The Census Bureau’s Business Formation Statistics release on October 14, 2026 counted 534,947 seasonally adjusted business applications filed in the United States during September 2026, up 0.5% from August. A year earlier, in August 2025, the same series logged 473,679. Americans are filing paperwork to start companies at a pace roughly 13% faster than a year ago, and the number shows up in pitch decks as proof that “everyone is starting a business.”
The same release also contains a quieter figure that almost nobody quotes. Census projects 29,022 business formations within four quarters from recent applications, meaning businesses that will actually put someone on payroll. Divide one number by the other and you get about 5.4%. Roughly one in eighteen applications turns into an employer business within a year.
That gap is the reason this roundup exists. Entrepreneur statistics get recycled more than almost any other category of business data, often stripped of the year, the agency, and the definition that made them true. Below are the current numbers from the Census Bureau, the SBA Office of Advocacy, the Bureau of Labor Statistics, the Kauffman Foundation, the Global Entrepreneurship Monitor, PitchBook, Carta, and Muck Rack, each with its source and date in the sentence. After the data comes a simple test, the FRAME Check, for deciding which numbers you can safely put your name next to when a reporter is reading.
Business applications are running hot, and formations are not
Start with the definitions, because the Census Bureau is careful about them and most people quoting the agency are not. A “business application” in the Business Formation Statistics series is an application for an Employer Identification Number filed with the IRS. Anyone can file one: a freelancer opening a bank account, a family setting up a rental LLC, a founder incorporating a venture-backed startup. Census separates out “high-propensity” applications, the subset with characteristics that historically predict the business will hire, such as a planned wage start date or a corporate structure.
In September 2026, 147,331 of the 534,947 applications were high-propensity, according to the October 14 release. That is about 27.5% of the total, which tells you nearly three quarters of the filings are sole proprietors, side projects, and holding entities rather than companies that plan to hire.
The annual series tells a similar story. Census data put 2023 at roughly 5.5 million applications, a record for the series at the time, and the monthly run rate in 2026 suggests the post-pandemic surge has not faded. There is one wrinkle to remember before you compare months. The Census release for December 2025, published January 14, 2026, recorded 497,046 applications, down 7.3% from November, and flagged that an IRS outage from December 30, 2025 to January 5, 2026 blocked online EIN applications. Anyone who quotes that December drop as a sign of cooling entrepreneurship is reading an IT incident as an economic trend.
“Over 500,000 Americans file to start a business every month” is accurate and sourced. “500,000 new businesses launch every month” is not, and a business reporter who knows the series will notice. The first sentence makes you look informed. The second makes you look like you copied a listicle.
How many small businesses does the US actually have?
The SBA Office of Advocacy is the standard reference here, and its June 30, 2025 report put the count at 36.2 million small businesses. Advocacy also reported that these firms account for almost 46% of private sector employment and created 1.2 million net new jobs in the period from March 2023 to March 2024.
That 36.2 million figure needs the same care as the Census applications. Most of those businesses have no employees at all. The SBA count includes nonemployer firms, which are overwhelmingly self-employed individuals, and they make up the large majority of the total.
The Global Entrepreneurship Monitor offers a different lens. GEM measures people rather than firms, through a survey of adults. Babson College, which runs the US team, released the GEM 2025/2026 US report on March 30, 2026, based on a 2025 survey of 11,000 American adults. It found 18.5% of US adults engaged in early-stage entrepreneurship, close to the highest level the survey has recorded. Babson’s Jeff Shay summarized it this way: “Entrepreneurship in the U.S. continues to be favorably viewed.”
The same report holds the more interesting number. Established business ownership, meaning businesses older than three and a half years, fell from 10.6% of adults in 2019 to 5.5%. Twice as many Americans are starting as are sustaining. GEM also found that 67% of US entrepreneurs say AI tools are important to their business model and strategy, which is useful context if you are pitching an AI angle and need a number that is not from a vendor survey.
The Kauffman Foundation’s 2025 National Report on Early-Stage Entrepreneurship, published in May 2026, measures the flow of new founders from Census survey data. Kauffman put the rate of new entrepreneurs at 0.36% of adults per month in 2025, or about 360 per 100,000 adults, still above pre-pandemic levels. It also found 83.3% of new entrepreneurs started by opportunity rather than necessity, up from 69.8% in 2020 but below the 86.9% recorded in 2019. If you want founder demographics on top of these totals, our breakdown of founder statistics for 2026 covers age, team size, and ownership.
Read survival rates before you read anything else
The most misquoted entrepreneur statistic is some version of “90% of startups fail” or “half of businesses close in the first year.” Neither comes from the federal source that actually tracks this.
The Bureau of Labor Statistics follows cohorts of new private sector establishments through its Business Employment Dynamics program. In a January 12, 2024 summary, BLS reported on establishments born in March 2013: 79.6% were still operating one year later, 50.6% after five years, and 34.7% after ten years. Put plainly, about one in five new establishments closes in year one, about half are gone by year five, and roughly a third make it a full decade.
Kauffman’s measure lands in a similar place. Its 2025 national report put the startup early survival rate at 77.9%, slightly lower than the prior year and below pre-pandemic levels. Kauffman also reported startup early job creation at 5.3 jobs per 1,000 people, unchanged from the year before.
Survival is the stat that gives a milestone meaning. A company that reaches its fifth anniversary has outlasted about half of its cohort, by the BLS numbers, and that is a fair, sourced framing for an anniversary announcement. The honest survival number is less dramatic than the myth, and reporters prefer it. A pitch that opens with “90% of startups fail” signals that the sender did not check. A pitch that cites BLS cohort data with the year signals the opposite.
Venture money is concentrating, and that changes who gets covered
Funding data moves faster than any other series in this roundup, so dates matter even more. SiliconANGLE’s April 3, 2026 write-up of the PitchBook-NVCA Venture Monitor reported a record $267 billion in US venture funding for the first quarter of 2026, dominated by AI deals involving OpenAI, Anthropic, and xAI. By the Q2 edition, covered by SiliconANGLE on July 9, 2026, first-half US deal value had reached $412.7 billion, nearly 30% more than all of 2025, with $355.9 billion, about 86% of the total, going to AI companies.
A record dollar figure driven by a handful of mega-rounds says very little about whether a typical seed-stage founder can raise. When 86 cents of every venture dollar flows to one category, the median founder outside it is competing for a smaller share, even in a “record” year.
Carta’s data adds the founder side. Its Founder Ownership Report 2025, built on anonymized data from more than 45,000 startups founded between 2015 and 2024, found that about 35% of new startups in 2024 had a single founder, up from 17% in 2017. Yet solo founders made up only 17% of startups that closed a venture round before the end of that year.
That report is also a clean example of how entrepreneur statistics become press coverage. Carta published the data, and on March 21, 2025, Funds Society ran it under the headline “Startups Led by a Solo Founder Have More Than Doubled, But Are Less Successful in Raising Venture Capital.” The story worked because the data had a tension built in: more solo founders, less funding for them. Carta did not pitch a product. It pitched a finding with a twist, and the finding carried the brand’s name into a publication that would never have covered a feature announcement.
The FRAME Check: is this entrepreneur statistic safe to repeat?
The FRAME Check is the five-question test we use before any number goes into a founder’s byline, pitch, or press release. Each letter covers one way a statistic goes wrong in the wild, and a number has to clear all five before it earns a place next to your name.
F is for Freshness. Ask when the data was collected, not when the article quoting it was published. The BLS survival figures above were published in January 2024 but describe a cohort born in 2013. That is fine, as long as you say so. A stat with no visible collection year fails here by default.
R is for Reach. Ask what population the number covers. The SBA’s 36.2 million includes self-employed individuals with no staff. GEM’s 18.5% describes adults, not firms. Census applications describe filings, not businesses. Most misuse comes from quoting a number about one population as if it described another.
A is for Agency. Ask who counted, and whether they had a stake in the answer. A federal statistical agency, a foundation, an academic consortium, and a vendor survey can all produce valid numbers, but a reporter will weigh them differently. When a vendor stat and a federal stat say the same thing, cite the federal one.
M is for Measure. Ask what exactly was counted. “Applications” versus “formations,” “establishments” versus “firms,” “survival” versus “profitability.” The September 2026 Census release contains both 534,947 and 29,022, and they describe very different things. Name the measure in your sentence.
E is for Echo. Ask whether the number has already been repeated so often that it no longer carries news. “Small businesses are the backbone of the economy” is an echo. The fall in established ownership from 10.6% to 5.5% in GEM’s latest US report is not, because few people have quoted it yet. Echo does not make a stat wrong, but it does make it useless as a hook.
A stat that passes F, R, A, and M is safe to use. A stat that also passes E is a potential story. If you want to go further and produce a number that passes all five by design, our guide on how to create original research walks through the process.
What should founders do with these numbers?
Use them as context, and treat your own data as the headline. Muck Rack’s State of Journalism 2026 survey, as summarized by the University of Minnesota’s communications team on April 13, 2026, found that 40% of journalists want original data or research in pitches. A larger share, 70%, want clear relevance to their beat, and 58% want interview access to relevant sources. Meanwhile, 43% of journalists said it is rare to get pitches relevant to what they cover.
Those percentages describe the opening. Public benchmarks are free, sourced, and widely available, which makes them perfect as supporting evidence and weak as a hook. The founders who get quoted pair a public benchmark with a private one. A bookkeeping startup that knows how many of its customers reached year two can set that number against the BLS 79.6% first-year survival rate. A hiring platform can compare its own data with Census’s high-propensity applications. The public stat gives the reporter a frame they trust, and your number gives them something nobody else has.
The second use is in your own positioning. If you have survived five years, say so with the BLS cohort number beside it. If you are a solo founder who raised a venture round, the Carta finding that solo founders were 35% of 2024 startups but 17% of VC-backed ones turns your raise into a data point instead of a boast. For the mechanics of building a pitch around a number, see how to pitch a data story, and for what reporters say they need, our journalist statistics for 2026 collects the survey findings.
The third use is defensive. When you post entrepreneur statistics on LinkedIn or in a newsletter, every number becomes part of your public record, and AI answer engines increasingly quote those records back when someone asks about you. A misquoted stat attached to your name does more damage than leaving the number out. Run the FRAME Check before you post, and put the source and year in the same sentence as the figure, as every paragraph above does.
The short version: Americans filed 534,947 business applications in September 2026, but Census projects only about 29,000 per month will become employer businesses within a year. The SBA counts 36.2 million small businesses, most without employees. BLS data shows about 80% of new establishments survive year one and about a third last a decade. GEM finds 18.5% of US adults in early-stage entrepreneurship while established ownership has halved since 2019. Venture dollars hit records in 2026, with 86% of first-half money going to AI. Quote each with its source and year, check it against the FRAME test, and save your own data for the headline.