Independent audit · Y Combinator, 2005–2026 · data cut 6 November 2026
Beyond the Batch
Y Combinator still produces winners. Its 2020–21 batches produced them at a fifth of the rate of its 2015–17 batches, while the batches more than doubled in size.
We tested the common claims about YC, from believers and from critics, against 6,373 companies in YC's public directory and 1,447 companies on the CB Insights unicorn list. Every figure links to its source or to a calculation you can rerun. Where the data cannot settle a claim, this page says so.
Share of a batch year's companies valued at $1B or more within five years. Batches of 2015–17 against batches of 2020–21.
WP BΣCompanies per year in YC batches, 2019 against 2026. The 2026 figure spans four batches.
WP A✓YC companies among the 25 most valuable unicorns created since January 2023. The one is Kalshi.
WP C✓Consumer companies as a share of YC's batches in 2009, Airbnb's year, against 2026.
WP D✓Tick marks used on this page
The batch got bigger
YC funded 368 companies in 2019 and 725 in 2021. After a dip in 2023 it reached 830 in 2026, spread across four batches a year instead of two.✓
Companies per batch year
Hover a bar for the year's survival figures
2026 includes Winter, Spring, Summer and Fall 2026. Counts are companies listed in YC's directory today, so a few early companies that left the directory may be missing.
Size matters for two reasons. A fixed number of partners now reviews and coaches more than twice as many companies. And when a batch doubles, the number of outliers would have to double just to keep the rate flat. WP B tests whether it did.
Same age, same test
The usual mistake is to compare a 2016 batch, which has had ten years to grow, with a 2023 batch that has had three. We only compare cohorts at the same age.
- c is a batch year. N(c) is the number of companies from that year in YC's directory.
- Uw(c) is how many of them were first valued at $1B or more by the end of year c + w.
- A cohort only gets a value once its window has opened fully. The last year of each window is marked year-to-date.
Unicorn yield by batch year
Early years hold few companies (8 in 2005, 42 in 2009), so single years jump around. Read the era table.
| Batch years | Companies | $1B+ within 3 yrs | Rate | $1B+ within 5 yrs | Rate |
|---|
The five-year rate fell from 3.98% for 2015–17 batches to 2.48% for 2018–19 and 0.77% for 2020–21.Σ The three-year rate for 2022–23 batches is 0.09%: one company out of 1,125.
This did not happen because unicorns stopped being made. CB Insights records 123 new unicorns in 2025 and 181 so far in 2026, more than any year since 2022.✓ Recent YC cohorts are competing in a strong market.
Two effects that work against this result
The 2021–22 bubble. 61 of the 113 YC unicorns we counted first crossed $1B in 2021 or 2022. That flatters older cohorts whose windows included those years. The 2020 and 2021 cohorts' windows include them too.
Coverage. CB Insights lists companies that are private today. Companies marked down below $1B drop off the list, so for private companies we count "still valued at $1B+", not "ever touched $1B". The same rule applies to every cohort.
YC's share of new unicorns
Both sides of this ratio come from one dataset, so neither side is cherry-picked. YC companies made up 7.9% of new unicorns in 2021–22 and 4.5% since January 2023.Σ
YC-backed share of companies that became unicorns each year
2021 onward
Earlier years are left out because companies that later went public or were acquired leave the list, which would understate those years.
What critics leave out
YC is strong at the very top. 11 of the 91 private companies valued at $10B or more are YC companies (12.1%), double YC's 6.1% share of all unicorns.Σ
Seven of the eleven come from batches before 2019 (Stripe, Gusto, Helion, Scale, Rippling, Faire, OpenSea). Four come from 2019–20: Kalshi, Deel, Whatnot and Supabase.
What believers leave out
Of the 25 most valuable unicorns created since January 2023, one is a YC company. Most of the rest are frontier labs, AI infrastructure and defense companies.✓
| # | Company | Valuation | Became a unicorn | YC |
|---|
What a batch is made of
In 2009, half of YC's companies were consumer companies. In 2026 it is 5.5%. Business software and AI now dominate: 63–68% of 2023–25 companies were B2B, and 67% of 2024 companies carry an AI tag.✓
Share of each batch year, 2009–2026
B2B and Consumer are YC's own primary-industry labels and exclude each other. The AI tag cuts across industries.
Would Airbnb get in today?
No dataset can test that counterfactual, and we will not claim it either way. Two things are measurable. The category Airbnb came from has shrunk from half of a batch to one company in twenty. Airbnb itself was not a pedigree-free case: two RISD design graduates and a Harvard engineer. Paul Graham funded them although he disliked the idea.
Why the mix matters
Business software sold to other startups can show revenue within the 12 weeks of a batch. Consumer networks, hardware and frontier research usually cannot. A batch selected on short-term progress will drift toward the first kind, whatever the partners intend.
Who gets in
The thesis this page was built to test is "YC swapped track record for school brand." The founder data so far points that way. It comes from two outside firms, and we have not yet recomputed it ourselves.
| Measure | Earlier | Recent | Source |
|---|---|---|---|
| Founders from a top-20 university | Under half | Over half, for the first time† | Rebel Fund |
| Founders who worked at a top-20 employer | About 25%, a decade ago | Nearly half† | Rebel Fund |
| Average founder age | About 29 (2015–2023) | About 26 and falling† | Rebel Fund |
| Years of work experience before the batch | About 9 (historical) | 5.8 (W26); mean 6.5, median 5 (S25)† | Extruct W26, S25 |
Put together: founders are younger, have less work behind them, and more often carry an elite school or employer name. When there is less track record to judge, a school name and a crisp ten-minute interview carry more weight.
The awkward counterpoint. Across all venture capital, pedigree correlates with outcomes. That may be self-fulfilling: pedigreed founders get funded, so pedigreed founders win, so the data seems to prove pedigree. WP B is the check on it. If credential-heavy recent cohorts were better picks, their equal-age yield would be rising. It is falling.
Next version
v2 will compute school, employer, age and experience for every founder from public founder bios. It will publish the rows so anyone can recount them. Until then these four figures carry the † mark.
Where the frontier went
The largest AI companies of this wave were built outside YC. That is a fact. "YC missed them" is only partly true, because several were never in any accelerator's applicant pool.
| Company | YC | Founded | Valuation | Basis | Revenue | Basis |
|---|
A model mismatch, not a scouting error
Frontier labs needed billions before they had revenue, and their founders came out of research labs that large funds already knew. YC's standard deal is $500,000: $125,000 for a fixed 7% plus a $375,000 uncapped SAFE.✓ That deal is built for software that ships in weeks. That is a limit of the model more than a failure of judgment.
Not in the pool
DeepSeek is a hedge fund's in-house lab. xAI is Elon Musk's. Safe Superintelligence is Ilya Sutskever's. OpenAI was co-founded in 2015 with YC's president at the time. Counting these as YC misses inflates the critique, so we don't.
Paper and real
A valuation is a price someone paid for a small slice of a company. Revenue is money that came in. Choose the multiple you think a mature software business deserves, and see which marks hold. YC and non-YC companies are listed together because this problem is not specific to YC.
| Company | YC | Valuation | Revenue | Paid multiple | Implied at M | Gap |
|---|
Revenue figures mix company statements and analyst estimates. The basis is shown on each row in WP F. Dates of valuation and revenue can differ by a few months.
| Mark-downs on the record | Peak | Later | Source |
|---|---|---|---|
| Brex (YC W17) | $12.3B (2022) | Sold to Capital One for $5.15B (Jan 2026), 58% lower | Fintech Futures✓ |
| Zepto (YC W21) | $7.0B (Oct 2025) | Pre-IPO round at $4.5B, listing delayed (Jul 2026) | YourStory✓ |
| Embark Trucks (YC W16) | ~$5B SPAC (2021) | Wound down 2023 | Public filings |
| Cruise (YC W14) | $30B (2021) | GM ended the robotaxi program (Dec 2024) | GM announcements |
Dogmas, audited
Eight claims you hear in San Francisco, tested against the working papers above. Some come from believers, some from critics. Both sides lose a few.
The ledger
These are all the YC companies we counted as reaching $1B, with the year each first got there. If a company is missing or misdated, the corrections log below will record the fix.
| Company | Batch | Crossed $1B | Years | Route | Status | Source / note |
|---|
Method, limits and what we threw out
Batch counts
From YC's public directory as mirrored by the open-source yc-oss project, fetched 6 Nov 2026. Batch year is the calendar year in the batch name.
Unicorn matching
CB Insights names were matched to YC directory names, then filtered by city and date, then reviewed by hand. Name collisions were removed: Ramp, Cohere, Circle and Runway ML are not YC companies, although the directory lists others with those names.
Exits
Companies that went public, were acquired or shut down leave the CB Insights list. We added 25 by hand, using the year each first reached $1B. Where that year was unclear, we used the later, conservative year (EquipmentShare: IPO year).
Pivots
A company counts for the batch YC's directory gives it. Jasper is listed under its founders' 2018 batch (as Proof). Ramp is not counted, although its founders went through YC with Paribus.
What this cannot show
Revenue for most private companies is not public, and nothing here measures founder wealth, returns to YC's fund, or companies that are good businesses below $1B.
Independence
Published by YokaiOS. Not affiliated with, sponsored by or endorsed by Y Combinator. Company names identify companies and imply no endorsement.
Claims we tested and removed before publishing✗
An earlier draft of this page made these claims. None survived checking.
"0 of 16 frontier decacorns came from YC"The list was assembled by leaving YC companies out. It also counted OpenAI, co-founded with YC's then-president, and labs that were never in any accelerator's pool."A 119.8× Fundraising Privilege Index"All six inputs (close rates, weeks to close, caps) were unsourced, and multiplying unrelated ratios produces an arbitrary number."$1 of batchmate revenue creates $345 of valuation"This was a model with a constant tuned to hit its own answer. No public data measures how much YC revenue comes from batchmates."Series A graduation collapsed because of YC"Carta shows the same fall for all US seed companies: 17.6% of 2021 seed rounds and 13% of H1 2022 rounds reached Series A."Real unicorn yield 0.11% vs 4.7%"It compared batches that were 2–4 years old with batches that were 9–14 years old. WP B redoes it at equal age."Jasper (W21)", "DeepSeek $91B"Jasper's founders did YC in 2018 with Proof. DeepSeek's first outside round implies roughly $50–59B.
Corrections log. 6 Nov 2026: v1 published.