Large law firms are paying to train AI on their own deal and trial files
Kirkland & Ellis agreed this year to pay Palantir $500 million to build technology for its private equity fundraising advice, Business Insider reported. With Harvey saying it has users at 80% of the largest 100 firms, one consultant says the edge has to come from what the models know.
The Board Room · Leadership desk
Morgan & Morgan says it will spend $1 billion over the next decade on its push to build its own technology, which averages $100 million a year if spent evenly. Kirkland & Ellis agreed to pay Palantir $500 million to build technology for advising private equity clients on raising money.
| Figure | Value | Note | Claim |
|---|---|---|---|
| Morgan & Morgan says it will spend over the next decade | $1 billion | 4 | |
| A year, if spent evenly | $100 million | 11 | |
| Kirkland & Ellis agreed to pay Palantir | $500 million | for private equity fundraising tech | 1 |
What happened
- Latham & Watkins bought its own Nvidia servers to start customizing large language models.
- Morgan & Morgan has said it will spend $1 billion over the next decade on building its own technology.
- Harvey took an open-weight model and trained it further for legal work this year, and eventually wants to help firms do the same with their own matters.
- Thomson Reuters is offering customers Thomson, a model trained on 150 years of its own content.
Why it matters
- contradiction The in-house edge is partly bought. Kirkland's $500 million goes to an outside vendor, so the proprietary input is the firm's data.
- constraint Case files cannot go into a model as they are. Morgan & Morgan has to strip identifying details before it can look for patterns across cases, so data preparation comes first.
- exposure A firm with a thinner record of deals or cases has less to train on, so the approach favors whoever holds the largest file in its niche.
Shared tools have removed the old differentiator. Legora says it covers a majority of the country's 100 largest law firms by revenue [13], and Harvey says it has users at 80% of them, or 80 firms [12][19]. Mary O'Carroll, chief executive of LegalEng Consulting Group, said that once every firm has access to the same systems, the advantage must lie in what those systems have learned [14]. Her example is a firm that has "done every private equity deal for all of time and has a corner on the market" [15]. The prize, she said, is lawyers who can say, in effect, "I have the knowledge of 50,000 deals in my back pocket" [15].
The firms spent the past two years adopting tools from OpenAI, Harvey, Microsoft, Legora and others [17]. The vendors now want to sell the next layer. LexisNexis says it is aiming for models grounded in both its legal corpus and a firm's own data [9]. Sean Fitzpatrick, its chief executive of global legal, said: "All of the models can do things in a similar way" [16]. He sells these models, so the claim is a vendor's. It also fits O'Carroll's argument: if the models are alike, a firm's files are the variable. In our view that makes the files the asset, because the training method will not stay exclusive to one firm.
The two headline budgets are different units. Kirkland's $500 million buys technology for one business line [1]. Morgan & Morgan's $1 billion is for the firm's wider push to build its own technology over a decade [4], an average of $100 million a year if it is spent evenly [11].
Owning the files does not settle who may use them. "We have more data than anybody else," said Yath Ithayakumar, Morgan & Morgan's chief transformation officer. "We have to be good stewards of it." [5] He said the firm has to be careful about who gets access to its data, what they are allowed to do with it, and whether outside vendors can retain or learn from it [6]. We'd expect the last question to be answered in the vendor contract. A firm that signs one this quarter settles it before any model has produced a result.
Business Insider's account reports no measured result from any of these builds. The one worked example is Morgan & Morgan's own: the firm says its software can flag when a patient with symptoms of a compressed nerve close to the spine may need surgery down the line [20]. "Our best and experienced people will spot it almost immediately," Ithayakumar said. "But if you're a first-year attorney, this is your first job, maybe that's not as obvious to you. Technology makes a huge difference." [21] Business Insider describes the strategy as a bet that AI built on a firm's record of past deals, its approach to trials and what its clients prefer will prove more useful than anything a competitor can buy [18].
What to watch
- Whether any firm publishes a measured result, such as accuracy or time saved, from a model trained on its own matters.
- Whether Harvey and LexisNexis ship products that train models on a single firm's data, which would test how exclusive the method stays.
- Whether Latham's own-server approach and Kirkland's vendor-built approach produce different results once either reports outcomes.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence35
- Adoption50
- Hype gap+30
- Incentives70
- Confidence50
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Earlier this year, Kirkland & Ellis agreed to pay Palantir $500 million to build technology for one of the firm's core businesses: advising private equity clients on raising money.
- [2]
Latham & Watkins bought its own Nvidia servers to start customizing large language models.
- [3]
At Morgan & Morgan, America's largest personal injury law firm, lawyers have begun using an in-house AI model trained to recognize patterns in their cases.
- [4]
Morgan & Morgan has said it will spend $1 billion over the next decade on its push to build its own technology.
- [5]
"We have more data than anybody else," Ithayakumar said. "We have to be good stewards of it."
- [6]
Morgan & Morgan has to be careful about who gets access to its data, what they are allowed to do with it, and whether outside vendors can retain or learn from it, Ithayakumar said.
- [7]
This year, Harvey took an open-weight model and trained it further for legal work, and eventually wants to help firms do something similar with their own matters.
- [8]
Thomson Reuters is offering customers access to Thomson, a model trained on 150 years of its own content.
- [9]
LexisNexis says it is aiming for models grounded in both its legal corpus and a firm's own data.
- [10]
Before Morgan & Morgan can use its data to spot patterns across cases, it has to strip away any details that could identify the people behind them.
- [11]
$1 billion over ten years averages $100 million a year if spent evenly.
- [12]
Harvey says it has users at 80% of the country's 100 largest law firms by revenue.
- [13]
Legora says it covers a majority of the country's 100 largest law firms by revenue.
- [14]
Mary O'Carroll, chief executive of LegalEng Consulting Group, said that when everyone can buy the same systems, the edge has to come from what those systems know.
- [15]
O'Carroll said that if a firm has "done every private equity deal for all of time and has a corner on the market," the prize is lawyers being able to say, in effect, "I have the knowledge of 50,000 deals in my back pocket."
- [16]
"All of the models can do things in a similar way," said Sean Fitzpatrick, chief executive of global legal at LexisNexis.
- [17]
Large law firms spent the past two years adopting tools from OpenAI, Harvey, Microsoft, Legora, and others.
- [18]
The bet is that artificial intelligence steeped in law firms' deal histories, trial strategies, and client preferences will be more useful than whatever the firm across the street can buy.
- [19]
80% of the 100 largest firms is 80 firms.
- [20]
Morgan & Morgan says its software can flag that a patient showing signs of a pinched nerve near the spine could later require surgery.
- [21]
"Our best and experienced people will spot it almost immediately," Ithayakumar said. "But if you're a first-year attorney, this is your first job, maybe that's not as obvious to you. Technology makes a huge difference."
Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.
Sources
1 independent publisher whose own reporting we read for this story.
- businessinsider.comLaw firms are building AI around their most valuable asset: experience
1 article · October 14, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Topics
- Data governance workflowsFollow
- Legal AIFollow
- Enterprise AI customizationFollow