.com fell under half of YC startups in Summer 2022. .ai hit 37% in Summer 2025

We read the Y Combinator public directory, 6,190 companies, 6,189 of them with a batch label. 6,127 carry a website domain, 6,008 a one-line pitch and 5,758 a long description. We counted two things per batch: which top-level domain the company registered, and which words it uses to describe itself. The series runs Winter 2011 through Summer 2026, 5,962 companies.

We expected the .com majority to have ended in 2024 or 2025. It ended in Summer 2022. We also did not expect .ai to be the second most common suffix across the whole fifteen-year file.

Key findings

1. .com first falls under half in Summer 2022: 47.0%, or 109 of the 232 companies with a domain. The batch before it, Winter 2022, was 55.9%.
2. Winter 2011 was 95.5% .com, 42 of 44 companies with a domain. Every batch from 2005 to 2010 runs between 89% and 100% .com.
3. No batch after Winter 2022 has been above 52.2% .com. In the last ten batches, eight are under half.
4. .ai peaks at 37.0% in Summer 2025, 61 of 165 companies with a domain, and is 26.7% in Summer 2026, 63 of 236.
5. Across the whole series, 5,922 companies with a domain, .com holds 59.9% and .ai holds 14.6%, which makes .ai the second most common suffix on the file.
6. "Shopify for" has 19 companies all-time and none since 2023. "Cursor for" has 10, five of them in 2025.
7. "agent" in the software sense reaches 26.7% of Spring 2026 one-liners, 52 of
8. Before Winter 2022 it never passed 4.0% in any batch.
9. "marketplace" peaks in the first batch of the series, Winter 2011, at 11.4%, and is 0.4% in Summer 2026. "on-demand" is zero in every batch from Winter 2023 on.
10. AI, machine learning or artificial intelligence appears in 52.1% of Fall 2025 one-liners, 75 of 144, against 6.1% in Summer 2015, 6 of 98. Summer 2026 is 38.1%, 90 of 236, the third batch in a row below the Fall 2025 high.

Summer 2022 was the first YC batch where .com fell under half, at 47.0%

109 of the 232 Summer 2022 companies with a domain use .com, 47.0%. Source: FundingWatcher, September 2026. Winter 2022, the batch immediately before, was 55.9%, so the drop is one batch wide and it has not reversed: no batch since has been above 52.2%. Winter 2024 at 52.2% and Winter 2025 at exactly 50.0% are the only two of the last ten batches at or above half.

.ai went from nothing to the second suffix on the file. It is zero in every batch before Winter 2014, first passes 10% in Winter 2020 at 10.6%, and peaks at 37.0% in Summer 2025, which is 61 of the 165 companies with a domain. Summer 2026 is 26.7%, 63 companies.

.io and .co gave up share over the same stretch. .io peaked at 10.8% in Summer 2022 and is 3.4% in Summer 2026. .co peaked at 8.2% in Winter 2018 and is 3.0% in Summer 2026.

Across the full series, Winter 2011 through Summer 2026, 5,922 companies have a domain: 3,546 on .com, 867 on .ai, 340 on .io, 219 on .co, and 950 on everything else. After those four, the largest suffixes are .dev with 113 companies and .app with 92.

TLD is the public suffix of the domain, so .co.uk and .com.au count as themselves and land in "other". Shares are suppressed for batches under 30 companies, and counts under 5 print as <5.

We left Summer 2005 through Summer 2010 out of the series. Those eleven batches hold 207 companies between them, and most run under 30 companies each, which is our suppression floor, so they are reported as counts only. None of the 207 is on .ai. The largest of them is Summer 2010 with 36 companies, 34 of them on .com; the smallest, Winter 2006, holds 7 companies, all on .com.

"Shopify for" has 19 companies and none since 2023; "Cursor for" has 10, five of them in 2025

We searched the one-liner plus the long description for <Brand> for across 41 brands, counting distinct companies rather than sentences, so a company that writes the phrase twice counts once. 164 raw sentence matches came back. We read all of them and threw out 30, leaving 134 sentences from 129 companies.

"Shopify for" is the most used analogy on the file with 19 companies, and its last user was funded in 2023. By year: 1 in 2017, 1 in 2018, 1 in 2019, 4 in 2020, 5 in 2021, 4 in 2022, 3 in 2023, then nothing. "Cursor for" has 10 companies and five of them are 2025 batches.

"Stripe for" with 11 companies and "Plaid for" with 11 are the fintech-era equivalents, and each has one user after 2023: Surge in Fall 2024 for "Stripe for", Hubble in Summer 2026 for "Plaid for" (medical records). "Lovable for" has 8 companies, 5 of them in 2025 and 2026. "Uber for" has 7 companies across fifteen years and its last user, Kamion, was funded in Summer 2022. "Rippling for" has 6, spread evenly from 2021 to 2025.

Fifteen more brands have two users or one: Snowflake, Amazon, Zapier, Square, Gusto, Slack, Coinbase, Waymo, Tinder, Tesla, Notion, Carta, Vercel, Salesforce and Toast.

The 19 "Shopify for" companies, by batch: OneFin (Winter 2017), Easol (Summer 2018), Namaste Business (Summer 2019), Explorex (Summer 2020), Hubla (Summer 2020), Sakneen (Winter 2020), Tough Commerce (Winter 2020), AgendaPro (Winter 2021), Preki (Winter 2021), Topline Pro (Winter 2021), Wabi (Summer 2021), ZOKO (Winter 2021), Blink (Winter 2022), Ekho (Summer 2022), Pulppo (Summer 2022), Take App (Winter 2022), Foundation (Summer 2023), Pair AI (Winter 2023) and Sanvivo (Winter 2023). The jobs they name are restaurants, real estate, construction, beauty and wellness, creators, WhatsApp merchants, vehicles, homebuilders and pharmacies.

The 10 "Cursor for" companies: Mito (Summer 2020), Glade (Summer 2023), Minded (Summer 2023), Alex (Fall 2024), Raycaster (Fall 2024), Dex (Winter 2025), Onlook (Winter 2025), SigmanticAI (Summer 2025), Slashy (Summer 2025) and Vesence (Spring 2025). They name Xcode, games, chip design, email, lawyers, designers, regulated life-sciences documents and knowledge work. Mito is filed under Summer 2020 with its current wording, which is the limitation described in Method.

"Copilot for" is not in the table. We removed all 21 matching sentences by hand, and that is the one call that changes the ranking. In this corpus copilot is a product-category noun, as in "an AI copilot for radiologists", not a reference to GitHub Copilot: the matches start in 2019, two years before GitHub Copilot shipped. The 21 sentences come from 19 companies, so left in, "Copilot for" would have tied "Shopify for" at the top of the leaderboard at 19 companies. The word is counted as a word family in the next section instead, and every removal is listed with its reason in analogy_hits_removed.csv so the leaderboard can be rebuilt without our judgment.

"agent" in the software sense reaches 26.7% of Spring 2026 one-liners, 52 of 195

This section counts word families in the one-line pitch only, so the denominator is companies in that batch with a one-liner.

"agent" needs an exclusion, because the word also means a person. We cut human-agent spans such as insurance agent, real-estate agent and call-centre agent before the test, unless a software modifier (AI, LLM, autonomous, agentic, voice, bot and similar) sits inside the span, in the 40 characters before it, or in the company's own name. Across the whole in-scope set that exclusion removes exactly two one-liners: GroMo in Winter 2021 and Pulppo in Summer 2022. A naive exclusion without the software-modifier guard removed 19, including Yuma AI's "The AI Support Agent for Ecommerce", which is the sense we want to count.

The family is zero in 12 of the 18 batches from Winter 2011 to Summer 2019 and never passes 4.0% in any batch before Winter 2022. It is 5.3% in Winter 2022, 16.0% in Summer 2023, and peaks at 26.7% in Spring 2026, 52 of 195 one-liners. Summer 2026 is 19.5%. All-time the family covers 490 companies. "AI agent" specifically peaks at 11.8% in Fall 2025, 17 of 144, and "agentic" peaks at 6.3% in Spring 2025.

"AI-native" is the family closest to its high in the last batch: zero before Winter 2020, 9.7% in Spring 2026, 9.3% in Summer 2026.

"copilot" peaks by batch at 2.1% in Fall 2024. By calendar year it peaks in 2023 at 1.6%, with 2024 close behind at 1.4%, then 1.0% in 2025 and 0.2% in 2026. 33 companies use it all-time.

"marketplace" peaks in the first batch of the series, Winter 2011, at 11.4%, and is 0.4% in Summer 2026. "on-demand" peaks at 3.2% in Winter 2012, last appears in Summer 2022, and is zero in every batch from Winter 2023 onward. Crypto, web3, blockchain, NFT and DeFi together peak at 3.5% in Winter 2018, are 2.6% in Summer 2022, and are 0.5% or zero in every batch after Winter 2023. "SaaS" peaks at 3.7% in Winter 2015 and is zero in Summer 2026.

Every batch is in families_by_batch.csv, along with the AI agent, agentic, open source, SaaS and no-code columns.

52.1% of Fall 2025 one-liners mention AI, against 6.1% in Summer 2015

We matched AI, A.I., artificial intelligence and machine learning against the one-liner, after stripping the company's own name from the string, so a company called Foo AI is only counted if its pitch says AI somewhere else. That step never moves a batch by more than 2 companies.

Fall 2025 is the peak and the only batch above half: 75 of 144 one-liners, 52.1%. Summer 2015 is 6.1%, 6 of 98. Winter 2022 is 21.1% and Winter 2023 is 39.0%, a jump of nearly 18 points across two batches, and no batch since has fallen below 38.1%.

The last three batches are below the Fall 2025 high: 43.7% in Winter 2026, 44.1% in Spring 2026, 38.1% in Summer 2026. Summer 2026 may still be filling, so we are not calling that a turn.

All 36 batches are in ai_by_batch.csv.

The funnel runs from 6,190 directory rows to 6,169 in scope and 5,962 in the series

The all-time TLD split in the first section covers the 5,922 companies with a domain in the Winter 2011 to Summer 2026 batches. The 2005 to 2010 companies are counted but are not part of any share in this report.

The directory holds 166 companies in Summer 2025 against Extruct AI's count of 160

The largest limitation applies to every number in this report. The YC directory shows each company's current text and current domain, filed under the batch it was founded in. A company funded in 2020 that rewrote its pitch in 2026 puts 2026 wording in the 2020 row, which is why "Cursor for data analysis" shows up against a Summer 2020 batch. Domains behave the same way: a company that moved from .com to .ai counts as .ai in its founding batch, which if anything makes the .ai shift look older and gentler than it was. The directory carries no text or domain history, so we could not correct for it. Read the by-year series as "companies from batch X that describe themselves this way today".

For batch sizes we checked one batch against an outside count. Our file holds 166 companies in Summer 2025. Extruct AI's published analysis of that batch, https://www.extruct.ai/research/ycs25/ from 25 September 2025, describes it as 160 startups and 231 founders. The gap is 6 companies, in the direction you would expect from a directory that keeps adding entries after demo day. We could not find a first-party YC page that states a batch size as a number: the directory at ycombinator.com/companies renders its counts in JavaScript and publishes no total in the HTML.

Method

The source is the Y Combinator public company directory as held in FundingWatcher's copy: name, batch, website domain, one-line pitch, long description. 6,190 rows, the full directory, no sample. One script produced every number in this report and wrote the published files in the same run. tldextract supplied the public suffix from its bundled snapshot, so no lookup left the machine.

The series stops at Summer 2026. Fall 2026 holds 19 companies and Winter 2027 holds 1, so both are obviously still filling and are excluded. Summer 2026 is included because it is not obviously incomplete, but its numbers may still move.

Three denominators are used and never mixed. TLD shares divide by companies in the batch with a non-empty domain. Word-family and AI shares divide by companies in the batch with a one-liner. Analogy phrases are searched across one-liner plus long description, for the 6,114 in-scope companies that have either.

Shares are suppressed for any batch under 30 companies, and counts under 5 print as <5 to avoid identifying individual companies in a small batch. The share next to a suppressed count is still exact.

Analogy matching is the regex \b(<Brand>) for\b, case-insensitive, for 41 brands listed in METHOD.md. The leaderboard counts distinct companies. Every raw match was read by hand. The 30 removals are 21 "Copilot for" category-noun matches, 3 sentences describing the named brand's own business rather than an analogy, 2 employment histories ("led product initiatives at Amazon for over 7 years"), 1 customer list ("companies like Google, Yahoo, Linkedin, Twitter, and Airbnb for years"), 1 product name, 1 verb object and 1 common noun. Each is listed with its reason in analogy_hits_removed.csv. The removals are stored in a file that is re-applied on every rerun, so the script reproduces these numbers exactly.

The word families are matched on the one-liner only. The regexes, including the human-agent exclusion and its software-modifier guard, are in METHOD.md.

Two things we did not check. No company's batch label was verified against a funding record, and the brand list for analogies is fixed at 41, so a brand outside it could rank. Databricks, Deel, Twilio, Replit, Zillow, Peloton and OpenAI were searched and scored zero.

Data

Everything in this report is in the files below, under CC BY 4.0.

• tld_by_batch.csv, one row per batch with the .com, .ai, .io, .co and other counts and shares
• pre2011_by_batch.csv, the eleven Summer 2005 to Summer 2010 batches, counts only
• tld_other_suffixes.csv, every suffix outside .com, .ai, .io and .co with its company count
• families_by_batch.csv, one row per batch with all eleven word families
• ai_by_batch.csv, AI mentions per batch, with the count before the company-name exclusion
• agent_exclusions.csv, the two one-liners the human-agent exclusion removes, and naive_exclusions.csv, the 19 an exclusion without the software-modifier guard would have removed
• phrases_by_batch.csv, one row per analogy phrase and batch, companies and sentences
• analogy_hits.csv, every kept match with the company, its batch and the matched sentence
• analogy_hits_removed.csv, every hand-removed match with its reason
• README.md and METHOD.md