Harvey is raising $500 million at a $15.5 billion valuation, roughly doubling its price tag in five months after its annualized revenue jumped from $190 million in January to $350 million today. That growth rate is the whole argument for the round. The money likely funds Harvey’s announced push to build its own foundation models tuned for legal work, reducing dependence on Anthropic and OpenAI while those same vendors are building competing legal AI products<\/a> of their own.<\/p>\n The firms most exposed here are the ones still treating legal AI as a 2027 budget line. Harvey’s revenue trajectory, nearly doubling in under eight months, signals that law departments and outside counsel are signing contracts right now, not running pilots. If your general counsel is already spending on Harvey or a competitor, that spend is about to get a lot more defensible at the board level. If they aren’t, the gap to peers is widening faster than most transformation timelines assumed.<\/p>\n The strategic wrinkle is what Harvey’s model ambitions mean for enterprise procurement. Right now Harvey is a software layer sitting on top of Anthropic and OpenAI. When it ships its own legal-specific foundation models, the value proposition shifts from “better interface” to “different outputs you can’t replicate by giving ChatGPT a good prompt.” That distinction matters to any enterprise that is currently running a self-serve ChatGPT Enterprise deployment alongside a Harvey contract and assuming the two will converge over time. They probably won’t.<\/p>\n The sharpest risk isn’t Harvey’s valuation, it’s the dynamic where your AI infrastructure vendor and your AI application vendor are building toward the same product from opposite directions. OpenAI and Anthropic selling directly into legal is not a rumor, it’s a stated roadmap. Harvey’s $500 million bet is essentially a race to make itself irreplaceable before that happens. Enterprise buyers sitting on annual renewal decisions for legal AI platforms should weight vendor independence from hyperscaler competition more heavily than the sales cycle typically invites them to. The companies that survive as application vendors in a world where foundation model makers go vertical are the ones with proprietary models and deep workflow lock-in. Harvey is spending to get both.<\/p>\n A foundation model is a large AI system trained on broad data that can handle many tasks, similar to a generalist employee on day one. Vertical tuning means retraining or extensively customizing that model on a specific domain’s data and tasks, like years of case law, contract language, and regulatory filings, so it performs better than a generalist on those specific jobs. The business case is lower inference costs and outputs that a general-purpose model structurally cannot match, which is the moat Harvey is trying to build.<\/p>\nWhat this means for your business<\/h2>\n
Concept deep-dive: Foundation model vertical tuning<\/h2>\n