AI Harness 的 ARR 倍数：25-125 倍区间与加速趋势

We now have enough data points to draw some initial conclusions. Harvey, Legora & Sierra each announced crossing $100m in annual recurring revenue within nine months of each other, & Ramp & Decagon bracket them at $1.4b & $35m.¹

The answer varies between 25-125x current ARR at scale, with multiples increasing recently still further.

These are among the fastest growing businesses at scale. The dashed lines above show multiple bands ; amazingly, each company seems to trade within just one or two bands, despite significant growth in ARR.² Typically, multiples compress with scale.

In January of 2026, we can see the acceleration of multiples for Legora, Sierra, & Ramp. Some of this may be due to sustained acceleration (Databricks’ is another example of this) ; another component, a more favorable fundraising market.

In 2021, I wrote about the 100x ARR multiple,³ a premium valuation for private startups & public companies. Five years later, we’re back at those levels, but the growth underneath is about 3x faster.

1. Harvey announced crossing $100m on 4 August 2025, with CEO Winston Weinberg quoted in CNBC. Sierra published “Sierra hits $100m ARR milestone in 7 quarters” on 21 November 2025. Legora announced passing $100m in its newsroom on 2 April 2026. Valuations are post-money from the round preceding each milestone: Harvey’s $5b Series E on 23 June 2025, Sierra’s $10b round in September 2025 & Legora’s $5.6b Series D including its April 2026 extension. Reported revenue at the time each round priced was roughly $75m for Harvey, per TechCrunch, & undisclosed for Sierra, whose last figure before the milestone was $26m at the end of 2024. Growth rates are annualized from each company’s prior disclosed revenue figure. None of these companies is audited & all are private. The time series adds Ramp & Decagon. Ramp’s revenue includes interchange earned on customer card spend rather than pure contracted subscription, so its multiple is better read against fintech comparables. Decagon has not disclosed revenue since late 2024, so its 129x divides a January 2026 round by an October 2025 estimate & should be treated as an upper bound rather than a print. ↩︎

2. The revenue figures behind both charts combine company statements with public extrapolations, so there is meaningful error in these estimates. Nine of the fifteen observations come from a company announcement or a founder post ; the rest are third-party estimates filling the gaps between disclosures. Sub-$100m figures are estimates in every case. Treat the direction of each line as the finding rather than the precise level. ↩︎

3. The 100x ARR Multiple, November 4, 2021. ↩︎