BusinessMarkets, money & strategy
Anthropic tells investors Q3 revenue rose to $16.8bn and says it cut back distillation by rival labs ahead of IPO
Anthropic told prospective investors that third-quarter revenue reached $16.8bn, roughly 45% above the prior quarter, with an 11% adjusted operating margin, and that distillation by rival labs, once 5–7% of revenue, was materially cut back ahead of a planned IPO.
From the Simon Willison article of October 15, 2026 and cited sources: Anthropic presented third-quarter revenue of $16.8bn at an October investor day in San Francisco, according to people briefed. [5] [6] [7] [3]
SOURCES[5] Bloomberg: Anthropic Revenue Hits $16.8 Billion bloomberg.com[6] The Next Web: Q3 revenue $16.8bn thenextweb.com[7] CNBC: Anthropic's third-quarter revenue cnbc.com
AT Pulse research deskPublished · Updated · 5 min read · AI-assisted analysisAI-assisted, not human-edited. Methodology Corrections Who owns this site
- What happened
- Anthropic presented third-quarter revenue of $16.8bn at an October investor day in San Francisco, according to people briefed. [5] [6] [7] [3] [4]
- Original source
- Simon Willison article, independent perspective
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- October 15, 2026 article published
- Covered
- Oct 16 edition
- Evidence
- 11 sources across 10 sites
- 1Simon Willison: Anthropic tells investors quarterly…simonwillison.net
- 2Anthropic: Detecting and preventing distillation attacksanthropic.com
- 3Verdict: Anthropic SEC draft filingverdict.co.uk
- 4The Next Web: Q2 revenue $11.5bnthenextweb.com
- 5Bloomberg: Anthropic Revenue Hits $16.8 Billionbloomberg.com
- 6The Next Web: Q3 revenue $16.8bnthenextweb.com
Reach: not measured yet
Anthropic presented third-quarter revenue of $16.8bn at an October investor day in San Francisco, according to people briefed. That is about 45% above the roughly $11.5bn reported for Q2. The figures arrive months after the company confidentially filed a draft S-1 for a planned Nasdaq listing. [5] [6] [7] [3] [4]
Anthropic described itself as profitable on an adjusted basis, with an 11% operating margin that excludes stock-based compensation. It also reported about $1.70 of revenue for every $1 of compute. Coverage notes the company now leads with quarterly figures rather than annualized run-rate claims. [1] [6]
Rival labs paying for heavy Claude use to train their own models made up 5–7% of revenue, and Anthropic said it reduced this materially in Q3. In February, Anthropic had accused DeepSeek, Moonshot AI and MiniMax of extraction campaigns and introduced classifiers and tighter account verification to counter them. [1] [2]
anonymously sourced figures show about 45% quarterly growth and adjusted profitability after Anthropic shed distillation revenue.
Read the full assessment
Implication: demand appears broad beyond rival labs, while heavy API users may face tighter verification and pattern-based detection.
Sources are gathered; writing the dossier now.
Executive brief
Anthropic reportedly told prospective investors that "distillation" made up 5% to 7% of its revenue, and that it cut this back "materially" in Q3 2026. Distillation here means rival labs paying for heavy use of Claude so they can train their own models on its outputs. That disclosure came alongside Q3 revenue of $16.8bn, about 45% above Q2. Anthropic also said it was profitable on an adjusted basis, with an 11% adjusted operating margin that excludes stock-based compensation. All of these figures come from anonymous attendees at an investor day held ahead of a planned Nasdaq IPO. Anthropic has not published them, and no public S-1 exists yet.
What changed and event timeline
Anthropic names distillation campaigns
Anthropic said DeepSeek, Moonshot AI and MiniMax generated more than 16 million exchanges through about 24,000 fraudulent accounts to extract Claude's capabilities ().
Confidential IPO filing
Anthropic announced a confidential draft S-1, following a $65bn round at a $965bn post-money valuation ().
Q2 figures leak
Preliminary Q2 revenue was reported at $11.5bn, against $787m a year earlier, with positive adjusted operating income ().
Investor day in San Francisco
Anthropic presented Q3 revenue of $16.8bn and adjusted profitability, according to people briefed. Reports say it declined to comment to Bloomberg (;).
FT details and commentary
The FT reported the distillation share, the compute ratio and margins. Simon Willison noted that Anthropic no longer leads with annualized run-rate figures (;).
Capabilities and access
- The disclosures are financial. The reviewed sources do not link revenue to any specific Claude model or version.
- The access change that matters is the distillation crackdown. Anthropic tightened verification on its education, security-research and startup programs.
- Reports say the Q3 cut to distillation was "material" but give no method or scale (Simon Willison).
Read the full section
- The disclosures are financial. The reviewed sources do not link revenue to any specific Claude model or version.
- The access change that matters is the distillation crackdown. Anthropic tightened verification on its education, security-research and startup programs. It also deployed classifiers and behavioral fingerprinting to spot coordinated extraction traffic in its API (Anthropic).
- Reports say the Q3 cut to distillation was "material" but give no method or scale (Simon Willison).
Technical analysis for researchers and developers
- Detection approach: Anthropic described classifiers trained on extraction patterns, including elicitation of chain-of-thought reasoning, very repetitive prompt structures, and coordinated "load balancing" across many accounts.
- Where the extraction focused: MiniMax (more than 13M exchanges) and Moonshot (more than 3.4M) targeted agentic coding, tool use and computer use.
- Compute efficiency: $1.70 of revenue per $1 of compute in Q3 (Simon Willison).
Read the full section
- Detection approach: Anthropic described classifiers trained on extraction patterns, including elicitation of chain-of-thought reasoning, very repetitive prompt structures, and coordinated "load balancing" across many accounts. It also described safeguards at the product, API and model level that make outputs less useful for distillation (Anthropic). There is no published evaluation of false-positive rates, and no independent reproduction.
- Where the extraction focused: MiniMax (more than 13M exchanges) and Moonshot (more than 3.4M) targeted agentic coding, tool use and computer use. Those are the same areas legitimate heavy API users depend on, so developers running high-volume, templated agent workloads could fall within the same detection surface.
- Compute efficiency: $1.70 of revenue per $1 of compute in Q3 (Simon Willison). The sources do not say whether "compute" includes training as well as inference.
Claims and evidence
All of these figures were reported by journalists from what Anthropic told investors. None has been audited or filed publicly. One outlet describes its numbers as resting on a single attendee and calls them "indicative" (investingLive).
Read the full section
| Claim | Evidence |
| Q3 revenue was $16.8bn | Consistent across Bloomberg, CNBC and the FT via Willison |
| Earlier quarters | Figures differ by outlet: Q1 was $4.7bn or "about $4.8bn"; Q2 was $11.5bn or $11.6bn (TNW vs Willison) |
| 11% adjusted margin; distillation at 5–7% of revenue | FT, citing four people with direct knowledge (Willison) |
| Valuation target | Reported as above $2tn (36Kr) or $1.8–2.0tn (TNW) |
All of these figures were reported by journalists from what Anthropic told investors. None has been audited or filed publicly. One outlet describes its numbers as resting on a single attendee and calls them "indicative" (investingLive).
Context and prior work
- In May, Anthropic told investors its annualized run rate was about $47bn, then about $65bn by the end of July, up from roughly $9bn at the end of 2025 (AI Weekly).
- Earlier leaks put gross margins above 80%, but that figure was calculated before revenue sharing with Amazon and other partners and before training costs (cryptobriefing).
- Willison estimates that if distillation was 5% of Q2 revenue, rival labs spent more than $0.5bn distilling Claude in that quarter.
Read the full section
- In May, Anthropic told investors its annualized run rate was about $47bn, then about $65bn by the end of July, up from roughly $9bn at the end of 2025 (AI Weekly).
- Earlier leaks put gross margins above 80%, but that figure was calculated before revenue sharing with Amazon and other partners and before training costs (cryptobriefing).
- Willison estimates that if distillation was 5% of Q2 revenue, rival labs spent more than $0.5bn distilling Claude in that quarter. This is his own inference (Simon Willison).
Limitations, safety and contested findings
- "Adjusted" profitability excludes stock-based compensation, and coverage says the full list of excluded items was not disclosed (TNW).
- The Q2 figures were labeled preliminary and could be revised (TNW).
- The sources do not say whether the 5–7% distillation share refers to one quarter or the year to date.
Read the full section
- "Adjusted" profitability excludes stock-based compensation, and coverage says the full list of excluded items was not disclosed (TNW). One outlet still describes Anthropic as "heavily loss-making" (investingLive).
- The Q2 figures were labeled preliminary and could be revised (TNW).
- The sources do not say whether the 5–7% distillation share refers to one quarter or the year to date.
- The accusations against the Chinese labs come from Anthropic alone. The reviewed sources contain no independent forensic confirmation.
Business and practitioner implications
- Revenue quality: Up to 7% of revenue came from customers Anthropic now treats as adversaries, and it chose to give that up.
- API users: Expect more account verification and pattern-based throttling.
- Investors and buyers: Treat these as marketing numbers until a public S-1 appears.
Read the full section
- Revenue quality: Up to 7% of revenue came from customers Anthropic now treats as adversaries, and it chose to give that up. Growth of about 45% after the cut suggests the underlying demand is broad.
- API users: Expect more account verification and pattern-based throttling. Teams generating large synthetic datasets from Claude outputs should check them against the terms of service.
- Investors and buyers: Treat these as marketing numbers until a public S-1 appears. The listing is targeted after the November midterms and could slip (TNW).
- Rivals: If distillation spending reached hundreds of millions of dollars a quarter, lower-cost models built on frontier outputs now face a tighter supply.