Claude AI Statistics 2026: Revenue, Users and Which to Cite

Last verified: October 1, 2026.

Claude AI statistics are easy to find and hard to quote, because the figures look comparable and count different things. A run rate is not revenue, a visit is not a person, and a share needs a denominator.

This page gives each Claude number its measure, date and primary source. It also shows where credible sources disagree and which figure to cite for which claim, from Anthropic revenue to Claude active users and Claude market share. If you have to defend a number in a deck or an article, start with the cite-this-for-that table near the end.

Claude AI statistics come from three kinds of source: Anthropic's own announcements, outside measurers such as Similarweb and Ramp, and Anthropic's Economic Index, and each counts something different.

No single Claude number exists, so Claude statistics split into revenue run rates, visit and app counts, shares with named denominators, and usage samples. If you also track how AI engines send traffic, AI SEO statistics covers the cross-engine referral side.

Here are the 12 figures worth quoting, each with its measure:

• Anthropic said its run-rate revenue crossed $47 billion in May 2026, according to its Series H announcement.
• An investor update showed a $65 billion run rate by late July 2026, according to three anonymous sources who spoke to CNBC; Bloomberg was first to report it and Anthropic has not published it.
• On May 28, 2026, Anthropic's Series H raised $65 billion and valued the company at $965 billion post-money.
• More than 1,000 Anthropic customers were spending over $1 million a year as of April 6, 2026, double the February count, Anthropic reported.
• Between December 31, 2025 (about $9 billion) and May 2026 ($47 billion), the run rate rose 5.2x, by our calculation from Anthropic's two figures.
• Similarweb recorded roughly 950 million to 969 million claude.ai visits a month in its September 2026 analysis.
• Similarweb counted 103.9 million Claude app monthly active users worldwide in May 2026.
• Claude took 9.6% of website visits among major AI chatbots in August 2026, up from 2%, by Similarweb's count.
• Apptopia put Claude at roughly 17% of the US consumer mobile market for gen AI chatbot apps in its June 3, 2026 brief, which reports May data; its September 2 brief shows 14.4% for August.
• Ramp's July 2026 data shows 43.5% of US businesses paid Anthropic for subscriptions or tokens.
• Menlo Ventures put Anthropic's share of 2025 enterprise LLM spend at 40%.
• Tasks tied to Computer and Mathematical occupations were 35% of Claude.ai conversations in Anthropic's February 2026 Economic Index sample.

Last verified: October 1, 2026.

A run rate is the annualized pace of recent revenue, which means run rate is not revenue. The pace itself keeps changing, and the sections below show how fast.

Anthropic has published no monthly user total; the measured figures are Similarweb's 103.9 million Claude app users in May 2026 and roughly 950 million to 969 million claude.ai visits a month.

Anthropic's funding announcements for Series F, G and H and its Google and Broadcom compute announcement give run-rate revenue and customer counts, with no monthly user total. TechCrunch wrote in March 2026 that Anthropic had not disclosed the figure.

So every "Claude users" number you meet is an outside estimate of one surface. The table puts them side by side with what each counts:

Download CSV (CC BY 4.0)

Read the table as a set of different questions. Sign-ups answer how fast accounts are created. Unique visitors answer how many browsers reached the site. App monthly active users answer how many phones opened the app. None of them answers how many people use Claude, and adding them double counts anyone on two surfaces.

One vendor's modelled estimate puts Claude at 271.3 million monthly active users in August 2026, and the same model gives 148.2 million weekly users. That is 1.4x the sum of Similarweb's 86 million web unique visitors (March to May 2026) and 103.9 million app monthly active users (May 2026).

That sum is an undeduplicated ceiling for web plus app. Similarweb doesn't measure Claude Code or the API, so the gap is a signal and not proof. Anthropic has not confirmed any figure of 200 million or more.

Anthropic's only weekly figure is that Claude Code weekly active users doubled since January 1, 2026, with no absolute count. The 148.2 million weekly number comes from the same vendor model.

A daily count sets a floor under any monthly count for the same period, so you can reject stale figures with one comparison. Similarweb found 11.3 million daily active users in the Claude app on March 2, 2026, which makes this a floor test: any app monthly active users figure stated for March 2026 or later must be at least 11.3 million.

The 2.9 million to 7.38 million range sitting in a 2026 roundup fails it. But the range isn't wrong for its own period.

7.38 million is within 1.1% of Similarweb's June 2025 app figure of 7.3 million, so the range describes mid-2025 and is stale for a page that presents it as current. The 2.9 million end carries no date of its own.

The 18.9 million headline needs a different test. That page names no measure, and no row of its own Semrush website visitors table matches it: the table runs from 18.8 million in November 2024 to 16 million in January 2025. The vendor's 271.3 million is 14.4x that headline, so the two figures sit at opposite ends of the spread.

Anthropic's run-rate revenue crossed $47 billion in May 2026, up from about $9 billion on December 31, 2025, and anonymous sources put it at $65 billion by late July.

Claude revenue is reported as a run rate, so the timeline below uses only Anthropic's own disclosures, plus the one July figure from anonymous sources. The growth column is our calculation from the two nearest points, using actual day counts.

Download CSV (CC BY 4.0)

Growth uses actual day counts and dates the May figure to May 15, the midpoint of "earlier this month" in a May 28 announcement. The May-to-July rate stays between 11.5% and 16.4% for any May date, so the slowdown holds. The April-to-May rate moves between 30.7% and 72.7% with that date, so we don't use it for comparisons.

The run rate has grown in every interval, but the growth is slowing. Monthly growth fell from 54.9% per month (February 12 to April 6) to 13.7% per month (mid-May to July 31). The dollar gain per month went from $9.2 billion to $7.1 billion.

If you're building a forecast, don't extend the early-2026 curve. Anthropic's disclosures show a business still adding billions each month, at a slower percentage pace.

Quarterly revenue is a completed period, and a run rate is a pace at a point in time. One source told CNBC that preliminary Q2 revenue was $11.5 billion, and the company declined to comment. A draft prospectus leaked to Reuters on September 28 and 29, 2026 reports the same $11.5 billion, but Anthropic has not published or confirmed it, so treat the figure as reported.

Times four, Q2 revenue annualizes to $46 billion, which is 98% of the $47 billion May run rate. The two agree because run rate is a pace measured at one moment, and a quarter averages it.

The same logic bounds a full year. Anthropic's run-rate points were about $1 billion at the start of 2025, over $5 billion in August and about $9 billion at the end.

Suppose each level held until the next point arrived: implied 2025 revenue is (8 x 1 + 4 x 5) / 12 = $2.3 billion. Suppose each level was already in place at the start of its interval: (8 x 5 + 4 x 9) / 12 = $6.3 billion. Those 2025 revenue bounds, $2.3 billion to $6.3 billion, are what any circulating figure has to fit.

Here is how the circulating 2025 figures fare against that range:

Revenue can't exceed the highest run rate in the period, so roughly $10 billion fails. No audited revenue figure for 2025 is public: Anthropic's S-1, filed in June, is still confidential. The draft prospectus leaked to Reuters puts the 2025 figure at nearly $4.6 billion, inside those bounds; Anthropic has not published or confirmed it.

CNBC's own multiples show the same problem. Its 14-fold jump to $11.5 billion implies Q2 2025 revenue near $0.82 billion, which fits the run-rate path. Its "about sevenfold" for $65 billion implies a July 2025 run rate near $9.3 billion, well above Anthropic's August 2025 point of over $5 billion. Quote Anthropic's own points and skip the multiples.

Anthropic's post-money valuation rose from $183 billion at its Series F on September 2, 2025 to $380 billion at Series G on February 12, 2026 and $965 billion at Series H on May 28, 2026. By our calculation that is 5.3x in about nine months and 2.5x since February.

A valuation prices the future and isn't revenue. No audited profit figure is public either, because the S-1 is confidential. A draft prospectus leaked to Reuters in late September 2026 reports a 2025 operating loss of $8.06 billion, which Anthropic has not published or confirmed.

Claude Code's run-rate revenue was over $2.5 billion on February 12, 2026, and its weekly active users had doubled between January 1 and that day.

Anthropic added that business subscriptions to Claude Code had quadrupled and that enterprise use was over half of its revenue. It gave no absolute user count, so the doubling can't be converted into one.

Across five audience, spend and adoption measures, Claude's market share runs from 9.6% to 43.5%, because web visits, US mobile users, unique audience, enterprise API spend and paying businesses each use a different denominator.

Each share is true for its own question. If you're quoting Claude's market share, check the denominator first, then pick the row below that matches your claim.

Download CSV (CC BY 4.0)

The spread across these five measures is 4.5x (43.5 / 9.6). It reflects the denominators, and also differences in period, region and measurer. Consumer web visits sit at the bottom, and paid business activity sits at the top.

Two limits matter. Similarweb says Claude's API consumption "sits beyond our measurement," so its 9.6% is a consumer figure. And the 9.6% covers major AI chatbots, which is a different set from a share of all generative AI traffic, so never read one as a trend of the other.

Referral share, meaning Claude's share of AI chatbot referral sessions, is a far lower ruler that belongs with our search engine market share data. For the wider field, see how AI search engines compare.

ChatGPT leads Claude by more than 17 times on 12-month average web visits, but its lead on unique visitors shrinks from 12.6 times on 12-month averages to about 5 times in May 2026, while on app users it is 10.8 times in May 2026.

Similarweb's figures show why the multiple falls. Claude.ai visits grew from 100 million in May 2025 to 953 million in May 2026, which is 9.5x, so a 12-month average mostly measures when you sample Claude. ChatGPT's visits did not grow anywhere near that pace.

Download CSV (CC BY 4.0)

Never put a 12-month multiple beside a May multiple without the window column. The 12.6x is 467 million over 37 million, and the 10.8x is 1.12 billion over 103.9 million.

Overlap moved too. The share of Claude users who also used ChatGPT in the same month fell from 78% to 61%, while the share of ChatGPT users who also used Claude rose from 2% to 14%. ChatGPT's own figures are in our ChatGPT statistics guide.

Anthropic had over 1,000 customers each spending more than $1 million a year in April 2026, and Ramp counted 43.5% of US businesses as Anthropic payers in July 2026.

Anthropic reported over 300,000 business customers in September 2025, then switched to counting the $1 million cohort. That cohort went from over 500 in February to over 1,000 in April, doubling in under two months.

Menlo's 40% and Ramp's 43.5% count different things. Menlo estimates dollars of enterprise LLM spend from about 500 US decision-makers plus a bottoms-up model, and Menlo is an Anthropic investor. Ramp counts US businesses that paid for subscriptions or tokens, with a sample that skews tech.

In our judgment, a business paying for one seat counts fully in Ramp's 43.5% and barely in Menlo's 40%. If you're sizing the enterprise opportunity, use Menlo. If you're measuring how many companies pay at all, use Ramp.

Claude is not losing users on the measures Similarweb publishes: visits flattened near 950 million a month after a fourfold jump, while six-month retention rose to 45%, still behind ChatGPT's 59%.

Three signals tell the story:

• Visits plateaued. Monthly visits to claude.ai rose from 203 million in February 2026 to 824 million in April (4.1x), then sat at roughly 950 million to 969 million.
• Retention improved. Six-month retention was 25.8% for the January 2025 sign-up cohort and 44.8% for the February 2026 cohort.
• A gap remains. Six-month retention is 45% for Claude against 59% for ChatGPT, a 14-point difference.

Visits stopped climbing after the jump. That makes the retention gap to ChatGPT the number to watch, and the visit count adds little. Similarweb itself notes that recent Claude cohorts soften at months three and four.

One caveat applies to the visit signal. Similarweb counts each session as a visit, so a user who comes back daily counts many times, and a plateau in visits can hide a growing or shrinking audience. Treat the plateau as evidence of slower acquisition. It isn't a user count.

Coding is Claude's largest single use: Anthropic's Economic Index sample for February 2026 put 35% of the Claude.ai conversations it sampled in tasks tied to Computer and Mathematical occupations, which Anthropic calls coding.

The Economic Index measures a share of conversations, so its percentages describe what was asked and never how many people asked. The table collects the shifts that matter most:

Download CSV (CC BY 4.0)

Don't merge the two 35% figures. One is the Computer and Mathematical share of Claude.ai conversations, and the other is the November personal-use share.

Several shifts come from where coding runs, the school calendar and who just signed up, according to Anthropic's own explanations. Three mechanisms sit behind the table:

• Coding moved to the first-party API. Claude.ai's top-10 concentration fell partly because coding tasks migrated from Claude.ai to Anthropic's first-party API, where Claude Code has grown to a large share of sampled traffic.
• Claude Code splits work into small calls. The API's own concentration stayed roughly flat because Claude Code splits coding work into smaller calls that are labeled as distinct tasks. These are two claims about two surfaces and don't form one causal chain.
• Winter break and new sign-ups. Coursework fell partly because many students were on winter break in the sample, and personal use rose as sign-ups beginning around February brought more casual users.

Which means a shift in the table is a hypothesis about users only after you've checked it against these mechanisms. A drop in coursework in January and February says little about students leaving.

Mind the sample sizes. The March report samples 1 million conversations from Claude.ai and the first-party API. The 81,000 interviews and about 9,700 linked survey respondents in the June 2026 report are research samples that say nothing about audience size. For turning AI traffic into something you can measure on your own site, see AI search analytics.

Seven widely repeated Claude figures need a correction before you quote them: two are stale, two are unconfirmed estimates, two mislabel their measure and one is wrong at the source.

Each failure is one of four kinds, named in the verdict column:

Download CSV (CC BY 4.0)

Count them: 2 stale, 2 unconfirmed, 2 mislabeled and 1 wrong.

Cite a run rate for Anthropic's pace, Similarweb's unique visitors for web audience, app monthly active users for mobile reach, Ramp for US business adoption and Menlo for enterprise API spend.

Use this table to match the claim you want to make with the number that supports it:

The rule behind the table: name the surface and the window every time, and a Claude number becomes quotable. For how these figures fit the wider shift in AI search, see AI search trends 2026.

Sources are Anthropic's funding announcements and Economic Index reports first, then named measurement firms. Every figure carries its month. Figures relayed through press, the $65 billion run rate and the $11.5 billion Q2 revenue, from anonymous sources (the Q2 figure also appears in a draft prospectus leaked to Reuters), are labelled as reported and unconfirmed by Anthropic. Vendor models are described as models, and no first-party data is used.

Next step: before your next deck or article, replace every "Claude has N users" with a surface, a window and a source from the table above. Claude's numbers are real once they carry their measure.