Europe's forgotten AI winners
Harvey hired Jude Law. Legora collected the VCs. But the real winner in legal AI might be a 50-year-old database company you've never thought twice about.
On the left is a grey-scale version of a rat with inaccurate and oversized reproductive organs recognisable from an academic article which contained AI slop. The image has been edited to be sliced, and blue painted torus icons overlay the image. The background features a green mountainous range with a magenta tiled floor and gradient sky.Image: Marcin Wilkowski / Better Images of AI / CC BY 4.0
If you've walked through Liverpool Street Station or Piccadilly Circus recently, you might have caught Jude Law staring down at you. He isn't advertising a film. He's the new face of Harvey AI, the legaltech juggernaut that's spent the past two years convincing investors — and now London commuters — that his last name is its destiny.
But while Harvey pays movie stars to advertise its name and Swedish rival Legora collects VCs on its cap table, the actual winner in legal AI could be a 50-year-old database company you've never thought twice about.
Before you yawn, let me explain. I'm talking about LexisNexis — the legal research database founded in 1973. It's where lawyers go to find cases, check citations, and figure out how a law has been interpreted, and its biggest competitor to date has been Thomson Reuters.
When ChatGPT launched in late 2022, LexisNexis's parent company RELX did something distinctly un-incumbent: it panicked productively. Sources tell me the company called an internal code red, froze non-essential product development, and redirected resources toward generative AI.
By October 2023, it had shipped Lexis+ AI, a generative-AI legal research tool built on retrieval-augmented generation grounded in its own proprietary corpus — more than 200 billion interconnected legal documents. In February this year, it replaced that with Lexis+ with Protégé, an agentic AI platform that layers natural-language search, automated drafting and document-grounded answers on top of the same legal knowledge graph.
RELX's legal division posted 9% annual revenue growth in 2025, up from 7% in 2024, and delivered the strongest profit growth of any of RELX's four divisions. The legal segment alone generates £1.8 billion in annual revenue across more than 150 countries. Despite this, RELX's share price is down 20% this year. The market just doesn't seem to accept an incumbent that isn't backed by the who's-who of Sand Hill Road.
Under the hood
Are the VCs seeing something we aren't? Harvey says it has more than 1,000 customers and hit $190 million annual recurring revenue (which means actual 2025 revenue was lower — ARR is an annualised run-rate, not a bank statement). Legora says it works with 800 law firms and legal teams across 50 markets. It hasn't disclosed ARR.
The question is how much of that revenue is durable. Legal AI products are still early enough that many contracts were signed on the basis of "we need to be doing something with AI." The renewal cycle will be the real test.
On the surface, all three companies now sell similar things: agentic flows for legal work. Harvey launched its workflow builder last June. Legora shipped its answer the same month. Protégé has more than 300 pre-built workflows.
But underneath, they diverge sharply. Protégé is built on top of what is arguably the single most valuable proprietary dataset in legal: more than 200 billion documents, Shepard's Citations, and decades of curated authority. Harvey and Legora run on general-purpose models from OpenAI and Anthropic — a much shallower foundation. Shallow enough that last year, Harvey entered a partnership with LexisNexis to gain access to a slice of the incumbent's U.S. legal library. That slice represents less than 1 percent of LexisNexis's full repository. Harvey needs LexisNexis's data. Not the reverse.
The billable hour problem
This matters because of something the new legal AI tools tend to ignore: the primacy of hourly billing. A tool that lets a junior associate do in 20 minutes what used to take four hours doesn't make the firm more money. The partner still needs their margin. The best legal AI products will be the ones that expand what lawyers can do, not compress it — research tools that surface arguments a lawyer wouldn't have found, not tools that find the same arguments faster.
That's where LexisNexis has an underappreciated edge. Its AI sits on top of the dominant legal research corpus, the thing lawyers already use every day. Harvey and Legora are selling a new workflow — which means they're competing against the billable hour itself.
It's surprising when investors ignore this. Is RELX just not talking about it enough? The tie-up with Harvey is also confusing — does RELX want to buy it? (Too expensive.) Put it out of business? Or just collect rent on its data?
We all know what happens when highly funded, unprofitable companies hire celebrities and sponsor sports teams. The best AI trade might not be the one with the best pitch deck. It might be the one that already owns the law.
The bottom line: if you're competing against a VC-backed rival with a war chest and a celebrity spokesperson, the worst thing you can do is let the numbers speak for themselves. The market still doesn't believe RELX is an AI story. If you don't claim the narrative, your better-funded competitor will define it for you.
Continue reading
This analysis is for Second Chapter members. Join the executives, founders and policymakers reading our reporting from inside the room.
From €250/year. Become a member to read in full.
The conversation
14 comments · 3 from our teamSecond Chapter is a multi-directional resource. Comment, challenge our reporting, or suggest an angle for the next piece in this series — our journalists are reading.
The "incumbent with the proprietary dataset" pattern isn't unique to legal — I think it repeats in healthcare, accounting and engineering. If you work somewhere with a deep, decades-old data moat that's quietly shipping AI on top of it, I want to hear from you for the next one.
As one of the VCs you're gently roasting: the durability question is the right one. We underwrite ARR assuming a real renewal cliff. But you're underrating distribution — Harvey's brand spend buys it the top-of-funnel an incumbent struggles to match. Data moat vs. distribution moat is the actual fight.
Fair pushback, Mathilde — and a good framing. Distribution vs. data moat could be the whole follow-up. Noted for the Advisory Group.
The billable-hour point is the sharpest thing I've read on legal AI. Our own internal counsel resisted the "do it faster" tools for exactly this reason. Expand-what-you-can-do beats compress-what-you-do, every time, in any margin-protected profession.