The YC Mortality Table

Key findings

• As of September 2, 2026, 33.9% of the 224 Y Combinator companies in the 2016 batch are no longer operating, and 21.4% have been acquired or gone public.
• As of September 2, 2026, 17.6% of the 727 Y Combinator companies in the 2021 batch are no longer operating, against 33.9% of the 2016 batch.
• 78.8% of tagged Y Combinator companies in the 2024 batch carry an AI tag, up from 20.3% of the 2021 batch, measured on September 2, 2026.
• The United States accounts for 87.7% of Y Combinator's 2026 batch, up from a low of 55.0% in the 2021 batch, according to YC's own company listings as of September 2, 2026.
• Consumer startups fell from 26.3% of Y Combinator's 2014 batch to 5.6% of its 2026 batch, while Industrials reached an all-time high of 17.4% in 2026, as of September 2, 2026.
• The median Y Combinator company in the 2026 batch reports 2 employees, against a median of 15 for the 2018 batch, measured on September 2, 2026.
• No inactive and no acquired Y Combinator company is flagged as hiring, against 34.1% of the 4,295 active ones, as of September 2, 2026.
• 269 people, 2.4% of the 11,230 founder profiles in this dataset, have founded more than one Y Combinator company as of September 2, 2026.
• In all 3 batch-year bands between 2011-2015 and 2020-2022, Y Combinator companies listing three or more founders were acquired or went public more often than two-founder companies, which in turn exited more often than solo-founded ones, as of September 2, 2026.

01 How does a batch look, years later?

A YC cohort loses a third of itself by year ten, and about a fifth of it exits.

Each bar is one batch year, as it stands today. The pattern is monotone and boring in the best way: the older the cohort, the less of it is still operating. The 2016 batch is 33.9% inactive. The 2021 batch, five years on, is 17.6% inactive. The 2025 batch has barely begun to register losses at 1.4%.

Caveat: Not a cohort study

This is a cross-section, not a longitudinal one. There is no history in the data, only today's status, so the downward slope mixes age with era. It cannot tell you whether the 2011 batch looked like the 2021batch does now, when it was five years old. Nothing here supports “YC is getting better” or “worse”.

02 What is a batch made of now?

AI went from a fifth of the batch to four fifths in three years.

Share of tagged companies carrying an AI tag: 20.3% in 2021, 32.1% in 2022, 62.4% in 2023, 78.8% in 2024. This is the sharpest move anywhere in the dataset, and it is not close.

Caveat: Watch the denominator

Tag coverage collapses on recent batches: 97.9% of the 2017 cohort is tagged, but only 76.2% of 2025 and 54.1% of 2026. Divide by the whole cohort instead of by tagged companies and you manufacture a fake AI decline after 2024. The 2026 point rests on 357 tagged companies out of 660 and should be read as provisional, not as a peak passing.

03 Where do the companies come from?

YC internationalised, then reversed it almost exactly.

The US share of a batch fell from 85.9% in 2012 to 55.0% in 2021, then climbed back to 87.7% in 2026. India ran 8.4% of the 2020 batch and is 0.8% of 2026. Mexico went 4.3% to zero. Whether that is fewer non-US companies or the same companies moving to the US is exactly what this field cannot tell you.

Caveat: This measures offices, not founders

The region field is a company's currentlocation, so a Mexican founding team operating out of New York counts as United States here, and always did. Mexico's fall to zero is therefore two different effects added together: fewer LatAm companies admitted, and admitted companies relocating to the US. This data cannot separate them, and it is not a headcount of founders by nationality. Read the chart as where YC companies are registered, nothing more.

04 And what do they build?

Consumer collapsed from a quarter of the batch to one in 18.

Consumer was 26.3% of the 2014 batch, the last one in which it was a quarter or more, and 5.6% of 2026. B2B absorbed most of it, peaking at 68.0% in 2023. Fintech had its own arc, cresting at 21.0% in 2022 and falling to 5.8% by 2025. The newest move is Industrials, which at 17.4% of the 2026 batch is the highest it has ever been.

Caveat: YC's taxonomy, not a neutral one

These are YC's labels, applied by YC, and “B2B” is broad enough to swallow a great deal. A company that would have been labelled Consumer in 2014 may be labelled B2B today without changing what it does. Treat the direction as solid and the exact levels as soft.

05 How big are they?

The median company in the 2026 batch has 2 people.

Median headcount by batch: 15 in 2018, 10 in 2021, 5 in 2023, 3 in 2025, 2 in 2026. This is the statistic everyone wants to read as the small-team AI thesis. Read the caveat before you do.

Caveat: This is mostly age

Headcount is measured today, and the 2026 batch is the youngest in the dataset: 99.5% of it is still marked active because almost none of it has had time to be anything else. A company that launched a few months ago has 2 people because it launched a few months ago. To make the small-team claim properly you would need headcount at a fixed age across eras, and this data has no history, so that comparison cannot be built from it. The honest reading is: younger cohorts are smaller, which is what younger cohorts always are.

06 Do more founders help?

The solo-founder penalty vanishes when you control for batch year. The exit gradient does not.

Compare naively and solo-founded companies look worse: a larger share of inactive companies list a single founder than of active ones. Split by era and the failure gap disappears. What survives the control is a different thing: in all 3 eras, companies listing more founders exit more often.

The failure rates cross and re-cross. The exit rates do not: 3+ beats 2 beats solo in every band, without exception.

Caveat: “Founders listed” is a page property

This counts LinkedIn links on a YC page, not the true founding team. The share of companies listing exactly one founder falls from 33.3% in 2007 to 15.8% in 2025, which is far more likely to be older pages carrying fewer links than a real shift to larger founding teams. That same bias plausibly inflates the exit gradient: a company that got acquired may simply have a better-maintained page. Suggestive, not established.

07 Which of them are actually alive?

The hiring flag is the cleanest liveness signal in the dataset.

Zero inactive companies are hiring. Zero acquired companies are hiring. 34.1% of active companies are, and 39.1% of the 23 public ones. If you want the subset of YC that is doing something right now rather than merely not yet marked dead, that is the filter.

Caveat: The newest batch has not started

Hiring peaks at 40.9% for the 2025 batch and drops to 20.6% for 2026, which is a cohort that has not begun recruiting, not a cooling market. Also note the flag is mechanically zero for inactive and acquired companies, so it measures “operating and recruiting”, not recruiting intensity.

08 And the founders themselves?

269 people did YC more than once. It is 2.4% of everyone.

256 of them appear at 2, 12 at 3, 1 at 4, counting YC companies per person. Their outcomes are close to everyone else's: 71.7% at an active company against 71.9% overall, and 12.6% inactive against 15.5%. The one visible edge is exits, 15.6% against 12.2%. Whatever the repeat-founder advantage is, at this resolution it is small.

Caveat: Neither gender nor school is in here

LinkedIn has no gender field, so any “% female founders” figure is name-based inference, and these names were themselves reconstructed from URL slugs. That is a guess stacked on a guess and it is not in this report. Education lives only on the profile page, which would mean paid per-profile scraping for 11,230 people. Also not here.

The founder profiles behind this section are searchable one by one in the Rocketlist YC founder directory, which is built from this same snapshot.

Methodology

• Company data, all 6,201: yc-oss.github.io/api/companies/all.json, a free community index of YC's own listings. Status, batch, industry, headcount, regions, tags and the hiring flag all come from there.
• Founder data, 11,230 founders across 11,513 founder-company roles, over 5,974 companies (96.3%): the LinkedIn links YC renders in plain HTML on each ycombinator.com/companies/<slug> page. No auth, no paid API. The other 227 companies list nobody. The two founder figures are different units and both are reported: a person who did YC twice is one founder and two founder-company roles, so per-person questions use 11,230 and per-company questions use 11,513.
• Company-level questions use the full 6,201 so founder coverage introduces no bias. Founder-level questions are reported against their real denominator.
• Every figure on this page is read from a frozen snapshot captured on September 2, 2026, not fetched live, so the page and the numbers in it cannot disagree.

Limitations: what this cannot answer

• No history. Every field is today's value. There is no way to ask what a cohort looked like at a past date, which rules out any fixed-age comparison across eras.
• No demographics. No gender, no school, no prior employer. Those need the profiles, which means paying.
• Roles are not distinguished. A YC page's LinkedIn links are usually founders, but an early employee or advisor on the same page is captured too.
• YC's “top company” flag is unusable as a time series. It reads a few percent for the early 2010s and zero for every recent batch, which is designation lag, not a finding, so it is not charted here.