Harvey's $550m raise prices it at 38.75 times its own disclosed revenue

A $15.5bn price against annual recurring revenue above $400m works out at 38.75 times revenue ^[1], and it is 3.1 times the $5bn Harvey carried fourteen months ago ^[2]. For that multiple to compress to a merely expensive 10 times, ARR has to reach $1.55bn, which is 3.875 times where it sits ^[6]. The per-account version is the more interesting one: more than 3,000 organisations against $400m averages roughly $133,000 each ^[3], and with about 80% of the Am Law 100 already on the platform ^[16] and the customer count up 2.31 times since March ^[8], the next billion has to come mostly from deepening accounts Harvey has already won.

Set against the sector forecast, that disclosed ARR is about 7.7% of the $5.21bn the legal AI software market is projected to be worth in 2026 ^[4], while the valuation is close to three times that entire 2026 market and about 38% of the $40.94bn projected for 2034 ^[5]. Tech Funding News notes that most of the forecast growth is going toward tools that speed up existing legal work rather than displace the billable hour ^[19], so the price assumes Harvey holds a large share of a market whose buyers are still buying efficiency.

Post-training Moonshot AI's Kimi K3 with help from Fireworks AI ^[6] does not take a supplier out of the stack, it swaps which supplier and which contract: Harvey stops paying per call to a third party on that workload ^[9] and takes on someone else's release cadence and licence instead, four years after routing legal work through OpenAI, Anthropic and Google ^[8], with OpenAI on the cap table since 2022 ^[7]. Chief executive Winston Weinberg's stated case is that post-training models becomes a muscle a software company needs in order to compete ^[11]. The disclosure contains no gross margin, no inference spend, and no figure for what share of production workload Tenet now carries, so the cost benefit is asserted rather than shown.

The counter-thesis is that the defence is procurement rather than weights: Latham & Watkins and Microsoft's in-house legal team ^[16] bought an approved vendor, and Legora, in talks at more than $10bn against $5.6bn in March ^[18], which is 1.79 times in a matter of months and about 65% of Harvey's new mark ^[7], is scaling in Europe without the same in-house model story. Three ways this reads differently a year out. Tenet lowers cost of revenue enough that 38.75 times looks like a margin bet rather than a growth bet. OpenAI's direct law-firm partnerships and Anthropic's Claude legal plug-ins ^[17] compress Harvey's pricing before the model layer pays for itself. Or the documentation chain that Tech Funding News says now starts in Beijing rather than San Francisco ^[10] becomes a procurement objection at precisely the privileged-work accounts that carry the revenue. My read is that owning post-training buys a negotiating position against two suppliers who now sell against Harvey ^[17], and that it lands in gross margin long before it lands in retention. It would be falsified by a disclosed margin that does not move once Tenet carries production volume.