Legal AI hits $15.5B while its biggest buyer builds its own

Two legal AI funding stories landed inside a week, and the interesting one is neither. Harvey is reportedly in talks to raise $500 million at a $15.5 billion valuation. Legora is reportedly seeking a round at more than $10 billion — four months after closing at $5.55 billion. Meanwhile the largest law firm in the world by revenue looked at that market and committed half a billion dollars to building its own platform instead. Vertical AI valuations are running ahead of the buyers, and the buyer's move is the one worth copying.

What actually happened

Harvey is in talks for roughly $500 million at a $15.5 billion valuation, up from $11 billion in March, per The Information's reporting as relayed by SiliconANGLE. Annualized revenue is above $350 million, of which about $300 million is ARR — up from roughly $190 million ARR at the end of 2025.

Legora closed a $550 million Series D led by Accel at a $5.55 billion valuation in April, bringing total funding to about $866 million with Nvidia's NVentures and Atlassian among the backers. It is now reported to be seeking a round above $10 billion. Both rounds are in talks, not closed — treat the numbers as reported intent.

The third number is the one that isn't speculative. Kirkland & Ellis is investing $500 million over three to four years to build its own AI platform, starting with $100 million in 2026, working with partners including Palantir. The architecture is deliberately model-agnostic so the firm can swap foundation models without a rebuild, and the internal AI and innovation group runs to roughly 180 people.

Why it matters for your business

You are not spending $500 million on anything. The transferable part isn't the budget — it's the decision structure, and it's the same one we argue for on every engagement.

Kirkland didn't conclude that legal AI is overpriced. It concluded that the durable asset is its own matter data, its own workflows, and its own lawyers' judgment, and that renting a wrapper around those things is a worse trade than owning the layer that touches them. So it bought commodity parts, kept the model layer swappable, and put the integration work in-house. That's a build-versus-buy answer with a specific shape: buy the parts that are interchangeable, own the parts that aren't.

The second read is about your renewal. A vendor whose price expectation doubles in four months has to grow revenue per customer to justify it, and there are only three levers — more seats, higher prices, or usage-based repricing. We have watched all three run through this market in the past year. If a vertical AI tool is now load-bearing in your operation, the questions to answer before the next renewal are boring and urgent: can you export your data in a usable format, does the workflow logic live in their product or in something you control, and what happens to your unit economics at plus-30 percent?

Vertical AI is genuinely better than horizontal AI for narrow, high-value work — the revenue numbers say so. That's an argument for buying it. It is not an argument for letting it hold the only copy of how your business actually runs.

Sources: Bloomberg Law, SiliconANGLE, Crunchbase News, City AM.