Harvey's $15.5B Price Tag on Legal AI • The Weighted Average

Harvey is raising at least $500 million at a $15.5 billion valuation, a 40% jump from the $11 billion it commanded in March, on annualized revenue that has grown 80% since January to more than $350 million. SiliconANGLE’s report on the raise, citing The Information, makes legal AI the most aggressively priced vertical in enterprise software: divide valuation by run-rate and investors are paying 44 times revenue for a company that resells foundation-model capability through a legal harness.

The vertical that proved the model

The growth underneath the multiple is real. Harvey’s own customer page counts 200,000-plus professionals across 2,400-plus organizations in 70 countries, which puts average annualized revenue near $146,000 per organization—$350 million spread across that base. That is enterprise-grade contract size, not seat-count arithmetic, and it explains why Lightspeed is reportedly leading with Sequoia and Coatue returning from March’s $200 million round. Legal was supposed to be the vertical where AI stalled on liability and privilege; instead it became the one where usage-based expansion revenue showed up first.

The product has quietly become an agent platform. Harvey’s May agentic-platform update added subagent parallelization, step-by-step plan previews, and a Command Center dashboard for tracking how legal teams actually use agents—the harness-as-moat pattern that OfficeQA’s enterprise benchmarks identified as the durable layer between models and money. Harvey also plans a custom legal foundation-model series, announced in June, which would cut its dependence on the Anthropic and OpenAI models that power it today. Both of those suppliers are now building competing legal tooling of their own, which makes the custom-model bet less a margin play than an independence hedge.

The timing is not incidental. Workhorse model prices keep collapsing—Gemini 3.7 Flash halved output prices just this week—and Harvey appears in Google’s own launch materials as a reference customer. Every deflationary turn in inference expands Harvey’s gross margin on contracts priced against lawyer time, not tokens. The application layer captures what the model layer gives up.

The customer mix matters as much as the count. Harvey sells to the largest law firms and to in-house departments at companies like Bayer, and Dentsu reports savings of four to ten hours per lawyer per week. At blended big-firm billing rates, a single recovered hour a week covers an enterprise AI contract several times over—which is why legal departments sign six-figure deals without the procurement trench warfare that greets most software. The ROI case writes itself in the customer’s own timesheets.

What 44 times revenue has to believe

A 44× multiple prices in three assumptions. First, that legal remains a winner-take-most vertical: switching costs in matter history, precedent libraries, and compliance sign-off are high, and Harvey’s early lead compounds with every firm that standardizes on its workflows. Second, that model costs keep falling faster than contract prices, so margin expansion funds the custom-model program without external capital becoming a habit. Third, that the frontier vendors stay partners rather than predators.

The counterevidence is equally concrete. OpenAI is reportedly building legal tooling into ChatGPT, Anthropic already ships legal capabilities with Claude, and incumbents like Clio and Relativity own distribution Harvey must rent. There is also a ceiling question: 2,400 organizations is a strong start, but the global legal-services market is finite, and sustaining 80% growth requires either international expansion at lower contract values or new practice areas beyond the diligence-and-drafting core. If a frontier vendor ships a credible legal suite at seat prices, the harness premium compresses toward the model price—and 44× becomes a number investors mention with a wince. There is also the training-cost question: foundation-model ambitions have strained balance sheets far larger than a $350 million run-rate, and a custom model that merely matches frontier legal reasoning buys independence at the price of perpetual catch-up.

The funding itself is unlikely to be the last. Harvey’s valuation history—$3 billion in early 2025, $5 billion mid-year, $8 billion late in the year, $11 billion in March, $15.5 billion now—describes a company raising ahead of need while the window is open, a pattern that usually precedes either an IPO conversation or a very expensive lesson in capital discipline.

The verdict for operators splits by chair. Law firms and legal departments should pilot now—$146,000-per-customer pricing means vendors will negotiate—but negotiate portability of matter data and agent configurations as hard as price, because the moat being built is partly your own workflow history. Investors should treat 44× as a bet on harness durability, not on legal AI adoption, which is already proven. Builders in other verticals should read the Harvey round as the market’s current answer to what a proven, instrumented, agent-driven workflow is worth—and note that the answer went up 40% in five months while model prices fell by half. The evidence that breaks the bull case is a frontier vendor shipping legal at seat prices; the evidence that confirms it is Harvey’s custom model reaching parity on legal reasoning evals.