Kirkland & Ellis Decided Not to Buy Legal AI. It Is Spending $500 Million to Build It.

The world's highest-grossing law firm has just told the legal AI industry that its products are table stakes.

Kirkland & Ellis has committed $500 million over the next three to four years to build proprietary artificial intelligence rather than rent it, chair Jon Ballis told the Financial Times in May. More than $100 million of that lands in year one. On June 4, the firm revealed the first name attached to the money: Palantir Technologies, whose Artificial Intelligence Platform now underpins a custom engine for private equity fund formation -- the documents, side letters, investor terms and compliance obligations that sit at the center of Kirkland's most lucrative franchise.

The number is large. The reasoning is larger.

"The idea is that we're going to take the collective intelligence of our institution and be able to deploy that throughout our firm," Ballis told the FT. Widely available AI tools, he said, were "raising the floor for everyone" in the legal industry -- and Kirkland, in his framing, does not get hired for the floor.

The Numbers Behind the Bet

Kirkland is spending from a position most firms cannot imagine. It booked $10.6 billion in revenue in 2025, becoming the first law firm in history to clear $10 billion, on a jump of roughly 20 percent. Profits per equity partner reached about $11 million across 595 equity partners, and headcount now exceeds 4,000 lawyers. The $500 million comes out of revenue, meaning partners absorb the hit to distributions, at least in the near term.

The infrastructure is further along than the May announcement implied. Kirkland's AI and innovation group now numbers more than 180 engineers and data scientists, with roughly 50 dedicated AI engineers currently engaged -- about 35 through outside partnerships and more than 15 in-house. Some 250 attorneys, including about 100 equity partners, have been feeding the system a record of how they actually do their jobs, through internal groups the firm calls AI Pods. Nearly 40 AI and legal-technology roles are posted in Chicago and Houston, including an AI Infrastructure Director to run on-premise GPU clusters.

Critically, the architecture is model-agnostic: Kirkland can swap underlying foundation models without rebuilding, insulating it from lock-in to any single vendor, Palantir included. And unlike several rivals, Kirkland is not building to resell. Ballis told the FT that outside companies working on the platform would not be able to sell it to others, and that the firm would "own, or have the right to own, all of it."

What Palantir Actually Builds

The fund formation engine targets a practice where Kirkland supported nearly $500 billion in capital raised or targeted for clients in 2025 alone. More than 1,000 lawyers in its Investment Funds Group are expected to use it.

"Private equity fundraising has become significantly more complex, requiring fund managers to manage enormous volumes of information, transaction history and investor-specific requirements across global commercial and legal frameworks," said Erica Berthou, a partner and executive committee member at Kirkland. Combining the firm's funds expertise with Palantir's infrastructure, she said, would deliver better value to both general partners and limited partners.

Palantir's contribution is less about model quality than structure. It was brought in for what Kirkland calls its ontological expertise: linking funds, obligations, market terms and transaction history into one operational system, so knowledge scattered across attorneys, documents and spreadsheets can be applied in the flow of execution.

"Kirkland is defining what the next generation of professional services will look like," said Ryan Taylor, Palantir's chief legal officer, describing an ambition to build "an enterprise operating system that compounds every transaction relationship, decision and obligation into a continuously improving platform."

Why This Matters

Build-versus-buy is normally a cost question. Here it is a moat question.

The legal AI vendor market is extraordinarily well capitalized. Harvey is valued near $11 billion on roughly $300 million in annual recurring revenue. Legora closed a $550 million Series D at a $5.55 billion valuation in March and is reportedly in talks toward $10 billion. Both sell to everyone. That is precisely the problem for a firm whose pricing depends on being demonstrably better than the firm across the street: if every Am Law 50 competitor licenses the same agent, the tool raises the industry baseline and erodes no one's relative position.

Kirkland's alternative is to treat institutional judgment as proprietary data and workflow ownership as strategy. Rivals are hedging differently. Freshfields struck a deal with Anthropic in April to co-develop specialist tools that could later be sold to other firms, in exchange for early access to future models. A&O Shearman plans to sell its own fund formation AI externally, and Fried Frank is weighing client access to its version. Kirkland has emphasized exclusivity.

There is a second motive beneath the first. Ballis was unusually direct that AI accelerates the shift away from hourly billing. "We already do a number of matters on value-based pricing, and that trend will only continue and it will accelerate," he said, adding that the firm is "looking forward to leaning into it." Owning the system that produces the work is a precondition for pricing the outcome rather than the hours.

The risks are largely unhedged. Kirkland has published no return-on-investment benchmarks and has not said what would trigger a reassessment if frontier models commoditize this capability faster than expected. The profession's cautionary tales are fresh: Sullivan & Cromwell told a US federal bankruptcy court this spring that a major filing contained multiple AI hallucinations, and a London court reprimanded Pinsent Masons over false AI-derived submissions.

What to Watch

Three signals will separate strategy from spectacle: whether Kirkland's 2026 financials show the margin compression a self-funded half-billion-dollar build implies; whether the fund formation engine produces client-visible pricing changes rather than internal efficiency claims; and whether any peer firm matches the commitment at scale.

If none does, Kirkland will have bought something more durable than software -- three or four years of separation, encoded in a system its competitors cannot license.