Harvey said on September 9 that it had raised $550 million. The company's own post put the mark at $15.5 billion. The Next Web, matching Bloomberg, printed $15.6 billion, a touch above the August ask. Lightspeed and Diffusion co-led. Diffusion is Kris Fredrickson's new firm after Coatue. Sapphire Ventures and Whale Rock came in. Sequoia, Kleiner Perkins, a16z, Coatue, GIC, and Goldman Sachs Alternatives were among the holders who re-upped.
Harvey has now taken more than $1.5 billion since 2022. ARR has passed $400 million. Customers sit above 3,000 organisations, up from about 1,300 in March, when a $200 million round marked the company at $11 billion. At $400 million of ARR, $15.6 billion is about 39 times revenue. Lightspeed's Sebastian Duesterhoeft, joining as a board observer, said legal is the second-largest AI market after coding, and that in-house teams are the path to the whole legal-services spend.
I read this as a stack-control round. Harvey grew up on OpenAI. The new money is earmarked for models Harvey trains itself. Tenet, the first post-trained legal model, sits on Moonshot's Kimi K3. The Next Web says Harvey already routes a task to the cheapest model that is good enough, and fine-tunes on historical documents the firms provide. A legal vendor that can park a Chinese open-weight base under a named product is treating the frontier API as a cost line.
The Guardrails AI purchase is Harvey's fourth acquisition of 2026. Terms were not disclosed. Shreya Rajpal, Zayd Simjee, and their team join product and engineering. Guardrails built an open-source validator layer Harvey says is downloaded more than 250,000 times a month, plus Snowglobe, a simulator that stress-tests agent behaviour before a live client sees it. Weinberg said Harvey is shopping teams more than products. The cash is paying for people who can keep an agent inside a policy while it works a matter for hours.
Harvey says 80 percent of the Am Law 100 uses the product, along with five of the Fortune 10, Latham & Watkins, and Microsoft's in-house legal team. Law firms bill by the hour. A tool that compresses a junior's night on a diligence memo can shrink the hours the firm invoices. Duesterhoeft's answer is corporate legal departments, where cutting time is the point of the budget. I do not have a published mix of firm seats versus in-house seats inside the $400 million. Without that mix, 39 times revenue is a claim about where the next dollar of ARR comes from.
Anthropic has legal plug-ins for Claude. OpenAI is customising ChatGPT with law firms. Sweden's Legora hit $100 million of revenue in 18 months. The Next Web puts Harvey at roughly four times Legora's revenue and about three times its valuation. That is a lead. It will not stay one if the labs sit inside the same workflow.
Harvey LAB is how Harvey wants that fight scored. A vendor that owns the eval and the watchdog can tell a general counsel that unsupervised hours were simulated first. The $550 million pays for weights Harvey can move. Guardrails is the kit for letting those weights run longer than a prompt.
Canonical essay: https://deanlee.info/essays/harvey-legal-ai-round/
Top comments (0)