On May 28, the Financial Times reported that Kirkland & Ellis is committing $500 million to build its own AI platform. Not a Harvey license. Not a Copilot rollout. Its own system, built by roughly 180 technologists working with input from 250 of its lawyers, including 100 partners. Kirkland will own all of it. The outside companies helping build it are contractually barred from selling it to anyone else.
In the same three weeks, Fried Frank launched FundAssist, a proprietary platform that mines its own historical deal work to surface the firm’s preferred drafting language. Linklaters stood up Applied Intelligence, a practice that pairs lawyers with data scientists to build bespoke AI for individual matters. Freshfields deepened its co-development deal with Anthropic.
Four elite firms. Three weeks. One direction.
The trade press is calling it the “build versus buy” moment for Big Law. That framing is half right, and the half it gets wrong is the half that matters for every firm that isn’t Kirkland.
What Kirkland is actually buying for $500 million
Kirkland’s chairman, Jon Ballis, gave the FT the line that explains the whole decision. Off-the-shelf tools, he said, are “raising the floor for everyone.” Then: “we don’t get hired for the floor.”
Read that twice, because it is the most important sentence published about legal AI this year, and a competitor said it, not a vendor.
Ballis is not knocking Harvey. He is conceding that Harvey is good, that it makes every firm that uses it more capable, and that this is precisely the problem. A capability every firm can buy is a capability that differentiates no firm. When the litigation boutique down the street runs the same model, trained on the same public corpus, accessed through the same interface, the tool raises everyone’s floor and lifts no one’s ceiling.
So Kirkland is not spending half a billion dollars to be faster. It is spending it to build something no competitor can replicate: a system trained on how Kirkland’s own lawyers actually do the most complex work in private equity, restructuring, and litigation. As one analyst put it, nobody else can buy that platform because nobody else has the source material to build it from.
That is the entire thesis. The value was never the model. The value is the institutional knowledge you feed it, and whether that knowledge compounds into something you own or evaporates into a vendor’s general-purpose product.
Kirkland just spent $500 million to prove the point.
The false choice the headlines are selling
Here is where the “build versus buy” story breaks down.
The way it is being told, a firm has two options. Buy: license Harvey or CoCounsel, accept the generic floor, move on. Build: assemble 180 engineers, extract knowledge from 250 lawyers, and spend 1% of revenue a year until you have your own platform.
For Kirkland, 1% of revenue is $100 million a year, because last year the firm booked $10.6 billion and paid its equity partners an average of $11.1 million each. It can fund a half-billion-dollar bet out of pocket and barely feel it.
No firm below the AmLaw 50 can do that. A 150-lawyer litigation firm cannot hire 180 technologists. A regional powerhouse doing $400 million a year cannot put $4 million a year into an internal AI lab and wait three years to see if it works. The press framing quietly tells every firm that isn’t a global giant that “build” is off the table, which leaves them with “buy” - the generic floor Ballis just told you he refuses to compete on.
That is a false choice, and it is costing the middle of the market the most important strategic decision it will make this decade.
Owning your AI is not the same as building Kirkland’s AI
The thing Kirkland is actually buying is not 180 engineers. It is ownership: a system trained on its own matters, running on infrastructure it controls, that compounds in value as more of the firm’s work flows through it, and that stays the firm’s property no matter what any vendor does next.
You do not need $500 million to own that. You need the right architecture and a partner who builds it on your infrastructure and hands you the keys.
The mechanics Kirkland is paying a premium to assemble in-house already exist as a delivery model for firms a fraction of its size. Fried Frank’s FundAssist mines historical deal work to surface precedent and preferred language. That is not a capability reserved for billion-dollar firms. A landlord-tenant practice running hundreds of matters a month has the same raw material: years of filings, outcomes, and judgment calls that currently live in senior partners’ heads and walk out the door when they retire. A personal injury firm sitting on thousands of closed case files has a settlement-pattern engine waiting to be built. The source material is the firm’s own history. The only question is whether that history compounds into an asset the firm owns or stays trapped as unsearchable dead weight.
What Kirkland understands, and what the middle market is being talked out of, is that the firm’s accumulated work is the moat. Not the model. The model is the floor, and the floor is for rent. The moat is what you build on top of it, and the moat is only a moat if you own it.
The decision in front of everyone else
Ballis was clear-eyed about the bet. The firm will fund it out of revenue, which means a short-term hit to partner distributions, on the conviction that owning the intelligence layer is where long-term value accumulates. He is also using it to push Kirkland off the billable hour toward outcome-based pricing, because when the system compresses the low-value work, hours stop being the honest unit of value.
Every firm leader reading the Kirkland headline should sit with one question. Not “can we spend $500 million,” because the answer is obviously no. The question is the one underneath it: in five years, do we want to own the system our most valuable work runs on, or rent it?
Kirkland answered. So did Fried Frank, Linklaters, and Freshfields. The firms that can write a nine-figure check have already decided that owned intelligence beats rented capability.
The firms that can’t write that check face the same decision with less margin for error, not more. Renting the floor is safe right up until the moment every competitor is standing on the same floor. Then the only firms with a ceiling are the ones who built one.
You don’t need Kirkland’s budget to make Kirkland’s choice. You need to stop believing those are the only two options on the table.
About the author. Jon Ventoso is the founder of LISA — Legal Intelligence Sovereign Architecture. LISA designs and builds purpose-built AI infrastructure for law firms, deployed inside the firm’s environment and owned permanently. The views above are the author’s own and do not constitute legal advice.