Updated for Amendment No. 1, filed
Some lawyers have Claude because they, their firm or their legal department pays for it. More are modeled as reaching it inside Harvey or Legora, where the vendor pays for Claude. The evidence for that second group is weaker.
US legal users, direct and through vendors
48,400 direct
about 104,000 through vendors
a year reaching Anthropic from US legal work
$37.5M direct
about $46M through vendors
A model produced all of these figures and nobody has measured them. The vendor part is less certain, because it rests on assumed vendor costs and Claude shares. Other cases, and sliders to change the assumptions, are further down.
48,400 direct, modeled as people whose firm, legal department or own account pays.
about 104,000 through vendors, counting Harvey and Legora users modeled as reaching Claude.
$37.5M direct, modeled as paid by those people, their firms and their legal departments.
about $46M through vendors, modeled as what vendors pay for Claude.
The largest group of direct users pays for its own plan. Law-firm seats come next, and legal departments after that.
US users on the direct route, modeled. Plan-by-plan breakdown: Note 7Notes 3 4 5 8
The money ranks differently from the people. People paying for themselves are the biggest group and one of the smaller bills, because only the legal part of a subscription counts. Firm seats and the tools firms build carry most of the spend.
Most of this money comes from a small group. The Base case has 25 Am Law 100 firms building their own tools at about $350K a year each. If you know what a firm actually spends, this is the first assumption to replace.
Firms in each size group × share that build × spend per builder, from about $2K to $30K a year.
Freshfields runs Claude on its own platform, and it is not a US firm. Quinn Emanuel has a partner-built litigation platform; how it pays for Claude is not stated. Kirkland & Ellis plans to spend $500M over three to four years on its own AI platform, Reuters reported, and declined to say whether it would rely on a specific model. None of that is counted as Claude spend.
$350K a year across a firm of 1,805 people firmwide is about $16 a person a month. Across the 1,444 assumed to be in the US, the count used for whole-firm seats, it is about $20. The S-1/A puts Enterprise users without Claude Code at under $75 a month. The per-firm figures cover what a firm pays for Claude itself and leave out developers and hosting.
A firm's own tools count whether or not it also buys seats. Nothing comes off for offices outside the US, so spend at global firms may run high. People using tools their firm built appear in no user row, which leaves the user count short by an unknown amount.
API spend a year, part of the direct route, modeled.Notes 3 6
Claude's mark sits far below Thomson Reuters, RELX and Wolters Kluwer. Read it for scale and nothing finer, because the bars mix revenue with annual recurring revenue, and the vendor money in the mark also sits inside the vendors' own bars.
They mix fiscal-year revenue with ARR (annual recurring revenue), and different dates and regions, and they overlap, so they do not add up. Vendor ARR may include money that ends up with Claude. On assumed shares, about $64.2M a year of global spend reaches Claude through legal vendors in the Base case. The US part, about $46M, is in the Claude mark and is also inside those vendors' own bars, so the mark and the bars overlap.
Westlaw and CoCounsel product revenue is not disclosed. No reliable total exists for contract lifecycle management (CLM) software, and published estimates run about $1.1B to $2.3B. Docusign IAM is about $353M, calculated, global and broader than CLM. Icertis was approaching $350M ARR in Aug 2025, reported by Sacra.
Gartner forecast in 2024 that global legal tech would reach $50B by 2027. Menlo puts enterprise legal AI apps at $650M in 2025, leaving out general assistants, and Menlo lists Claude as a portfolio company. The Census puts US legal services revenue at $447.4B in 2025.
Revenue and ARR on one scale, $ millions.Note 9
The total is five counts added together, four on the direct route and one on the vendor route. Each narrows a starting group of people or a revenue figure through a chain of assumptions, and the notes listed below give them.
US firms in five sizes × share using or exploring Claude × a discount because survey respondents chose to answer × share buying seats × seats per firm. 80% of seats count as users.
Legal professionals × share using AI × share choosing Claude × share paying for themselves. Spend counts only the part of the subscription used for legal work.
In-house, compliance, legal operations and government staff, split by who bought access.
Firms running their own Claude tools × Claude spend per firm, plus tools legal departments build, at an assumed US share.
Vendor revenue × model costs × Claude's share, cut to the US at 72.2%. Harvey and Legora claimed users × Claude's share, at the same 72.2%.
Steps 1 to 4 are the direct route, and step 5 is the vendor route.
Modeled.Notes by step: law firms 3 individuals 4 in-house and staff 5 US API 6 vendors 8 total 2
Everything above is the Base case. Here it is beside three other cases, with no odds on any of them. Pick a case to load its inputs, then move the sliders.
total US legal users, modeled
total a year reaching Anthropic, modeled
| Line | Users | Spend |
|---|---|---|
| Law-firm seats | ||
| People paying themselves | ||
| Legal-department seats | ||
| API | not counted | |
| Direct | ||
| Through vendors, US |
Moves vendor figures only, modeled.Note 8
a year worldwide reaching Anthropic through legal vendors, modeled. The US part goes into the total.
| Vendor spend | Inputs | Result |
|---|---|---|
| Harvey | ||
| Legora | ||
| Other vendors | ||
| US | ||
| Vendor users | Inputs | Result |
| Harvey | ||
| Legora | ||
| US |
Adjusted results are arithmetic on assumed inputs. The first four sliders move only direct figures and the two vendor sliders only vendor figures. Claude's share moves vendor spend and vendor users, but model costs moves vendor spend alone. Law-firm seats are counted by firm size, which no slider touches. Every other input, including the 72.2% US share, stays at the loaded case's value. The four direct sliders run past the Low and High values. The vendor sliders stop at the low and high in Note 8. Loaded on Base they run from $20.0M to $150.8M and never reach $201M, which also needs $800M of revenue at other vendors.
Every figure attributed to Claude is modeled. Figures are US and at a yearly rate unless a line says otherwise. The shaded column is the Base case.
Related: How many people pay for Claude? · For lawyers