Three AI firms hit $100m ARR. Multiples ran 50x to 100x

Harvey announced passing $100m in annual recurring revenue on 4 August 2025, with chief executive Winston Weinberg quoted in CNBC. Sierra said it reached the same mark in seven quarters, in a post dated 21 November 2025. Legora announced it on 2 April 2026.

The valuations attached to those three businesses were not close. Tunguz, a general partner at Theory Ventures, took the post-money valuation from the round preceding each milestone: Harvey's $5bn Series E in June 2025, Sierra's $10bn round in September 2025, and Legora's $5.6bn Series D including its April 2026 extension.

Set against $100m of recurring revenue, that is 50x, 100x and 56x. Same milestone, twice the price at one end of the range.

Tunguz's argument is that growth rate does not account for the spread, because on his data the fastest grower priced near the bottom of it. He attributes the premium to category position instead.

The wider set he looked at runs from 25x to 125x current ARR at scale, and he says multiples have been rising recently rather than compressing, which is the usual pattern as revenue grows. He puts part of that down to a friendlier fundraising market.

He also notes he wrote about the 100x ARR multiple in November 2021. Five years on, he says the market is back at those levels, but with roughly three times the growth underneath.

For a founder, the practical reading is about what the deck argues. Two companies can present the same revenue line and be priced 50x or 100x, so the variable worth working on in a pitch is the claim to own a category, not the slope of the chart.

The caveats matter and Tunguz states them himself. None of these companies is audited and all are private.

Reported revenue at the time Harvey's round priced was roughly $75m, per TechCrunch. Sierra's revenue at the time of its round was undisclosed; the last figure before the milestone was $26m at the end of 2024. Growth rates in the analysis are annualised from each company's previous disclosed figure.

His time series also includes Ramp at $1.4bn and Decagon at $35m. He says Ramp's revenue includes interchange on customer card spend rather than contracted subscription, so it reads better against fintech comparables, and that Decagon's 129x divides a January 2026 round by an October 2025 estimate and should be treated as an upper bound.

Of the fifteen observations behind the charts, Tunguz says nine come from a company announcement or a founder post and the rest are third-party estimates filling the gaps. Every sub-$100m figure is an estimate.

His own instruction to readers is to treat the direction of each line as the finding, not the precise level. That is the honest way to use these numbers in a fundraising conversation, and the way they will be used against you if you overstate them.