rate | Austria, 6.2% | Eurostat (bd_size) |
| Highest death rate | Estonia, 27.5% | Eurostat (bd_size) |
| Lowest death rate | Hungary, 2.6% | Eurostat (bd_size) |
Source: Eurostat, Business demography statistics, data from October 2025, reference year 2023. Retrieved August 8, 2026.
Here is the caveat that most comparison articles skip. The current Eurostat edition does not publish cohort survival percentages at one and five years in the way BLS Table 7 does; its survival section reports employment growth among surviving enterprises instead. According to Axis Intelligence Research, this means published claims of the form “EU startups survive at X% versus US at Y%” cannot be sourced to the current edition of either statistical system, and figures circulating for EU one-year and five-year survival generally trace to earlier editions on a different methodological basis.
The two systems also count different units. BLS counts establishments (physical locations); Eurostat counts enterprises (legal entities). Eurostat excludes agriculture, forestry, fishing and public administration; BLS Total Private includes agriculture. A single-multiple-site U.S. company generates several BLS establishment records and one Eurostat-equivalent enterprise record. Birth and death rates are comparable in direction, not in level.
Do 95% of AI Startups Fail?
No, and the figure that generates this claim measures something else entirely.
The widely repeated 95% comes from MIT’s Project NANDA report The GenAI Divide: State of AI in Business 2025, which found that about 95% of enterprise generative-AI pilots delivered no measurable profit-and-loss impact, with roughly 5% of integrated pilots producing significant value. That is a statement about corporate pilot programmes inside established companies. It is not a startup failure rate, and the companies in the denominator are overwhelmingly not startups. Published descriptions of the study’s sample also vary — some report 150 executive interviews and 350 employee surveys, others 52 interviews and 153 leaders, alongside an analysis of 300 public deployments — so the sample should be described with care by anyone citing it.
The actual AI-startup failure evidence is more specific and less dramatic. SimpleClosure’s 2025 data puts AI at 15.9% of shutdowns, down from 17.7% a year earlier, with AI companies representing roughly 12% to 20% of shutdowns within most funding stages. The median AI company that shut down had raised about $2.4 million, slightly below the $2.8 million dataset median.
The concentration is in one product shape. AI shutdowns skew heavily toward wrappers and application-layer tools — copilots, content generators and vertical software whose differentiation is a model front-end rather than proprietary data or infrastructure. Infrastructure and developer-tool AI companies shut down less often, but those that do have raised roughly twice the capital of their application-layer peers, because they sell into longer enterprise integration cycles and compete directly with hyperscalers. The counter-example at the top of that market is documented in our Cursor AI statistics dataset, where an application-layer company built enough enterprise depth to avoid exactly that trap.
Mia Scarlett: Two things are true at once and the coverage keeps picking one. AI is not over-represented in shutdown counts relative to its share of formation. But within AI, the wrapper cohort is failing exactly the way a credit analyst would expect a business with no switching costs and a variable input cost it does not control to fail. Gross margin compresses when the model provider reprices, and there is no contractual defence. That is a business-model problem wearing a technology label, and it will not be fixed by a better model.
The capital-formation side of the same market is covered in our AI investment statistics report, and the two largest privately held model developers are tracked in our OpenAI statistics and Anthropic statistics datasets.
Methodology
How this dataset was assembled. Every figure in this report was retrieved from a primary source document during the production session on August 8, 2026. No statistic is drawn from model memory or from secondary aggregation. Source URLs and retrieval dates are recorded per row in the accompanying CSV.
Cohort survival data. U.S. survival and failure rates come from the Bureau of Labor Statistics Business Employment Dynamics programme, Establishment Age and Survival Data, Table 7, “Survival of private sector establishments by opening year,” for Total Private and for NAICS 51 (Information). BLS publishes survival rates since birth; failure rates in this report are the arithmetic complement (100 − survival) and are labelled as Axis-calculated in the CSV. The series runs to the year ended March 2025; the fourth-quarter 2025 BDM release was scheduled for July 29, 2026 and may extend the age tables in a future edition.
SMDI™ construction. The Startup Mortality Divergence Index divides the Information-sector cumulative failure rate by the Total Private cumulative failure rate for the same birth cohort at the same observation date. Both inputs are BLS-published survival rates. No weighting, smoothing or normalisation is applied; the index is a direct ratio, reproducible from the two cited BLS tables by anyone with a calculator. Readings are reported to two decimal places. The March 2025 readings published here are the baseline for the series. Methodology is dated rather than numbered, so that any future revision is identifiable by the date it takes effect.
Post-mortem cause counts. CB Insights published cause shares against a stated classified denominator of 385 companies (from a 431-company shutdown dataset). Axis Intelligence Research multiplied each published share by 385 and rounded to whole companies. These counts are estimates derived from published percentages, not observed tallies, and are flagged as such in the CSV. Because companies cited multiple causes, shares sum to 161% and counts cannot be added.
Capital-per-failure calculation. $17,500,000,000 ÷ 431 companies = $40.6 million. The implied denominator behind the published $48 million average is $17,500,000,000 ÷ $48,000,000 ≈ 365 companies. Both results are disclosed rather than reconciled, because reconciling them would require access to the underlying company-level funding records.
Source selection. Priority went to statistical agencies (BLS, Eurostat) for population-level rates, and to organisations holding the underlying operational records (CB Insights, SimpleClosure, Carta) for venture-specific data. Secondary aggregators were used to locate primary documents and are not cited as sources of any figure. Where a widely circulated statistic could not be traced to a current primary edition, it is either excluded or explicitly flagged.
Scope and Caveats
This report covers what the cited datasets measure, and several things it does not:
- Establishment closure is not the same as business failure. BLS records an establishment death when a location stops reporting employment. That includes voluntary retirement, sale, relocation and consolidation alongside insolvency. The 22.1% first-year figure is an upper bound on genuine failure.
- Venture datasets are platform-limited. Carta observes companies on Carta; SimpleClosure observes dissolutions it processes; CB Insights observes publicly announced shutdowns. Each undercounts, and none is a random sample of the venture population.
- Cohort tables mix vintages. The ten-year failure rate describes companies founded in 2014-15. It is not a forecast for companies founded today.
- Cause attribution is self-reported. Post-mortem causes come from founders and shutdown announcements, which have obvious incentives around framing.
- EU and U.S. figures are not directly comparable. Different statistical units, different sector coverage, different reference periods.
About This Dataset
startup-failure-rate-statistics-2026.csv contains every quantitative claim in this report as a single row, with metric, value, unit, as-of date, source organisation, source document, source URL, retrieval date, primary-source flag, Axis-calculated flag and method note.
License: CC BY 4.0. Free to share and adapt, including commercially, with attribution.
Attribution line: Axis Intelligence Research, Startup Failure Rate Statistics 2026, axis-intelligence.com
Temporal coverage: 1994-2026 (BLS cohort series); 2023-2026 (venture shutdown datasets) Spatial coverage: United States; European Union
Cite This Report
APA: Axis Intelligence Research. (2026, August 8). Startup Failure Rate Statistics 2026: Survival Rates, Post-Mortems and the Real Numbers. Axis Intelligence. https://axis-intelligence.com/startup-failure-rate-statistics/
MLA: Axis Intelligence Research. “Startup Failure Rate Statistics 2026: Survival Rates, Post-Mortems and the Real Numbers.” Axis Intelligence, 8 Aug. 2026, axis-intelligence.com/startup-failure-rate-statistics/.
Chicago: Axis Intelligence Research. “Startup Failure Rate Statistics 2026: Survival Rates, Post-Mortems and the Real Numbers.” Axis Intelligence, August 8, 2026. https://axis-intelligence.com/startup-failure-rate-statistics/.
Frequently Asked Questions
If my company survives year one, what does the data say my odds look like after that?
Better than most founders assume, and the improvement is steep. BLS conditional survival — the share of one year’s survivors that make it to the next — sits above 93% for every year after the fourth across recent cohorts, against roughly 78% in year one. Year one removes about a fifth of a cohort. Each subsequent year removes about one twentieth of what remains. For a technology company the effect is stronger still: the Axis SMDI™ reading falls from 1.29 at year one to 1.07 by year ten.
Which is the honest number to quote to my investors: 22% or 90%?
Neither in isolation. If you are describing the population your company belongs to statistically, the BLS cohort figures are the defensible ones. If you are describing the probability of delivering a venture-scale return, the venture figures apply and you should say that is the definition being used. The failure mode in board materials is quoting the 22% survival-basis number alongside a returns-basis growth case.
Does the record 2022 formation wave actually predict more shutdowns ahead?
Partly, and the mechanism matters. The cohort born in the year to March 2022 was the largest on record at 1,065,228 establishments and had already lost 43.7% of its members by March 2025. Because the cohort is so large in absolute terms, even an ordinary failure rate produces an unusually high count of closures. Expect elevated shutdown volumes without concluding that per-company risk has risen proportionally.
How long after our last round should we worry?
The CB Insights median from last raise to shutdown is 22 months. That is a distribution midpoint, not a deadline, and it partly reflects a standard 18-to-24-month runway plan colliding with a raise that does not close. The more useful signal in that dataset is what happens beyond three years post-raise: nearly a quarter of the failures had passed that mark, which describes companies that stopped raising but never formally decided to stop.
We are an AI application company. Is the shutdown data actually about us?
The shutdown data is disproportionately about your product shape rather than your sector. Application-layer AI — copilots, generators, vertical tools built on a third-party model — dominates AI closures, and the recurring pattern is margin compression when model pricing shifts combined with low switching costs. Infrastructure and tooling companies fail less often but more expensively. The median AI company that shut down had raised about $2.4 million.
Is “ran out of cash” a real cause of failure or a placeholder?
A placeholder, and treating it as a diagnosis is the most common analytical error on this topic. It appears in about 70% of classified post-mortems — roughly 270 of 385 companies by Axis Intelligence Research’s derivation — which is close to a definitional statement, since almost every shutdown ends in cash exhaustion. The causes that carry information are the ones upstream: product-market fit at 43%, timing at 29%, unit economics at 19%.
Which sector should I be benchmarking against, not the all-industry average?
For a software or digital-media company, NAICS 51 (Information) is the closer comparator, and it fails materially faster in the early years: 28.4% first-year failure for the cohort born to March 2024, against 22.1% for all industries. If your company is capital-intensive with a long validation cycle — biotech, climate hardware — the venture shutdown data is more relevant than establishment data, because your failure event is a failed readout or a failed raise rather than a location closing.
Are startup failure rates actually getting worse, or does it just feel that way?
On the establishment data, modestly worse and concentrated in the newest vintages. First-year survival for the three most recent cohorts (76.3%, 78.2%, 77.9%) sits below the series median, and the 2022 reading was the weakest since the 2008 recession cohort. In the Information sector the deterioration is sharper: the year-to-March-2024 cohort’s 71.6% first-year survival is the lowest since 2001. On the venture side the count of shutdowns rose through 2024 and the mix moved up-market in 2025, which is a different phenomenon — later-stage companies reaching the end of capital raised years earlier.