Who Is Buying Into the $1.1TN Legal Market?

The Information reported last week that Harvey is in talks to raise at least $500 million at a $15.5 billion valuation. Harvey has not confirmed it. Five months ago it raised at $11 billion.

The same month, Crosby, an AI-native law firm with around thirty lawyers, said it is arranging professional liability insurance for its AI agents so they can work without a lawyer checking every output.

One is the biggest valuation legal AI has produced. The other is a small firm buying an insurance policy. I think the insurance policy matters more.

Legal is a $1.1 trillion market and you cannot simply buy your way into it. In most of the United States, non-lawyers are not allowed to own a law firm, and that includes the venture funds behind every company in this story. And however good the models get, somebody has to carry the blame when the work is wrong, which for now means a human being with insurance behind them. Those two rules set the pace at which AI takes legal work, not model quality, and both of them moved this month.

So this edition is about the rules and the ways round them. What the tools are doing with their money, the two jurisdictions everyone is using to get past the ownership ban, and who ends up buying into legal as a result.

In edition one I mapped four positions in this market. AI-native firms take the floor, the Kirkland tier defends the ceiling, the tools face a squeeze, and the middle becomes acquisition inventory. Five weeks on I would not change any of that. What I got wrong was the timetable.

Harvey first. When I wrote edition one it was worth $11 billion on about $190 million of ARR, which is 58 times revenue. ARR has since reached roughly $300 million, with total annualised revenue past $350 million. At $15.5 billion that works out at about 52 times ARR, or 44 times if you count everything.

So the valuation went up 40% and the multiple came down, which is not what I expected to be writing five weeks later. The worry changes with it. At 58 times you had to ask whether the price had much to do with the business at all. At 52 times you only have to ask what happens if growth slows for a couple of quarters.

Meanwhile Harvey and Legora have bought eight companies between them this year, and two of the targets are not legal businesses at all. Benchmark sells to asset managers, Cadastral does commercial real estate. And in June, Harvey signalled it would start building its own models rather than rely entirely on OpenAI and Anthropic, which supply it and compete with it at the same time.

In edition one I said the tools would eventually have to cross into legal services because the arithmetic forces them to. They are still not crossing. They are buying capability, buying adjacent industries and reducing their dependency on the labs, which is everything except the one move I predicted. The reason is legal rather than strategic, and it is the first of the two rules.

The rule is about ownership, not staffing. Plenty of lawyers work at Harvey and Legora. What neither company can do is own the firm that does the work, because their own shareholders are venture funds. That is why the largest valuations in legal AI sit with companies selling software to lawyers, and why every serious entrant has had to find a jurisdiction that allows something else.

There are only two of those, and both are busy.

Arizona dropped the ownership ban completely in 2021, and now hosts alternative business structures for KPMG, Axiom, LegalZoom, Elevate and Eudia Counsel. Utah runs a narrower sandbox. So outside capital reaches US legal services through one state and a half, which is a strange position for a trillion-dollar market to be in.

England and Wales has allowed outside ownership since 2007, subject to the regulator approving it. Garfield went to the Solicitors Regulation Authority and became the first fully AI-driven firm it has passed. Tacit Legal is another SRA-regulated firm built around its own AI.

Inside those two jurisdictions you can build a firm or buy one, and buying is quicker. Carta did that in May with Avantia, a UK firm used by more than 200 asset managers, including nearly a third of the largest funds in the world, on transactions Carta says are tied to over $15 trillion of assets under management. It became Carta Law. Henry Ward put it bluntly at the time. Large PE firms pay top-tier firms for high-volume routine work, and they should not have to.

I mentioned that deal in passing in edition one and I underrated it. Carta was not entering legal for its own sake. Its customers already run cap tables, valuations and fund administration on Carta, the legal work sits inside those same workflows, and buying a firm turned legal into another line on a bill they were already paying.

Outsiders are not the only ones buying their way in. Lawhive, which runs an AI-native firm for consumers and small businesses, opened in New York with the stated purpose of acquiring US law firms and moving them onto its own platform. On the deal page that looks the same as a software company buying one.

I have not seen anyone copy the Carta move yet, and it is the most copyable thing in this market. Any software company that already holds the system of record for an industry can buy a small regulated firm and start selling the outcome instead of the tool. Norm Ai has done a version of it with Norm Law, and has raised a further $120 million at a $1.2 billion valuation.

My call is that the next serious entrant into legal services will not come from legal at all. It will be a vertical software platform with a regulated firm attached, aimed at one industry rather than at lawyers generally.

The second rule is less visible and probably harder to move.

The AI-native firms all keep a person in the loop today. Crosby's lawyers review each output its agents produce. Garfield works under solicitor supervision and is not permitted to propose case law at all. The models could very likely do more of that work unsupervised. The review is there because somebody has to be insured, and at the moment that somebody is a human being.

Which is why Crosby's insurance project interests me more than the Harvey round. The company says capability is no longer the constraint, and that it is working with auditors, bar associations, state regulators and insurers to get cover for the agents themselves. It is trying to remove the last person from a process that technically stopped needing one a while ago.

If an underwriter agrees to price that risk, the review layer becomes optional. If none will, it stays compulsory however good the models get. Which puts the timetable for legal automation in the hands of insurers rather than labs, and that is not where anyone was looking.

I wrote in the insurance edition that carriers are busy writing AI exclusions into their policies because they cannot price the exposure. The first customers for exactly that cover are now appearing. Whoever works out the pricing gets to set the speed limit for professional services automation, and gets paid for setting it.

The question is no longer whether AI takes your production work. It is which of the two rules protects you, and for how long.

The ownership rule is being routed around rather than removed, and it will not protect a mid-market practice from an Arizona-structured competitor with permanent capital behind it.

The liability rule lasts longer, and it tells you where the value sits. Judgment, accountability and putting your name on advice that might turn out to be wrong are hard to automate, because they are a form of insurance rather than a task. A firm that understands it is selling risk-bearing will price better than one that thinks it is selling document review.

If you are in-house, both rules are why your legal spend has not fallen yet, and roughly when it will.

The four positions from edition one stand. These are about timing.

Within a year, another vertical software platform buys a regulated law firm the way Carta did. Watch any industry where a single platform already holds the system of record.

Within a year, Harvey or Legora takes a position in legal delivery rather than only selling software into it. My guess is an ABS structure rather than an outright acquisition, because that route does not look like competing with their own customers.

Within two years, an insurer sells standalone cover for autonomous professional work as a named product, priced painfully at first. Crosby is among the first buyers and will not be the last.

And more states follow Arizona, because the ones that keep the ban will simply export the work to the ones that do not.

I could be wrong on any of these. Worth checking back when the dates come round.

Every services market I have looked at resolves into those same four positions, and legal is going the same way.

What changes from market to market is what sets the pace. In accounting it was permanent capital. In insurance it was private equity, which was already there. In legal it is a regulator in Arizona and whoever eventually agrees to insure an agent.

Subscribe and every edition lands in your inbox. If you work in legal and I have missed something you can see from the inside, reply on LinkedIn. I read everything, and the sharpest replies shape the next edition.

And if you are thinking about selling a professional services firm, or buying one... that is exactly what I do. I run Eilla, an AI-native M&A advisory. Message me.

See you next time.