Anthropic's S-1 Is Public. Which of Its Numbers Belong in Your YC W27 Application?
Anthropic's S-1 hit #1 on HN six days before the YC W27 deadline. Here is how to borrow its traction metrics (and avoid its traps) in your application.

Anthropic S-1 Numbers for Your YC W27 Application
YC Roaster
Anthropic's preliminary S-1 is public and the Hacker News thread hit #1 with more than 1,500 points. The filing (ticker: ANT, Nasdaq) lands six days before the YC W27 application deadline on November 2 at 8pm PT. That timing matters for anyone finishing an application this week, because the partners reading your form will have just read the same document, and it quietly resets what "traction" looks like on paper.
This is not another "should you build on Claude" post; we covered that when the IPO was first reported. This one is about what the S-1 teaches about presenting numbers, and which of its numbers you should and should not import into your own application.
What does the Anthropic S-1 actually say?
The headline figures, in millions, from the preliminary prospectus:
- Revenue: $386 (2024), $4,585 (2025), $16,286 for the six months ended June 30, 2026, against $1,163 for the same period in 2025.
- Net loss: $(8,309) in 2024, $(41,965) in 2025, $(80,748) for the first half of 2026.
- Net revenue retention of approximately 900% as of September 30, 2026.
- More than 11,900 customers with $100,000 or more in trailing-twelve-month revenue, more than 1,900 at $1 million or more, and approximately 250 at $10 million or more.
That is the whole traction story in four lines. Notice what is missing: no user count, no "downloads," no "waitlist." Every number is dollars, or customers bucketed by dollars.
How should a YC W27 applicant present traction?
The YC application asks how many users you have, how much revenue, and your growth rate. Most applicants answer with the biggest number they can find. Anthropic's filing shows the opposite instinct: pick the number that is hardest to fake, even if it is small.
Three things to steal directly:
Bucket customers by what they pay, not by how many signed up. If you have 40 paying customers, say how many pay more than $1,000 a year, how many pay more than $10,000, and who the top one is. "11,900 customers above $100K" tells an investor more than "two million developers" ever could. Your version might be "14 customers above $5K ARR, 3 above $20K."
State retention as a ratio, with a date. "Approximately 900% net revenue retention as of September 30, 2026" is a specific claim with a specific as-of date. In your application, "net revenue retention 130% over the last 6 months (cohort of 22 customers who paid in April)" beats "strong retention." If you only have three months of data, say three months. The precision is the signal.
Put the comparison period next to the number. Anthropic does not just say $16.3B for H1 2026, it puts $1.16B for H1 2025 beside it. Whenever you give a revenue figure in a YC application, give the same figure from the prior period on the same line. Make the growth rate trivial to compute.
Which Anthropic numbers should you NOT copy?
This is where applicants get hurt.
Do not cite the S-1 as your market size. By the deadline there will be W27 applications saying "Anthropic did $16B in six months, so the market for AI agents is enormous." That is the frontier lab's revenue, not your TAM, and the Oct 21 post on this blog already covered why laundered market-size numbers get an application tossed. If you build on Claude, Anthropic's revenue is your cost of goods sold, not your opportunity.
Do not mistake 900% NRR for a benchmark. That figure is what happens when existing customers expand spend roughly ninefold in a year during a once-in-a-decade platform shift. It is not a target for a 6-month-old startup. If you report 140% NRR honestly, you are doing well. Reporting 300% because one of your two customers doubled is a number a partner will unpick in 30 seconds.
Do not pretend the loss line does not exist. An $80.7B six-month net loss is right there in the filing's summary financials. Anthropic does not hide it; it explains the compute and infrastructure spend ($8,945 million for the half) behind it. If your startup burned $90K to get $12K of revenue, say so and explain what you bought with it. YC has funded that ratio many times. What they do not fund is the applicant who leaves the burn number blank.
What does the S-1 tell YC about startups built on Claude?
Partners will read your application with two sentences from the risk factors fresh in mind.
The first: Anthropic says it has "changed, and may in the future change, usage limits, rate limits, or model and product capabilities." If your product's margin depends on a rate limit or a pricing tier staying where it is today, the partner knows that is not in your control, and so does Anthropic's own lawyer.
The second concerns cloud providers: "Certain CSPs from which we obtain substantial compute also develop and offer their own AI models," and those relationships "may not always be fully aligned with our long-term interests." Translate that down one level. If Anthropic's suppliers are also its competitors, then your supplier (Anthropic) is also plausibly your competitor, and the "why won't the lab build this" question from the Oct 24 post on this blog applies with the S-1 as exhibit A.
Neither of these is a reason not to apply with a Claude-based product. The W27 batch will almost certainly include plenty of them. It is a reason to have one sentence in your application that shows you have read the risk section too: what you do if the price doubles, and what you own that Anthropic does not.
Does an $80B loss change what YC funds this batch?
Probably not in the direction the HN thread is arguing about. YC partners are not public market investors. The deal is still $500K ($125K for 7% plus $375K on an uncapped MFN SAFE), interviews still run in November and December, and decisions still go out by December 11. What the filing changes is the baseline literacy in the room. A partner who just read the prospectus will be less impressed by a vanity metric and more impressed by a founder who can talk about retention cohorts and customer tiers without being asked.
Good news for a small company with real numbers. Bad news for a polished deck wrapped around a waitlist.
What should you fix in the next six days?
You have until Monday, November 2 at 8pm PT. In priority order: rewrite your revenue answer so the prior period sits next to the current one; replace every user count with a paying-customer count bucketed by spend; add an as-of date to every retention or growth claim; add one sentence about what happens to your product if your model provider changes limits or pricing; and remove any line that uses Anthropic's revenue as evidence of your market.
Then have someone who has actually been through a YC batch read it before you submit. YC Roaster matches applicants with YC alumni who will tell you which of your numbers a partner will believe and which they will skip. Six days is enough time for one round of honest feedback.
Ready to get your YC application roasted?
Get free AI feedback + a review from a YC alumni.
Submit Your Application